The rectifier read 3 750 kW, the same as the inverter.
Four mutations each turned the gate red:
a planted literal in transformer.js;
the present-load loss taken from the design load;
a wrong engine basis on a UPS row;
a lossless rectifier.
A code review found four switching cases the normal state cannot show. Each now has a check that was red on the
first P4a build: output breaker open (156 kW of loss on an idle inverter), bypass (rectifier still drawing),
CB-UI + CB-UO open (drawing NORMAL, sheet ON BATTERY), TX-AA dead (3.42 kW loss), CB-FA4 drawn 33.0 A against 165.2 A.
Fixed in the port
The seven other Conventional cockpits load js/conv-engine.js?v=2.6.0 again (the port had kept 2.5.0, so every
page but EPMS would have opened "Authority UNAVAILABLE"); test-conv-engine-version-pins red before, green after.
tools/test-conv-datahall-air.mjs, red since v3.89.0 (12 of 48): it assumed the old Plan projection and the Power
mode; it now pins Plan and presses Power as a reader does. 48 of 48.
tools/test-conv-epms-sld.mjs followed the P4b route: a breaker, switch or relay now opens its equipment modal, whose
action row carries the same page actions (sim-open/sim-close/tcc, select-before-operate kept). V7–V11 read that row
as well as the inspector: 13 of 92 checks red before, 122 of 122 after.
The equipment modal inerts every page element again, hidden ones included (test-conv-equipment-modal Q4 was red,
114 of 114); the time–current dialog lifts its own inert while it is open above the modal and gives it back on close.
DC AI Data Hall, 3D floor: it no longer shrinks by itself until it vanishes (owner: "bergerak membesar kemudian mengecil
kemudian hilang"). The floor renderer measured the bordered #hIso it then resized, losing 2 px per resize round;
it now measures a borderless wrapper. New law D8 in tools/test-dcai-hall-iso.mjs (6 s steady, follows a window
narrowed and widened back): 38/40 before, 40/40 after.
P4b — Breakers and relays open up, and are operated from inside
Owner (R186): a click on a block must show the process inside the machine. PRD-CONV-EQUIPMENT-INSIDE phase P4b
(P4b-BREAKER-RELAY.md) brings this to the switchgear and the protection on EPMS_Telemetry. Before this release a
click on a breaker or a relay opened the side inspector, and the Select / Execute buttons lived only there.
Added
Circuit breaker inside (js/conv/internals/breaker.js, class breaker): CT, main contacts in the arc chamber,
the trip unit or the relay that trips it, trip and close coils, spring mechanism, 52a/52b, the interlocked earthing
switch or disconnector (drawn only when the sheet has one), FROM and TO. All 129 breakers of E-S-101 and of
E-S-102 in the four halls open it with one click. Position, current and kW come from the page's own solve; the
fault level at the bus and the HV breaking capacity come from the engine; Ir / tr / Isd / tsd appear only on a
breaker an LSIG relay trips. Spring state, operation counter, contact wear and trip-coil supervision read
"not modelled".
Protection relay inside (js/conv/internals/relay.js, class relay): CT (and VT for 21) inputs, measuring
element, one block per ANSI function parsed from the relay name (IEEE C37.2), trip logic, the breaker it trips,
watchdog. All 28 relays open it. The graded chain shows its engine settings (R-FA1: Is 113.7 A, EI, TMS 1.08,
0.802 s, instantaneous 1 540.5 A); 87L, 21, 87T, 87B, 50BF, 51G and 50N/51N read "not modelled"; a setting the
element does not have reads a dash (an LSIG unit has no TMS).
Operate from the modal: the equipment modal gains an action footer with the page's own buttons (Select · Open,
Execute, Cancel, Time–current curves), dispatched to the same handler registry the inspector uses, so
select-before-operate and the interlocks stay the page's. The page's last message shows under the buttons and the
drawing refreshes at once.
js/conv/epms-internals.js 1.1.0: breakerUnits() (FROM / TO walk power edges only, skip earth switches, read a
busbar junction as "HV/MV/LV bus"; strips by voltage class and, on E-S-102, by stream, at most 14) and
relayUnits().
tools/test-conv-internals-p4b.mjs EB0–EB8 (3 910 checks, 1440 + 390), in the ship gate.
Fixed
The modal's source note on electrical units said "plant cooling values"; it now names the single-line solve.
A dialog opened from the modal (the time-current curves) now owns Escape: Escape closes it first and the modal
stays open and usable. Before, Escape closed the modal underneath.
A breaker tile in the unit strip showed "LOAD —"; it now reads its current or OPEN.
A TCC closed earlier stayed inert when it was opened again from a later modal session (no click worked); the modal
no longer inerts hidden dialogs.
The page's last operate message showed under every unit; it now shows only under the unit it names.
After Select the pressed button is replaced by Execute / Cancel; the keyboard focus now stays in the modal. A
handler that throws no longer skips the refresh.
The breaker's interrupting medium and the LV thermal-magnetic trip are labelled "typical" (they follow the voltage
class, not a per-breaker datum).
Proof
Red on the P4a tree (0 of 129 breakers mapped, no relay class) and on six mutations: Ir taken from Isd, LSIG on any
relay, a CT named as FROM/TO, relay name split on " · " only, TMS shown on an LSIG unit, Escape not yielded, hidden TCC inerted, message not scoped, focus not kept.
Deferred to P4c: the switchboard class (EB7) and the goto navigation.
v3.92.0MINOR
— The AI hall in 3D, read like a BMS
PRD-DATAHALL-BMS-ISO phase F4 (owner R185: "Kenapa tampilan UIUX datahall jadi 3d jadi kayak bukan BMS … floor plan dulu
bentuknya 3d tapi sekarang flat"). v3.89.0 gave the Conventional data hall a 3D floor drawn like a BMS graphic. The DC AI
Data Hall tab (datahallAI.html) was still a flat plan. It now has the same floor and the same 3D | Plan switch.
One renderer, an adapter.
js/datahall-ai/hall-iso.js (new) turns the hall's engine counts into the model js/conv/floor-plan.js draws.
The counts are rows, racks per row, rack kW, fan-wall modules installed / active and CDUs installed / running.
The 3D floor holds 20 rows as 10 contained pairs, one hot-aisle roof per pair, the north and south fan-wall
galleries, and the CDU line between the rack field and the south gallery. Standby units are drawn as outlines.
Temperatures are the aisle readouts the plan already draws (cold aisle = rack inlet). Before the first readout,
the engine's design supply and return air stand in. The adapter does no engine arithmetic.
Same rules as the Conventional hall.
A normal rack is grey with a green lamp. Only a warn or alarm rack wears its state colour.
Every rack is the same height. Both plan axes are drawn at 30°.
Same HMI on a click. Each 3D rack and unit carries the plan's equipment reference (rack-unit:Rnn-kk, crah:nn,
cdu:n), so a click opens the HMI the plan opens.
Plan is the engineering drawing, unchanged. In 3D the plan keeps only its header (hall line and the IT, liquid
and air cards). Its layer bar and key, which describe the plan, are hidden. Plan elements below the header are taken
out of the render, not clipped at the frame, so test-conv-geometry G2 stays at 0 clipped (it counted 3,084 before
this was fixed).
The switch. 3D is the default on a stage of 600 px or more; Plan is the default on a phone. The choice is stored
under rz.dcai.proj.
js/conv/floor-plan.js 2.1.0.
The line length follows model.racksPerZone.
model.cduUnits adds a CDU line.
racks[i].ref and unit.ref become data-rz-equipment.
opts.toggle: false hands the switch to the page; opts.zoneHead sets the row tag's head.
The Conventional hall's drawing is unchanged: its geometry is still GEO, 10 cabinets a line.
Gate.tools/test-dcai-hall-iso.mjs (new), laws D1–D7. Red on 7f7608e: 29 of 30 failed; the only pass was
page width at 390. It is green on this release (37/37).
v3.91.0MINOR
— The chiller runs its own cycle
Owner (R186): open chiller 1 and see the process inside, evaporator, compressor, condenser, expansion valve.
PRD-CONV-EQUIPMENT-INSIDE phase P2. The drawing was there. Most readings inside it said "not modelled" because the
engine never solved the refrigeration cycle.
Changed
js/conv-engine.js 2.5.0 solves the R-134a two-stage economised cycle of each running chiller. It returns eight
state points (t, p, h, s, x), the economiser pressure and flash fraction, both refrigerant flows, shaft power,
discharge temperature, compressor speed and guide-vane position. Properties are NIST WebBook rows; nothing is fitted.
The 0.58 kW/RT figure is now what it always was: the AHRI 550/590 RATING (lift 30.9 K). Until 2.4.0 it was applied
at this plant's warm-water lift (17.9 to 38.5 °C, 20.6 K). The same machine there needs 0.36 kW/RT, so chiller input
falls from 5,153.4 to 3,212.5 kW and COP rises from 6.06 to 9.73. Tower rejection is 34,462.5 kW, CDW flow 235.2 L/s
per unit, mechanical plant 4,622.1 kW, headroom 7,627.9 kW. PUE is an input and does not move.
— Conventional sheets: a tag struck by a foreign symbol's outline now fails the gate
Added
C5 in tools/test-conv-sheet-text-clear.mjs. C3 measured only the filled body box (.b) of a symbol, so an
outline stroke of another asset (a tank shell, a pipe stub inside a symbol, an instrument bubble) could run through a
tag and still pass, the limit v3.62.0 wrote down. C5 samples every stroked shape inside each foreign .hmi-node every
1 user unit and fails when a point lands inside a visible text box. Fire, chiller, water and fuel sheets are clean.
Proven red with RZ_TEXT_CLEAR_PLANT=1, which moves one tag onto a foreign outline: fire FACP-01 on the VESDA symbol
fails C5 while C3 still passes (the blind spot), and the other three sheets fail both. Standard: CONVENTIONAL_DC_OPERATIONS_STANDARD item 26.
v3.62.1PATCH
— datahallAI as an operator BMS: flow only with evidence, alarm badges, equipment finder, cooling overview and named equipment, SLDs without icon grids, schematic modals, 6 ms repaint
Gates that read the clock.audit-editorial-render.mjs pins the page clock to local noon: achievements.html unlocks "Night Owl" between 22:00 and 05:00, which added 4 off-scale values, so the unchanged v3.61.0 tree failed at 00:40 (487 → 491) and passed at 20:00. A zoom button now commits its viewBox at once (only wheel and pinch gestures defer it); the Conventional truth gate B5 read the viewBox right after the click and saw the old one.
Deferred panels kept deferred (gate D6). hmi-v2 drew all 14 sheets into hidden tab panels at boot, so 6,127 attribute writes landed in panels nobody could see (p-over 1,797, p-elec 1,378, p-fire 957 …). A sheet is now drawn when its panel is shown (tab click or key), the CDU opener no longer waits for the cooling sheet. tools/test-dcai-deferred-panels.mjs: 7 failures on db5284e, clean after.
Added
Electrical → DH-02 … DH-08 are HMI sheets (DH-E-102 … DH-E-108), drawn by the DH-01 builder for each hall; the
engine sizes every hall the same, so the sheets differ only in name. A hall sheet mounts when its sub-tab draws.
Cooling → DH-M-102 heat rejection (the dry-cooler field with every installed unit, HTW loop to the CDU line, CW loop
from the chiller condensers, chillers making CHW for the air path; temperatures from the engine's design planes) and
DH-M-104 TCS to the rack rows (supply / return headers, one manifold per rack row, flow per rack). The off-sheet
links on DH-M-103 now lead somewhere.
Chiller, CRAH and battery modals draw their L3 sheet like the CDU modal: chiller (evaporator, condenser, compressor,
expansion valve, CHW / CW transmitters with design temperatures, TE / TC), fan wall (filter, CHW coil and valve, four
EC fans, supply / return air), battery (DC bus, four strings with breaker and fuse, kWh per frame). The page modal
keeps its focus / inert / aria contract; the old mimic is retired, not deleted.
The sheets tell the operator only what the data supports (owner brief "BMS UI/UX master prompt"): a line animates
only when the pump, fan or equipment driving it runs with good data — stopped pumps stop their lines, missing samples
stop all motion; items with no data (isolation valves, strainers) are drawn neutral instead of looking healthy; every
node carries its data quality; an alarm shows a priority badge readable without colour, hollow once acknowledged; each
CDU pump shows its own state (pump 3 was hard-coded standby). Gate tools/test-dh-operator.mjs (12 laws, 9 red
before the change). Plan: documentation/prd/datahall-ai/PRD_BMS_OPERATOR_HMI.md. Electrical sheets take breaker, transformer, ATS and UPS states
from the page's electrical scenario engine (a tripped busway stops its lines; a lost utility de-energises its
transformer; UPS bypass closes the static switch); the bus tie reads open, as designed.
Find equipment on every sheet: a finder in the sheet bar lists every item by tag and name; choosing one zooms to it,
selects it and focuses it — on a phone the fitted sheet is too small to tap. Sheets fit without overflow at 360, 768,
1440 and 1920 px and at 200 % browser zoom; opening and closing an equipment modal keeps the sheet's zoom and selection.
BMS controller modals draw the controller dual-homed on the supervisory ring, its field I/O and the servers (from
the page's BMS model), above their detail.
Network modals draw the switch (uplinks, downlinks, OOB, dual power, per-switch counts from the engine's hall link
totals) above their detail. Fixed: the network sheets printed 400G ports while the engine's published port class is
800G — every port speed now comes from the engine.
More modals draw a schematic instead of text cards: SLD mimic (the full feed one-line, states from the scenario engine,
opened device selected — the old four tiny panels showed a null% gauge), rack power shelves, GB300 compute tray (the
old block named the GPU B200, the GB200 part; counts now from the engine), NVLink switch tray, STP, AHU.
Fire → Mimic → DH-F-102 clean-agent release: cylinder bank (release state from the workstation), manifold, one
selector valve and nozzle set per release zone of the fire inventory.
Legibility: no sheet renders below the scale its smallest label needs (≥ 8.5 px on screen); narrow screens scroll the
sheet inside its frame instead of shrinking it; stacked tag / name / rating lines 12 px apart. The geometry gate counted
515 label overlaps and 964 illegible labels on the v3.61.1 candidate; 0 and 0 after (4,816 labels, 96 views).
Fixed regressions from v3.60.0: the SIMULATED · ENGINE BASIS provenance banner was invisible (its slot sat in the hidden
header row) — it is back in the nav row with its 44 px dismiss; CDU pumps and the SLD battery opened no inspector.
Faster sheets: a repaint of the Room Layout sheet took 2.2–3.2 s (the whole electrical topology was re-evaluated
for every item) and now takes 6 ms; zoom and pan move the drawn sheet on the GPU and re-render it sharp when it settles
(60 fps median); an alarm reaches the sheet when its source changes (174 ms) instead of on the next 4 s poll. The
20 ms p95 frame target is not met yet (50 ms on this laptop's GPU) — measured and recorded in the PRD.
Fixed
DH-E-001 listed DH-02 … DH-04 instead of DH-02 … DH-08 (shipped in v3.60.0): it took site.hall_levels (4 levels)
as the hall count. The count now comes from the engine — installed CDUs on the campus ÷ installed CDUs per hall
(440 ÷ 55 = 8) — in the site sheet, the UPS split per hall and the hall builder; the hard-coded halls = 8 is gone.
The drawings gate and the local mock had stopped unlocking the cockpit (since v3.57.7 a session forged in
localStorage no longer unlocks it, and the page re-locks): both now clear the auth blockers with the shared audit helper's
observer, as every other cockpit browser gate does. A sheet builder that throws now leaves the old drawing in place.
Gate tools/test-datahall-ai-inspector-runtime.mjs: its file server wrote headers before reading the file, so the
first missing file crashed the run (ERR_HTTP_HEADERS_SENT); it now reads first and lists any 404 the page requested.
js/datahall-ai/hmi-v2.js (mount + payload binding; fire points keep their full inventory id and take their
state from the workstation snapshot).
Gate tools/test-dh-drawings.mjs in the ship gate: H0 ratchet (≥ 10 HMI sheets, may only grow), H1 paint only
from the renderer's own <defs> (no foreign gradient / filter / shadow), H15 the Shade toggle (deep by default,
flat after one click, flat after reload), H16 edges through a symbol body (reported), H17 shading filters per
layer, never per symbol (≤ 12 filtered elements per sheet — 599 per-node shadows made a tab take 25–36 s), H2 library-only, H4 title + focus, H5 ≤ 8 words per sheet, H6 fit-to-width at 1440
and 390, H13 first open, F1–F9 on the fire sheet (mounted, 12 zones, devices = inventory, no unknown state, no
overlapping symbols, no page overflow, no console error). H3 (own L3 modal per class) and H11/H12 (shell) are
reported and block only with --strict-all until their phases land.
Standard standarization/HMI_DRAWING_STANDARD.md (sheet object v1, live binding, laws, local mock seam).
Changed
tools/build-dh-drawings-mock.mjs: the local mock opens unlocked on 127.0.0.1 / localhost through the documented
premium test seam (owner R142); inert on any other host.
Not in this release (next)
The remaining cooling (101/102/104/105), electrical (DH-02…08, 002/003) and fire (102–104) sheets; the Room Layout building isometric and per-level plans as sheets; L3 schematics for every class
(H3 ratchet); R135 commissioning scenarios on these drawings.
v3.58.5PATCH
— Finance Terminal names a blocked gateway instead of timing out tile by tile
Fixed
On a network that blocks *.workers.dev (checked live on an Indonesian ISP) the gateway request hangs instead of
failing, so v3.58.4 showed "timed out" on every tile, no banner, and the ticker stayed on "Loading market data..."
for over a minute while queued requests each waited 10 s. Two timeouts in a row now count as unreachable: the
banner appears and the remaining requests fail at once until Retry. Gate law D7 (gateway hangs) failed on v3.58.4.
v3.58.4PATCH
— Finance Terminal reads real data through the gateway; admin tabs fit on desktop
Fixed
Finance Terminal showed "$0", "0 coins" and "0.0%" for data it never received. Every quote went through public
CORS proxies (corsproxy.io, allorigins, codetabs) that now fail (401 / network error), and CoinGecko was called
straight from the browser, where its 429 arrives without CORS headers. All market data now comes through the
first-party gateway (one GATEWAY_URL constant); a missing value shows "—", a short reason and Retry on every tab.
Crypto prices fall back to Binance's public data-api.binance.vision when the gateway's CoinGecko is rate-limited,
and the source line says so. A missing price no longer triggers "below" alerts or values a holding at its buy price.
Gateway calls are limited to 4 in flight and 90 per minute, identical requests share one call, and a 429/5xx is
retried once after Retry-After. A network failure to reach the gateway (some ISPs block *.workers.dev) shows one
banner, "Market data gateway unreachable from this network", instead of empty tiles.
The stock chart threw a page error whenever it had no data; the header now says "Delayed" (amber) instead of a red
"Offline"; the fake DXY tile is the real UUP fund.
rz-ops tabs were wider than the screen on desktop: User Activity 1350–1512 px at 1024–1440 (chart canvases
widened their grid tracks), Benchmarks 1331 px (tabs did not wrap), Users 1170 px at 1024. All fit now.
The coloured top stripes on the activity category cards are gone (1 px hairline only).
Removed
The public CORS proxies and the localStorage switches that could turn the gateway off or change its URL.
Gates
tools/test-finance-terminal-data.mjs D1–D6 (all six failed on v3.58.3); tools/test-rz-ops-access.mjs M2
(1024/1280/1440 px, failed on v3.58.3). See standarization/FIN_ENGINE.md "Data sources".
v3.58.3PATCH
— Admin suite theme aligned: embedded apps follow the console, light mode from tokens
Fixed
The apps inside the admin console ignored its theme. In light mode Finance Terminal, Stock Investment (DCA) and
IDR Stocks stayed dark: the console sent the theme only when the toggle was pressed and only to two of them, and DCA
and IDR had no listener. It is now sent on every iframe load and every toggle to all three; each app checks the
origin and applies it to <html data-theme>, and reads rz_admin_theme on first paint when embedded.
rz-ops light mode was 111 per-selector hex overrides. The toggle now switches <html data-theme> and light mode
comes from the --fs-* suite tokens; ~357 inline colours moved to tokens. Text below 4.5:1 (3:1 large) fell from
154 (dark) / 166 (light) to 0 on every tab. Shared change: css/rz-finance-suite.css darker light-mode
t4/grn/red/amb/cyn and dark t4 #808998, so Finance Terminal, account.html and StockMap read better too.
System Health invented numbers ("99.97 %", "42 ms", "0.02 %") and offered "Reset Mock Data". It now times one
real Supabase read (reachable + round-trip) and shows the real localStorage size; the mock buttons are gone.
Platform Config states only facts: hosting is GitHub Pages (not Cloudflare), 11 calculators from the registry,
sessions governed by Supabase Auth, environment from the host name.
The topbar clock is labelled with the viewer's real time zone instead of a fixed "UTC+7".
The Finance Terminal status line no longer scrolls off the left edge ("ket data unavailable").
Changed
One accent: violet (#8b5cf6, a rejected colour) removed from the DCA app (checked by hue), no modal blur, no glow;
"Add account" and "Promote to educator" use the amber outlined button; the login screen follows the console theme,
without the orb/grid backdrop, and says "Supabase Auth" instead of an unverified "256-bit encrypted".
Tier Manager v2 stays hidden while its worker is not deployed.
Gate
tools/test-rz-ops-theme.mjs: T1 every embedded app's theme equals the console's after load and after each toggle,
T2/T3 no theme-mismatched surface and no low-contrast text on any tab in either theme (gradient-aware), T4 no
invented health numbers, T5 real time zone, T7 no violet hue or blur, T8 login in both themes, T9 Platform Config
true, T10 status line on screen. It fails 19 laws on v3.58.1.
v3.58.2PATCH
— DCMOC: jspdf 4 + next 16.4.6, production audit clean; Country Intelligence PDF works
Security
DCMOC shipped jspdf 2.5.1 (critical PDF-injection advisories) and next 16.2.6 (critical). Now jspdf 4.2.1,
jspdf-autotable 5.0.9, next/eslint-config-next 16.4.6, and a non-forced npm audit fix for postcss, sharp, nanoid,
source-map-js, fflate and baseline-browser-mapping. npm audit --omit=dev in dcmoc/: 9 findings (2 critical,
5 high) → 0. No source change was needed: every builder already goes through initDoc() and calls
autoTable(doc, opts) as a function. Rebuilt with next build --webpack, one build id.
Fixed
Results → Full Report → Country Intelligence PDF never generated ("Error generating PDF"): its mitigation table
read annualBenefit / roiPercent, fields the disaster-risk engine does not have. It now shows the engine's own
Cost, Risk Reduction % and ROI; anything that is not a number prints "—".
Checked
All 27 PDF reports DCMOC offers download as valid PDFs (jsPDF 4.2.1) with the previous page counts and 0 page
errors; with exchange rates held fixed, report text equals the previous build's line for line (the only drift is
the live FX rate, which the old build showed between runs too). Trace parity 214/214, enum coverage clean, value
bindings green, Site Intelligence map renders.
v3.58.1PATCH
— PUE calculator: the Measured tab works again
Fixed
pue-calculator.html loaded no js/rz-pue-measured.js: the v3.58.0 rebase onto v3.57.7 resolved the cache-token
conflict at the end of the page in favour of the upstream copy and dropped the new script tag with it, so the live
Measured tab did nothing (the R125 gate had run on the pre-rebase tree). The tag is back after rz-engine.min.js.
Rebase check extended: script and link tags are compared with their tokens stripped, not skipped (tools/ship-gate.sh
unchanged; procedure in standarization/VERSIONING_STANDARD.md).
v3.58.0MINOR
— Measured PUE against a public reference; employer cards keep their size
Added
PUE calculator: a Measured tab beside Estimate (owner R125). PUE from 12 months of metered energy (peak kW for
category 0) per the Data Center Metrics Task Force *Recommendations for Measuring and Reporting Overall Data Center
Efficiency* v2 (2011, hosted by ENERGY STAR):
category select names where IT energy is metered (0/1 UPS output, 2 PDU output, 3 IT equipment input); the result
carries the subscript (PUE<sub>1</sub>)
every energy type crossing the boundary, weighted by the reference's source factors (electricity 1.0, natural gas
0.31, fuel oil 0.30, other fuels 0.30, district chilled water 0.31, hot water 0.40, steam 0.43, condenser water
0.03); electricity shown, the rest through "Add energy type"
each weighted line printed, the design estimate shown beside the measured value
starts on Appendix A Case B: 1.708 (the paper prints 1.70 for both worked cases; the exact quotients are 1.705 and
1.708)
refused inputs show a reason and no number: category 0 with non-electric energy, IT 0, negative energy, total
below IT
the "What drives this result" panel and the charts stay with the estimate: hidden on Measured (its registry anchor
is now #pmEstimateBench; the old section.calculator-section matched the new Measured bench first)
Engine:models.pue.measured(opts), DATA.pueSourceWeights, DATA.pueCategories, with source and refresh
entries. Gate tools/test-pue-reference.mjs (W1, F1–F5, red 6/6 before the change) and browser gate
tools/test-pue-measured.mjs (M1–M7, red 11/11 on the page without the tab). Gap file
documentation/product-transformation/gaps/pue.md; defect D1 (the "Green Grid Level" select changes no number)
recorded and answered by the category select.
Fixed
Homepage employer cards no longer change size when flipped (owner R126). The back used to leave the flow and the
card grew from 305 to 603 px at 390 (397 to 770 px at 1440). The back now fills the front's box and only its fact list
scrolls; the Back button sits below the list and never covers a fact. tools/test-index-r079.mjs L3: width and
height change ≤ 0.5 px for all five cards at 390/1024/1440 in both themes; the last fact is reachable by scrolling.
v3.57.7PATCH
— Feature overrides and the educator role come from Supabase
Security
Anyone could grant themselves features, and the admin's switches reached nobody. The rz-ops Feature Flags panel
and Educator promote/demote wrote to the admin's OWN localStorage (rz_admin_features_by_page,
rz_admin_educators), and every page read those keys from the VISITOR's localStorage. Now:
overrides live in Supabase feature_overrides (anyone may read, only root may write — RLS); pages read them after
first paint (never blocking it), cache them 5 minutes in sessionStorage — a cached entry can only restrict, a grant
is always re-fetched — and fall back to the shipped matrix when Supabase is unreachable;
the educator role comes from the server (educator_allowlist, is_educator()), re-checked at most every 10
minutes; a session claiming educator without a Supabase sign-in is treated as pro. The client seed list is gone;
the old localStorage keys are deleted on load; rz-ops reads/writes Supabase with saving/saved/error states, offers
any overrides left in the admin's browser once (upload or discard), and has an Educators card (list, promote,
demote).
The live project already has the tables (anon read returned 200) — no migration needed.
Gate
tools/test-feature-overrides.mjs: a forged local key stays locked (V1), a server grant unlocks and a server
restriction locks a real gated page (V2/V2b), a forged educator list gives no role (V4), Supabase down → defaults
with 0 errors (V3), the admin toggle calls setFeatureOverride once and leaves localStorage byte-identical (A1),
promote/demote call the RPCs (A2). 9 of its laws fail on v3.57.6.
probe-educator-access.mjs is unchanged at 61/65 — the same 4 stale expectations as before this release.
Owner action
rz-ops → Users → Educators → add educator@resistancezero.com if that account should keep educator access.
v3.57.6PATCH
— DCMOC map/sanitizer upgrades, auth worker hardened before deploy
Security
DCMOC shipped two libraries with published XSS advisories: maplibre-gl 5.24.0 (critical, sanitizer bypass) and
dompurify 2.5.8 (high). Now maplibre-gl 6.18.0 and dompurify 3.4.17 (forced under jspdf with an npm override); the
Site Intelligence map popup builds its content with textContent instead of HTML, since site names are typed by
users. maplibre 6 loads its worker from /dcmoc/_next/static/media/maplibre-gl-worker.*.mjs. Rebuilt with
next build --webpack, one build id shipped. Trace parity 214/214, enum coverage clean, value bindings green, map
and PDF report checked headless with 0 errors.
worker-auth (not deployed) hardened first: login and seed accept only application/json from an allowlisted
Origin (login-CSRF with a text/plain body is closed); the seed token moved from ?token= to an X-Seed-Token header
compared in constant time; login attempts are counted in a per-IP Durable Object instead of a non-atomic KV
read-modify-write (KV stays as a documented fallback). worker-auth/test/hardening-s12.test.mjs: 10 of its laws fail
on the old worker; the suite is 106/106.
Still open
DCMOC npm audit: jspdf ≤4.2.0 (PDF injection; the fix is a breaking major) and others, tracked in the security
ledger. next ≤16.4.0 advisories concern the server and do not apply to the static export.
Homepage motion (R121), live on index.html. The motion approved on the mock now runs on the homepage:
smooth scroll (Lenis 1.3.27) only with a mouse or trackpad; phones keep native scroll
the name rises word by word, each section reveals once, employer cards rise as a row, the portrait drifts slightly
(GSAP 3.15.0)
reduced motion: nothing moves; no JavaScript: everything is shown
the IntersectionObserver reveal it replaces is removed with its CSS, and the mock page is gone
CAPEX: one answer for one project (R120). The calculator and DCMOC priced the same project differently: 2.2 MW,
N+1, air cooling in Virginia read $18.1/W on the calculator and $8.8/W in DCMOC, against the city table's own $13.4/W.
A city's $/W is now the all-in total of a stated basis spec (N+1, air, medium density, purpose-built, seismic zone 2);
both engines land on it at that spec and move from it for every other choice. Without a city, the calculator uses the
region's median city and DCMOC the median US city times the country index.
Calculator, 1 MW in Virginia: $17.01M → $12.56M. With no city: $5.90M → $11.34M.
DCMOC default (2.5 MW, Indonesia): about $15.2M → $20.3M; the presets rise about 30 %.
The region list no longer shows multipliers the engine stopped using.
Added
models.tco.crf and models.tco.annualizedCapital: capital annualized by asset life, as in the Uptime Institute
TUI3011B worked example (19.1 M$/yr).
Gap file documentation/product-transformation/gaps/dcmoc-capex.md (reference: Koomey et al., Uptime TUI3011B, 2007).
Fixed
Reveals on the homepage hand the element back to its CSS when they finish. A leftover inline transform stopped the
employer card from turning over.
The homepage case cards now take part in the reveal; a forced opacity: 1 !important had kept them still.
Gates
tools/test-capex-reference.mjs (new, wired): F1 F2 D1 D2 D3 pass; 5 failed on v3.53.0 (the two engines differed by up
to 45 % off the basis spec; now 0.12 %).
test-index-depth D14 checks the GSAP reveal; test-index-micro M8 counts an identity matrix as no movement (a planted
−2 px hover still fails); test-index-motion runs on index.html; audit-a11y waits for GSAP tweens and fails a reveal
that never runs (planted: serious in both themes).
v3.53.0MINOR
— Ask about this page, answered only from the page's records
Added
Ask about this page (P21 stage 1). An Ask control on the 15 pages that carry the basis drawer: the 8 Conventional cockpits, the AI data hall and the 6 project calculators. Type a question; up to three answers come back. No AI is involved.
An answer is filled from a registry record: <label> = <value> <unit>. It is <evidence>: <note>. Formula: … Source: …, with Open basis (the drawer on that record) and a link to the closest manual section. A question about a manual topic returns that section.
A question the page's records do not cover returns "Not covered by this tool's records" and a link to the manual.
Matching is by keyword over the page's own records and manual sections, with a curated synonym list that includes Indonesian ("biaya", "emisi", "kenapa"). The index loads when the panel opens.
A value hidden for signed-out visitors stays hidden in the answer.
Failed index load: "Search of this page's records is unavailable" with Retry.
The seven manuals that had no section ids (datahall, capex, carbon, opex, tco, pue, roi) carry them now, equal to the ones the contents rail assigned at run time, so links to a section land on it.
Fixed
The local test server used by the P10 calculator registry builder crashed on a missing file (it sent a 200 head before reading the file). It now returns 404.
Gates
tools/test-explain-ask.mjs (22 checks):
Every cited record id resolves in the registry the page's drawer reads; every manual anchor exists in the manual the page links.
20 fixture questions answered in the top 3 (20/20, floor 18); "why is PUE 1.165" on the AI data hall answers pue.design_day first.
10 off-topic questions return the miss (10/10). 12 held-out questions written after tuning: 11/12 (floor 10); "maintenance cost per year" ranks the OPEX total above the maintenance line.
Slowest answer under 7 ms (limit 50 ms). Every sentence is a template over record fields.
In the browser: the card opens the drawer on its record; Escape closes drawer then panel and returns focus; loading and failed-load states with Retry; a gated CAPEX value withheld; markup in a record renders as text; 390 px; axe 0 serious; the control on all 15 pages at 1366 and 390 px.
RED: coverage threshold 0 (3 off-topic questions answered), a renamed manual anchor (2 checks), innerHTML rendering plus the gating check removed (2 checks).
tools/build-explain-index.mjs --check joins the generator freshness gate.
Not in this release
P21 stage 2 (an LLM answer layer) is BLOCKED on the owner: model, a server-held key, cost, and the risk of fabricated answers.
The palette action "Ask about this number" (P14 link) is not built.
Several AI data hall registry labels are terse ("design day", "pue"); the card shows the record id under the label.
v3.52.0MINOR
— Click a calculator result to see its basis; the engine as an ES module
Added
Basis on calculator results (P10 stage 1). 40 headline results on six calculators (CAPEX 9, OPEX 8, TCO 6, ROI 6, carbon 6, PUE 5) carry a small evidence dot. Click it, or press Enter on it, to open the basis drawer.
The drawer shows the live value, the value at the page's defaults, the inputs as they are now, the formula, the evidence class, the source and which inputs move the result.
It also says what did the arithmetic: an RZEngine function (15 results) or the page's own function, marked "page arithmetic, not in RZEngine" (25).
The registry data/calc-parameters.json (twin js/calc-parameters.js) is built by tools/build-calc-parameter-registry.mjs. It renders each page at its defaults, reads every hooked cell and moves each declared input by 10 % (a list: the next option). The inputs that move a cell are measured, not declared.
Values hidden from a signed-out visitor (CAPEX cost per kW, energy cost and system blocks; OPEX per-kW, per-kWh and component totals) get no dot and open nothing until the page shows them.
On the carbon page the dots exist on the Operational tab only; the other tabs reuse those cells for other quantities.
The drawer refuses to show numbers ("Runtime basis unavailable") when the registry was built from other engine data than the page loaded.
dist/rz-engine.mjs (P19 part A).rz-engine.js as an ES module: import RZEngine, { models, data } from './dist/rz-engine.mjs'. It is generated by tools/build-engine-esm.mjs, which wraps the browser file unchanged. A Node import writes nothing to globalThis. A JSDoc typedef lists all 54 model namespaces and their functions; return types are any. Not published to npm (part B waits for the owner).
Changed
The basis drawer opens at its title. Focusing the Close button used to scroll a long record to the bottom.
Static: schema, twin, engine functions exist and the page calls them, every record has a measured input, the six pages load the scripts.
Rendered: a fresh render equals the registry (parity at defaults); the hooks on each page are exactly its records; Enter and Esc; live inputs; the authority and missing-registry fallbacks; carbon tabs; gated values.
RED: a removed hook fails coverage; a changed page default fails parity; with the gating park switched off, B8 fails.
tools/test-engine-esm.mjs (10 checks): the ESM file is fresh, has the same model surface and data, and replays the 436 model calls tools/test-rz-engine.mjs makes (52 models) with deep-equal results. RED: a 0.01 % change to pue.dcie fails the replay; an engine edit without a rebuild fails freshness. dist/rz-engine.mjs is also in tools/test-generator-freshness.py.
Not in this release
P10 stage 2: every other numeral on these pages, and the other five calculators.
P19 part B: an npm package. It is public and permanent, so the owner decides licence, versioning and support.
v3.51.0MINOR
— Cx records for a real project, and the Cx plan exports to MS Project and calendars
Added
Cx record workspace (P06 stage 1). The Records panel on the Cx Analysis tab now has Demo | My records | Import file. Demo stays the default and is read-only. "Start your own set", or a validated asset-register CSV, creates a local record set. It is kept in this browser (rz_cx_records_v1), shown as local · rev n, and exported or imported as JSON.
New issue, Edit, Close and Gate decision are native dialogs.
Every write goes through one pipeline (js/rz-cx-workspace.js): clone, change, RZCxRecords.validate, then refuse or commit.
Refusal laws R1–R7 and R11 block a write. The dialog shows the law, for example "R4: L2 accepted on A-CRAH-01 without L1 — L(n) accepted implies L(n−1) accepted for the same asset".
Flag laws R8–R10 and R13 commit and are reported.
A blocking issue sets its gate's constraint to blocked. Accepting that gate is refused (R8) until the issue closes.
A gate decision is bound to a revision hash. The signer is a typed, self-declared name.
Other safeguards:
Deleting a referenced asset is refused, with the references listed.
One-level Undo.
A write in another tab reloads this one.
A damaged store is set aside to rz_cx_records_v1_quarantine with a notice.
Sets are capped at 2 MB, with a warning above 1.5 MB.
An imported set with refusal findings is marked invalid · n. It is never auto-fixed.
Cx plan export (P12). The Cx Analysis CSV row gains MS Project XML (MSPDI, also read by Primavera P6) and Calendar (.ics). Both export the dated plan the calendar panel shows.
L0–L6 are tasks, and L2–L4 are summary tasks with one subtask per system, split by the 02.4.8 planning-day share.
Gates are milestones.
FS links carry the lag in work days that keeps every date of the plan, including the witness-slot gaps.
The .ics holds one all-day event per gate (RFC 5545, folded at 75 octets).
Without a project start date both buttons are disabled, and the reason is on the button.
Samples are in documentation/product-transformation/05-test-evidence/P12/.
Fixed
R121 motion mock (LTC). Under load, hero words that were already drawn were hidden again. Chrome records its paint entry late, so the motion script took words on screen for unpainted ones. The mock now marks its first frame in <head>, and js/rz-index-motion.js counts that mark as painted. The new law T8b repeats T8 with paint entries hidden and the CDN held 800 ms. The old tree fails T8 (33 drops) and T8b (68).
A short form dialog on a phone is centred and fits its content. The first cut inherited the full-height side sheet of the block-detail dialog. The new gate measures an open dialog; the first version of that check measured a closed one and passed.
Gates
tools/test-cx-workspace.mjs (32 checks; Node) — 13 law fixtures, plus referenced delete, revisions, round-trip, quarantine and cap. 5,000 records validate in under 200 ms. RED: with R4 dropped from the refusal set, 1 check fails.
tools/test-cx-workspace-ui.mjs (browser) covers:
Demo read-only, and a set started from a register.
R4 and R8 refused in the dialog.
Issue create, edit and close, plus Undo and reload.
Export → import is deep-equal in a second browser.
An untrusted import renders as text.
Quarantine, two tabs, and a centred dialog at 390 px in both themes.
RED with R4 dropped: 2 checks fail. RED with the side-sheet dialog: the check fails.
65 tasks and 61 links whose lags reproduce every successor date.
Escaping, plus folding and UIDs in the .ics.
A DOMParser round-trip in the page whose counts, durations and links equal the plan.
2,000+ tasks export in under 500 ms.
RED: dropping the lags fails 3 checks, folding at 76 octets fails 2, and dropping escaping fails 4 (the level name "L0 Design & Cx Prep" breaks the XML).
tools/test-cx-analysis.mjs T20: in Demo, "Start a set with these rows" opens New record set; the demo set is untouched.
Not in this release
P06 stage 2 (cloud sets) is BLOCKED. The SQL is delivered in supabase/migrations/2026-11-product-transformation.sql (table cx_record_sets, own-row RLS, 2 MB cap) for the owner to apply.
Amendment C3. RFS gaps still go to the Tier 1 issue register even when a record set is active. A Tier 3 issue needs an asset, and an RFS gap names a system.
MS Project / P6 import itself is untested. There is no licence on this machine; the DOMParser round-trip is the proxy.
Manuals: sustainability, fire-safety and cx rewritten against their pages (Cx Analysis tab, 3D motion); four dead
manual links repointed; three hub-manual entries the hubs never linked removed.
Fixed
Ship gate before release: the manual hub still counted 42 manuals after pillar-standards was removed (now 41); the sustainability drawing's IT glyph rendered its label at 7 px (the IT node now uses the rack symbol); the sustainability generator now applies the site nav register to the menu it inherits from the fire pillar; two manual descriptions rewritten to finish inside 160 characters.
Cx 3D: callouts overflowed and overlapped at 390 px (now one short column inside the stage); the right callout
column covered the equipment it named at 1440 px (columns pick near / top / bottom by fewest anchors covered);
the Explode factor could go negative (rAF time vs performance.now()); a throttled rAF could strand the camera
half-way (every ease has a landing timer).
Removed
pillar-standards.html and its manual (K23, owner: unclear page); every inbound link repointed.
See standarization/CX_ANALYSIS_STANDARD.md §3 (motion), §3d, TELEMETRY_DOCS_STANDARD.md (Manual currency),
DIAGRAM_ENGINE_STANDARD.md §7.
v3.32.2PATCH
— "The commissioning program as a five-deep tree: an exact engine under the L0–L6 editor"
Owner K28 (2026-11-08) asked for a "plan editor L0-L6 … breakdown … sampai 5 level kedalaman … atau bisa hanya set
atasannya … perhitungan complex ada di engine … super accurate". This release ships the engine half. The editor UI is
built on it in a later release.
Added
js/rz-cx-tree.js — RZCxTree.programTree(input, overrides). It breaks each commissioning level into
system → subsystem → equipment type → activity (five deep; 1,269 nodes on a 10 MW colo).
Every number is an allocation of programPlan. A level node equals its plan level exactly. Every parent's
cost and man-hours are exactly the sum of its children (largest-remainder split), and a parent's window spans its
children.
Shares follow the handbook:
L2–L4: planning days per system (02.4.8).
L0: L2+L3+L4 days.
L1: factory-tested unit count; only FAT equipment appears.
L5: L4 days.
L6: assets handed over.
Below a system: unit counts from equipScale, with BMS points from bmsComplexity.
Activities are the level's sub-activities for that discipline, on the handbook dependency chain. A chiller's
start-up begins when electrical start-up ends.
Crew on any node uses the programPlan formula.
Edits:
A level-only edit (the existing plan overrides) moves every descendant.
A node edit (weight, exclude) re-shares the node's siblings with the parent total fixed.
Junk edits are ignored.
Every equipment row names its source. Four screening rules are labelled as such: CDUs and manifolds for
liquid cooling only, CRAH terminals for air only, one dry cooler per chiller, two DCIM servers per site.
tools/test-cx-tree.mjs: 128 laws on the 8 engine scenarios plus an edited plan, wired into the ship gate.
It was proven red twice by mutation. Plain rounding instead of largest remainder leaves parents off by $1–2 (T2).
Ignoring dep starts mechanical start-up at the level start (T9).
Fixed
Homepage footer: "DC operations and standards" pointed at pillar-standards.html, which the next release deletes
at the owner's request. It now reads "TIA-942 standards checklist" and links tia-942-checklist.html.
Docs
CX_ANALYSIS_STANDARD.md: a new "Program tree" section covering the invariants, the share table, timing, site
rules, edits and proof.
v3.32.1PATCH
— "Homepage career cards, redesigned calm: one logo height, no black card, no clock"
Owner 2026-11-08 on v3.32.0: the cards read black, big and odd, the "–" bullets looked wrong, and the Bekasi clock
line was not wanted. The new card was designed first in FigCraft (~/Documents/figcraft/index-cards/cards-v2.figcraft).
Changed
Career cards v2. Every card has the same surface.
Logos sit at one height (28 px).
The name is in sentence case, with the period in mono in the corner.
Skills are plain lines with a small square marker.
"Details →" is a quiet text button under a hairline.
The current employer is marked by a 3 px amber top rule and the word "Current", not by an inverted black card.
Hover and focus darken the border only.
A visible section head ("Career · Five employers, 2011 to now") replaces the screen-reader-only heading.
Details turns the card to its CV facts. The shown face sets the card's height, so a front no longer carries an
empty band sized for its longer back (163 px on a phone in v3.32.0). Reduced motion fades instead.
Removed
The [hh:mm WIB] · Bekasi, Indonesia · Independent personal project meta row and its clock script.
The IBM Plex Sans Condensed font request.
Added
New laws in tools/test-index-r079.mjs:
one card background, with the current card marked by a top rule;
no dash bullets;
names in sentence case;
every logo 28 px tall;
no clock row;
no empty band above Details: up to 40 px on a phone, up to 80 px of row alignment on wider screens.
The gate shows 10 failures on live v3.32.0 and passes on this release.
v3.32.0MINOR
— "Homepage: employer cards that flip, a capsule nav, a ResistanceZero footer"
Owner request R079 (2026-11-07, again 2026-11-08): adopt three ideas from the reference shot without rebuilding the page.
Changed
Career row → five employer cards. Each card shows a large vector logo (56 px), the company name in condensed caps,
the years as the big figure, the role and three skills, and a full-width Details button. Hover or keyboard focus
inverts a card; the current employer stays inverted and carries the row's one amber word. Details flips the card to
facts from the CV (scope, work, results, awards): one card open at a time, Esc or Back flips it home, focus follows,
the card never changes height, and reduced motion crossfades instead of rotating. Cargill and Pure Data Centres use
vector marks (assets/logos/); no raster logo is enlarged.
Navbar: the links sit in one capsule, with a dark Contact pill on the right. No hairline, blur or shadow; the
sliding hover indicator and the DC Solutions jiggle are gone. The hero Correspondence button and the contact
section's email button use the same pill. Wording is unchanged.
Footer: Pages, Topics, Based in and Contact over a full-width "ResistanceZero" wordmark. The wordmark is drawn
from IBM Plex Sans outlines (width 75, weight 700) by tools/build-wordmark.py, so no font loads at runtime. Below
it, a legal row (independent personal project, copyright, Terms, Privacy, Changelog). The mono slogan is gone. There
is no phone number and no street address.
Meta row under the hero: [hh:mm WIB] · Bekasi, Indonesia · Independent personal project.
Dates follow the CV: Astra Agro Lestari 2014–2019, Cargill 2019–2020.
Added
tools/test-index-r079.mjs checks 28 laws on the rendered page:
every pre-refit id is kept and every in-page anchor resolves;
the flip works, including with reduced motion;
phone width shows one column, no horizontal scroll and every logo at 56 px or more;
the nav is flat at 64 px, and there is one pill each in the nav, hero and contact section;
the meta row is present;
the wordmark spans the footer, footer links resolve and the footer has no phone or address.
The gate is red on the pre-refit page (0 cards). It is wired into the ship gate together with
build-wordmark.py --check.
Docs
ANTI_VIBECODE_STANDARD.md: one owner-shape exemption. An element carrying data-rz-owner-shape is not counted as
pill or round, on index.html only. tools/test-slop-precision.mjs proves that the same attribute on any other
page still counts.
UIUX_AUDIT_LEDGER.md: R079 is CLOSED.
Removed
77 styles-index.css rules left without a target by the refit: old career cards, footer and outline button.
v3.31.0MINOR
— "The Cx plan behaves like an application: drag a level, see why, undo it, keep it"
Owner, 2026-11-08 (PRD §41 K21, on K2): "Continue, ingat comment 2 dan ekspectasi. Sempurnakan, better" — K2 asked
for an application, not a form: schedule, cost and man-hours per level that can be shifted, with the monthly cost
moving and charted. The plan editor was a table of numbers; PRD v1.1 §17 / §22 / §23 promised more.
Added
Drag on the level timeline (js/rz-cx-workbench.js): drag a bar to move the level (start; effort and cost
unchanged), drag its right end to stretch it (duration; crew needed changes, L5 witness days move cost), 0.1-month
steps, Esc cancels. The same on the keyboard: a focused bar moves with the arrow keys, Shift+arrows change its
duration, Alt for 1.0-month steps. Every change goes through the editor's own override path, so the monthly cost,
cumulative curve, crew histogram, schematic and 3D plant follow it.
Undo / redo, 50 steps (buttons, Ctrl/Cmd+Z, Ctrl/Cmd+Shift+Z or Ctrl+Y); one drag is one step.
Impact panel: this scenario against the approved baseline, every figure a difference of two engine plans
(RZCxPlan.impact, nothing re-derived). Δ finish, cost, man-hours, peak crew, months that change. It says why each
level moved (edited, or carried from the level that moved first, because levels run in series). It warns when a
level is pulled in before its predecessor ends, and lists the months above the site crew cap. It also draws
scenario − baseline per month (amber = more cost that month, grey = less; "no change in any month" when nothing
moved), with a CSV of baseline / scenario / Δ.
Named scenarios (RZCxPlan.scenarios): Save as / Load / Delete, up to 5 per browser (the same cap as the
DCMOC store). Names are 1–40 characters, a repeated name replaces the old one, and a sixth new name is refused with
a reason. A table compares each scenario with the baseline and the editor.
Proof
tools/test-cx-plan-core.mjs: 37 → 66 laws. Red first: 2 failures (functions missing), then one mutation each for
carry attribution, the cap months and the 5-limit, each read red.
tools/test-cx-analysis.mjs: workbench laws W0–W9 with real pointer input (move +1 month = start +1.0, finish +1.0,
total unchanged; stretch L5 +0.5; Esc; undo/redo; Ctrl+Z; keyboard with focus kept; cap; Save as/Load/Delete;
cap 5; no-change chart). 88/88.
Render scan, strict, cx-calculator.html?tab=analysis in both themes: 0 tells; whole tree 215 × 2: 0.
Screenshots of the timeline, impact and scenario panels at 1440 and 390 px in both themes, looked at.
v3.30.4PATCH
— "The render scan is strict: every page reads zero tells in both themes"
Changed
water-system: the KPI band is one framed panel of rows (label, value, basis) instead of five stat tiles; ids,
basis-drawer hooks and state classes are unchanged.
cdu-mini-bms: the alarm bar carries ok / warn / alarm with its state (the green rail at rest is the normal
state), radius 4 px.
tools/audit-slop-render.mjs: a circle (an ISA pump symbol) is not a pill, and a 1 px offset with no blur (the rule
a sticky table column draws) is not a shadow. Red → green on tools/test-slop-precision.mjs (serves its own fixture).
tools/ship-gate.sh: the render scan runs --strict. All 215 sitemap pages read 0 tells in light and dark, so any
new tell in either theme now fails a release (proven red on the fixture).
Docs
ANTI_VIBECODE_STANDARD.md: the scan is strict; each exemption is listed with its proof fixture.
v3.30.3PATCH
— "Cockpits: named transitions, state rails that say which state"
Changed
datahallAI: 21 time-only transition: .15s-style rules (which animate every property) name their properties; the
theme toggle draws a moon or sun in SVG instead of the ☾/☀ glyphs. chiller-plant's switch thumb keeps left and
transform in its transition so it still slides.
dc-conventional: SVG theme toggle, no ⚠ glyph before "Legal Notice", no resting shadow on the light alarm strip.
fire-system: the gauge pin transitions transform only.
water-system alarm strip and pump units, EPMS_Telemetry status boxes: the coloured left rail stays, because it
is a plant state, and the element now carries a state-* class (state-ok, -warn, -alarm, -unavailable,
-running, -available) set by the same code that sets the state.
v3.30.2PATCH
— "Tools page: the separator dot, and no empty band under the head"
Fixed
tools.html: the dot between the three hero facts rendered as "�b7". The v3.30.1 rewrite passed the CSS escape
\00b7 through a regular-expression replacement, which turned \0 into a NUL byte. The file now carries the escape
again; every file changed since v3.28.1 was checked for NUL bytes (none left).
tools.html: a 108 px empty band above each category came from the site-wide section padding; the category blocks
set their own spacing.
v3.30.1PATCH
— "Tools page on one accent, cockpit transitions named, Tier IV shows its nines"
Changed
tools.html: one amber accent instead of 23 per-card colours (mint, green, blue, cyan); no aurora behind the head, no
shine sweep or glow on hover, icons without tinted tiles (the two emoji icons are Font Awesome glyphs), the hero
badges are one plain line, category titles are grey with a hairline.
Comparisons: diesel vs gas, online vs offline UPS, wet vs pre-action, N+1 vs 2N (152 → 0); PUE vs DCiE,
FM-200 vs Novec 1230, ASHRAE vs Uptime, Tier III vs IV (110 → 0). A winner is a word in an Edge column, not a
green fill with a rail.
Infographics: global PUE (35 → 0), build cost by category (41 → 0), sustainability scorecard (39 → 0).
cdu-mini-bms: chrome on one accent (flat head, sans H1, amber controls); instrument cyan stays inside the
instruments. Navbar and footer stay the cockpit family's.
LTC legacy reports draw rows, not tiles (S36: no radius > 4, no gradients, pills or pink wash).
Decorative rails and washes off water-system (state rails kept) and tools.html.
js/rz-article-chart.js: a series may carry an explicit color (a line chart had no way to leave the light-theme
signal colour, which sits outside the amber band).
Shared styles.css: footer links, share buttons, the nav dropdown arrow and the mobile TOC toggle no longer move,
scale or rotate on hover (about 100 pages; the 40 manual/ pages lose their only render tell). The author box is a
grey panel with a 1 px hairline instead of a gradient card with a blue rail.
dc-market-tracker (root-only): radii, blurs, gradients and the red PDF button gone; locked panels are solid, not
blurred. Render scan 35 tells → 0. Its teal accent stays until the owner decides on a deeper refit.
dc-conventional, chiller-plant, fire-system, manual/research-roadmap: hover transforms, large radii, a blurred
restricted overlay and decorative rails removed (alarm state strips kept).
pillar-fire-safety links underline with text-decoration; the datacenter-solutions 3D stage sits on a flat
background.
Shared styles.css: the 31 transition: all rules name their properties; no backdrop-filter glass on the
scrolled navbar, share buttons, gate overlays, cookie bar or open mobile nav (solid surfaces; the search scrim keeps
its 4 px dim blur); dark chart frames lose their resting shadow and sit at 4 px.
Fixed
Tier III vs IV availability estimate printed 4 decimals (every realistic MTBF/MTTR read "100.0000 %") and capped
Tier IV at 99.9999 %, below Tier III for a good component. Now 6 decimals, no cap, plus downtime a year
(8760 h / 4 h: Tier III 6.57 s, Tier IV 3.0 ms).
Global PUE slider said "Good" at load while its own bands said "Average", and compared the resting text with 2026
(1.55) but the moved slider with 2020 (1.58): one reference now.
Build-cost infographic called the $10M build "Baseline Tier III" while its tier table priced Tier II at 1.0× = $10M.
N+1 vs 2N quick table carried a stray "Full access" in a row label.
Docs
UIUX_AUDIT_LEDGER.md: a row per page or family, CLOSED with the gate that keeps it closed; cdu-mini-bms PARTIAL.
ANTI_VIBECODE_STANDARD.md: scope of the register gate beyond calculators and the explorer refit rules; the stale
"cx, RFS, spares pending" note updated.
v3.29.0MINOR
— "Cx Analysis 3D: a light canvas, the engine's plant, detail where the camera is"
Owner comments K16–K18 (harness PRD v1.1 §41): *"Jangan gelap2, kan sudah saya kasih PRD-nya, theme warna dan accuracy
gambar" · "Super detail dan accurate, tapi tetap responsive" · "accurate bener-bener dynamic data dan pintar"*.
Changed
The 3D campus is light in BOTH page themes (K16, PRD v1.0 §24/§38 "light workspace, neutral plant objects"): canvas #F4F6F8, ground #EEF2F6, text #17212F, lines #D9E0E8; .cx-3d-stage re-scopes the --rzt-* tokens so the dark page no longer darkens the instrument. Untagged equipment went from mid-grey #8a97a6 to light #d9e0e8; lights are a bright hemisphere + one sun (the fill light cost 4 fps on software GL and added nothing a drawing needs). Outline hulls stay off dense families — 334 rack outlines at 0.9 m pitch had read as one grey mass. Proven by a GL read-back of the rendered buffer (5 pixels, min luminance ≥ 0.6, mean ≥ 0.75) in both themes: the first run measured min 0.51 with the source already "light".
The plant is laid out from the engine's site basis (K18): RZCxCampus.layout(basis) is a pure function of IT kW, kW per rack, cooling type and equipScale — rack rows and racks per row, CRAH count, CDUs only for liquid cooling (2 per row), chillers, pumps, dry coolers, generators, 150/20 kV transformers and bays. Families are allocated at a capacity and drawn at the engine's count; past a cap the callout says "N of M drawn". Changing the IT load or density on the calculator re-lays the plant without a reload; loop particle speed follows the design flow (air-cooled ⇒ FWS 0 L/s). The note line prints the basis: IT kW, kW per rack, liquid share, FWS and CHW L/s at their ΔT.
Detail where the camera is (K17): every one of the 15 equipment families has a high-detail Blender twin (_hi: rack doors/handles/manifolds, CRAH fans + grille, UPS module slots, transformer fins/conservator/Buchholz/bushings, generator ribs/louvres/stack/silencer, GIS hall louvres/windows/roof fans …) drawn only inside the focused building (LOD swap in focus()), the low mesh everywhere else. Library 30 meshes, 19,144 triangles, 1.26 MB, still lazy-loaded; ≥ 24 fps kept on headless software GL.
Callouts carry the engine count, the one-line basis and the system's state at the viewed month (rack rows · 334 · 6 kW per rack · not started); building labels yield to the callout stack instead of sitting under it.
index.html contact copy: the section heading "Let's Talk Engineering" and the bento button "Contact Us" became "Correspondence" — owner rule 2026-11-07 (no "Let's Talk" / "Contact Us" / hire or availability copy anywhere; the site is a CV and an independent project). Found by the LTC session's render scan.
Added
tools/test-cx-campus-basis.mjs (19 laws, DOM-free): counts, CDUs, flows, capacity text and callout text follow the engine basis; palette constants are the PRD's. Wired in ship-gate.sh. RED on the previous module (no layout()).
tools/test-cx-glb.mjs: one low + one _hi mesh per family, per-mesh budgets for the twins, shade floor 0.40, CX_GLB= override for red proofs (RED on the v3.28.0 glb: no twins, shades 0.25–0.30). tools/test-cx-analysis.mjs +8 laws (K16 probe both themes + stage background, K17 LOD on/off, K18 counts = engine, air ⇒ no CDU, callout text, IT-load doubling re-lays).
DCMOC Tier 2 groundwork (source only, the app is not rebuilt in this release): dcmoc/src/lib/cx-plan.ts is a bridge to the SAME js/rz-cx-plan.js the site uses (browser: a deferred script in layout.tsx; Node: the rz-engine.ts eval('require') idiom) — plan and per-system status are never ported, so DCMOC cannot drift from cx-calculator.html; dcmoc/src/store/cxSchedule.ts holds the overrides, site crew cap, viewed month and up to five named scenarios (owner default (c)), clamped by the site's own sanitize, versioned payload, in-memory on a storage failure. Gate tools/test-dcmoc-cx-plan.mjs (19 laws: sanitize parity on 6 fixtures, plan + status deep-equal on 9 input × override pairs at 5 months each, store clamps/refusals/persistence/scenario cap/reset, layout tag) runs the TypeScript directly on Node 24 through tools/lib/ts-resolve-hook.mjs; RED when the site module is swapped for a copy with one planning-day value changed. js/rz-cx-plan.js now binds to globalThis under CommonJS (its this was the original exports object, replaced by module.exports, so compute() could never see an engine in Node).
Tier 3 vertical slice, record layer (source + data, no panel yet): js/rz-cx-records.js validates a commissioning record set against the owner PRD v1.0 invariants — unique ids/revisions, same-domain connections, accepted gate ⇔ accept decision with the same revision hash, L(n) needs L(n−1), FR-TST-01/02/03 (empty mandatory step, pass ≠ accepted, operator ≠ approver), open blocking issue ⇒ gate blocked, scenarios never touch actuals, allocations sum to 100 %, observations carry both times and never a Cx level — and overlays the ACTUAL accepted level per system on the Tier 1 plan status ("3 of 10 L3", "Model not verified (2 of 10)", "no register" instead of an invented count). importRegister rejects rows with a reason (duplicate tag, unknown system_id, missing parent, column count). Demo set data/cx/demo-records.json (one hall, 10 CRAH). A Records panel on the Cx Analysis tab (js/rz-cx-records-ui.js, under Status by system) shows planned vs actual per system with its denominator, the invariant findings (0 on the demo set, each naming law + record otherwise) and a CSV register import that lists every rejected row with its reason and adds the accepted rows to the set in-page. Gate tools/test-cx-records.mjs 28 laws, each RED on a mutated copy; tools/test-cx-analysis.mjs +4 laws on the panel. Discovery document: harness TIER3_DISCOVERY.md.
standarization/CX_ANALYSIS_STANDARD.md — the first standard for the Cx Analysis family: one source of numbers, four-field status, the 3D palette/LOD/basis rules, the seven gates and their red proofs, the K-ledger pointer. Four v3.26–v3.28 gates had been in no standard.
v3.28.1PATCH
— "LTC lab: charts that fit a phone, reports on one register"
Changed
ltc-system-modelling-lab.html — settings sit under the step they shape: Load 8, Capture 6, Loop 7, Rejection 9, Cost 5, Check 2. Design settings are in view; preset and tuning settings sit behind "More · n". The inputs are moved, not cloned, so ids, listeners, tier chips and screening only marks keep working; the old "Advanced · auto" panel keeps only calibration, guardrails and run settings.
Report compare card states its verdict per row (better / worse / same; scale for extensive figures when the IT load differs) and refuses to compare a plant run with a screening run.
Pareto scatter is drawn at its panel's width (redrawn on resize), so on a phone it no longer scrolls sideways inside a 360 px min-width svg with 8 px text; its legend is an HTML row under the chart (it sat off-screen at 390 px); status lines read "1 point, 1 feasible · select a point or row" instead of points=1 | feasible=1 | ....
Legacy audit-table reports (single, compare) on the lab register: 95 navy, slate, blue, teal, green and pink literals became greys of the same luminance; series take three distinct greys; the heatmap key reads "Grey >= 85".
The Glossary nav link is no longer inline teal.
Fixed
Pareto scatter with one point repeated its axis tick ($24.89M ×6); a degenerate span is centred in ±1 % of the value.
Gates
tools/test-ltc-lab-steps.mjs laws S29–S35 (distinct ticks, compare verdicts, nav hue, settings under their step, readable legend, scatter fits a phone, legacy report register), each proven red on the previous tree.
Docs
Standards brought in line with v3.27.0 – v3.28.1 (owner: every standardization change recorded in standarization/): ANTI_VIBECODE_STANDARD.md (LTC render gate and chart rules, audit-slop-render.mjs, the gradient-hides-contrast lesson), KNOWLEDGE_LABS_STANDARD.md v2 (per-card accents withdrawn, LTC lab workflow contract), ACCURACY_VALIDATION.md (Rule 1 on the LTC lab, golden fixture), AGENT_HARNESS_STANDARD.md (release-environment ledger E1–E4), UIUX_AUDIT_LEDGER.md CLOSED rows 21–23. The four gates added in v3.27.0 had appeared in no standard.
v3.28.0MINOR
— "Cx Analysis 3D: machines with shapes, drawn like an isometric"
Added
assets/cx/campus-equipment.glb — a 15-family equipment library modelled in Blender 4.5 (headless, tools/cx-assets/build-cx-assets.py): racks, end-of-row CDUs, CRAH / fan walls, UPS, unit substations, LV switchboards, BMS racks, fire panel + agent, chillers, FWS pumps, dry coolers, generators, 150 kV bays, 150/20 kV transformers (radiators, bushings), 20 kV GIS hall. Y-up, metres, origin on the floor at the footprint centre; 5,812 triangles for the whole library; per-face shade written as a FLOAT_COLOR attribute so the exporter does not darken it. The build is deterministic — the gate regenerates the file and compares bytes.
The 3D plant loads the library and swaps every family's geometry in place (js/rz-cx-assets.jsloadLibrary via GLTFLoader; the primitives stay as the fallback while it loads, data-cx-assets="primitives|glb" on the stage). Status tint, picking, dimming, roof lift and stats are unchanged.
Isometric presentation: true-iso campus preset (camera at [d, d, d], zoom 1.3), flat shading with a low front-left key light and a right fill so every box reads in three tones, an unlit slab plate (top + dark side) instead of a floor plane, theme-aware edge lines on every building, parapets, roof plant, louvre bands and doors on the halls, inverted-hull outlines on every unit of the focused building.
Callouts: focusing a building labels each equipment family in it (family · count [· level], mono, tag square) in two columns at the stage edges with a dashed leader and a dot on the unit; they clear when the campus returns.
Gates: tools/test-cx-glb.mjs (110 laws: names, footprints inside the declared sizes, triangle budgets, unit normals, colour range, byte-for-byte regeneration with --regen) over a 60-line Node GLB reader tools/lib/glb-read.mjs; tools/test-cx-analysis.mjs +6 laws (library replaces 15/15 within 20 s, ≤ 300 draw calls, ≥ 6 callouts with equal leader lines inside the stage, callout texts, callouts gone on the campus, hulls only on the focused building).
Changed
js/vendor/three-r170.min.js (RZ3 bundle) rebuilt from three@0.170.0 with GLTFLoader exported (61 → 62 exports, +69 KB; tools/three-vendor-entry.js). The hub 3D and DH01 pages load the same bundle and keep every export they used.
The six Cx gates are now wired into tools/ship-gate.sh (plan core, calendar, io, assets, glb, Cx Analysis browser). They existed since v3.26.0 and were run by hand only — the same "gate exists but is not wired" defect the gate-wiring test guards for CLAUDE.md audits.
v3.27.0MINOR
— "LTC lab: every figure from the plant, one step at a time"
Added
ltc-system-modelling-lab.html — a seven-step thread (Load, Capture, Loop, Rejection, Cost · carbon, Check, Review) in the left rail. Each step shows its own result line, outlines its equipment on the system overview, and opens an arithmetic strip that substitutes the engine's own numbers into each equation (holds when the substitution reproduces the engine value). Review compares the current design with the last saved scenario, recomputed from its saved inputs (better / worse / same, scale when the IT load differs). The right rail and the dashboard tab follow the step (1–5 Results, 6 Charts, 7 Report; ARIA tabs). Auto-run switch in view; Recompute only when it is off. Design status is a fact list that states its basis.
Settings the plant does not read are marked screening only (17 inputs: they move the legacy risk index, not the heat balance).
Gates: tools/test-ltc-lab-steps.mjs (laws S1–S28 at 1440 and 390 px, each proven red on the old tree or a mutation), tools/test-ltc-lab-golden.mjs (12 scenarios × 28 rendered outputs against a fixture), tools/audit-slop-render.mjs (computed slop tells per page: rails, pills, radii, blur, gradients; report only until a baseline is recorded), tools/test-glossary.py (G1–G4). All four wired into tools/ship-gate.sh; the LTC browser gates start their own server on 8166 and stop it by PID.
Fixed — LTC lab accuracy (figures that mixed the plant and the legacy model)
Primary sliders show the values the model runs on, also after a programmatic set; IT range 0.5–200 MW, capture 0–99 %.
Pump figures name their loop (TCS rack-side, FWS facility, both); the Results card, fx traces, Equation Inspector, PDF equation lines, tornado, impact matrix, fix suggestions and the operating-scenario readout all read or perturb the plant (LTCLabPlant.traceOf / equationLines / stepInput / computeLike).
Optimize To Target and Calibration Fit write back only changes that lower their objective, at the precision the form shows, then recompute — the shown figures are reproduced from the form (calibration had rewritten 12 inputs, 10 with no effect). With "apply" off the form is untouched and the results keep the form's model. The scenario row names its source (Optimized, Calibrated, Pareto point).
Pareto "Apply Best Feasible Point" compared the point with itself (baseline read after the apply); candidates no longer write float noise (1.3900000) into the form.
Energy Sankey adds up to the plant's facility power (it invented aux = 3 % of IT, dropped the TCS pumps and showed fans as 0 kW: 131,039 kW for a 129,750 kW plant).
The step strip refreshes on an ltc:rendered event, throttled, so it no longer shows the previous model after an edit.
Changed — off the slop register
LTC lab chrome: one accent, no tiles, neutral --text-secondary (was slate), status pills neutral / amber / red-as-state, 56 slate and navy panel fills removed, solid hairlines instead of dashed boxes; chart series amber and grey (red only for a high-risk point); result tables keep each figure on one line and scroll inside their frame on a phone; accordion titles no longer include their heading's tooltip text.
Glossary (EN + ID): cards are panels with a 1 px hairline, links hairline chips, no hover slide, blur or aurora; six article-28 terms filed under their own letter. Light theme gets a light hero: the dark one put the navbar links at contrast 2.27, which axe could not measure while the hero was a gradient.
cdu-calculator, infographic-dc-sustainability, asean-dc-report-2026, achievements: rails, pills, round radii, blur and gradients removed (render scan before → after: 8/13/1, 12/4/1, 4/17/1/1, 1/32 → 0); CDU KPI chips state the band verdict as text. test-cdu-calc stays 42/42.
Manual family callouts: one panel with one hairline.
Notes
WI-HARNESS-002's last criterion (A15) closed in v3.24.36; this release carries the LTC work item WI-LTC-LAB-001 under the same harness (per-criterion checkpoint, owner-request ledger R001–R088).
v3.26.0MINOR
— "Cx Analysis: the handbook's programme as an app"
Added
cx-calculator.html — a second page tab, Cx Analysis, beside the calculator (the register's .tool-tabs row; deep link ?tab=analysis). Everything on it is computed by RZEngine.models.commissioning.programPlan from the same inputs as the calculator (grand total parity within 1 %), with the Cx Field Manual as the source of every level, gate, tag and duration (ch 02 §2.4–2.7, ch 07/11/28/30, linked per row).
Plan editor: L0–L6 start / months / crew / shift, successors carried serially, cost delta and finish delta against the approved baseline, site crew cap, save / revert / reset (localStorage, foreign payloads ignored).
Timeline + month scrubber: grey baseline bars under amber scenario bars, a month cursor, play.
Status by system: 17 systems in three blocks, tag per month from the gate ladder (red L1 … white L5), in-test and changed-this-month marks, counts by tag.
Giant schematic: three sheets (A electrical 150 kV → 20 kV → 400 V, B mechanical TCS/FWS/CHW P&ID, C controls dependencies) drawn by RZDiagram with IEC/ISA symbols, node colour = tag at the viewed month, halo on a gate passed that month, click a node → inspector (gate table, handbook ladder), zoom chips, pannable at 390 px.
3D plant: a representative campus (hall, chiller/FWS plant, dry-cooler yard, generator yard, 150 kV switchyard) in three.js, equipment families tinted by their system's tag at the viewed month, particle flows for 150 kV / 20 kV / 400 V / FWS supply-return / TCS / CHW with speed ∝ design flow (labelled "design simulation"), click a building to lift its roof and show its equipment, presets Campus / Hall / Plant / Substation, breadcrumb, pause. Loads three only on approach after load, renders at 30 fps when idle, steps pixel ratio down on a weak GPU, says so in one sentence when WebGL is absent, static under reduced motion. Every equipment family is a shaped asset, not a box (js/rz-cx-assets.js): screw chiller with its two shells, compressors and flanged nozzles; end-suction pump set; V-bank dry cooler with fans; containerised genset with silencer and stack; 150 kV bay with post insulators and CT drums on a portal; power transformer with radiator banks, conservator and HV/LV bushings; unit substation, LV board, UPS, CDU, CRAH, racks, OT racks, fire panel + agent cylinders, GIS hall. One merged, vertex-shaded geometry per family on one InstancedMesh, so the draw-call count is unchanged and the status tag still tints the whole unit.
Calendar, witness slots and extension lines (Tier 1.5): set a project start date and the plan is laid on work days (Mon–Fri, no holiday calendar) by RZEngine.models.commissioning.calendarPlan; L4 and L5 take the Monday of the first full Mon–Fri week on or after the previous gate, so a slipped gate can cost a month and the row says which constraint did it (an edited start is a work-day shift from that level's dated baseline start, never an absolute date). Release date and delta in days against the baseline; extension lines (standby per work day between two dates or the two releases, lump sums on a date) bucketed by month with the largest-remainder rule; EAC = actual + ETC at the data date. CSV import of plan overrides and lines with per-row rejection reasons; CSV export of plan, monthly, status at the viewed month, calendar and lines; the page's PDF export gains a Cx Analysis sheet (KPIs, plan, monthly profile, status by system, dated windows, schematic snapshot). Saved with the scenario (store v2).
Charts: monthly cost (baseline grey vs scenario amber), cumulative cost, crew on site against the cap — one accent, grey for the baseline, as the tool register requires.
js/rz-cx-plan.js, js/rz-cx-schematic.js, js/rz-cx-campus.js, js/rz-cx-assets.js, js/rz-cx-io.js, js/rz-cx-analysis.js, js/rz-cx-ext.js, css/rz-cx-analysis.css; gates tools/test-cx-plan-core.mjs (37), tools/test-cx-schematic.mjs (13), tools/test-cx-assets.mjs (142 shape laws: footprint, floor, height use, unit normals, per-type triangle budget, determinism, campus wiring), tools/test-cx-calendar.mjs (33: the PRD §21 fixture — baseline, a slipped L3 losing the L4 slot, standby + lump lines, EAC, clamping, weekends, largest remainder), tools/test-cx-io.mjs (29), tools/test-cx-analysis.mjs (53 browser laws incl. no-WebGL, reduced motion, ≥ 24 fps under software GL, 390 px fields ≥ 16 px, register bar ≤ 6 controls, the calendar float: +11 work days on L3 is absorbed by the L4 witness slot, +33 crosses a month).
rz-engine.js: commissioning.calendarPlan(spec) — UTC work-day calendar, serial FS, witness slot rule, dated overrides with clamping explained, extension lines, EAC.
tools/test-rz-diagram-engine.mjs T3 now reads the page's linked local stylesheets as well as the inline styles when it checks that every var(--…) token a diagram uses is defined; proven red (1/129) with one token removed, green 129/129 restored.
rz-engine.js: calendarPlan levels may carry baselineWorkDays / baselineCost / baselineManHours; the baseline schedule is laid on those, so a duration or cost edit now shows as a release slip and a cost delta (before, the baseline silently followed the edit and a +1 mo level reported "on the baseline date" with no standby cost). RZCxIo.calendarSpec fills them from each level's approved baseline. Laws: calendar 37, io 34 (both proven red first).
Fixed
cdu-calculator.html — NPSHa on the closed-loop suction basis. The page subtracted the whole secondary-loop dP as a suction loss, so the DN50 default read a −24 % NPSH margin; in a closed pressurised loop the loop dP is pump head, not a suction loss. NPSHa now uses atmospheric + loop pressurisation + static head − strainer dP, with the basis in js/cdu-model.js (ILLUSTRATIVE, shown on the card and in the CSV); two red-first laws in tools/test-cdu-calc.mjs (42/42).
cx-calculator.html — the Cx Analysis tab strip inherits the register's .tool-tabs type (sans 13px, amber underline); the mono-uppercase override on its buttons was the §A34 tell (serif-display-chrome) and is removed.
Cx CSV: a leading =, +, - or @ in a text cell is neutralised with a leading apostrophe (plain numbers untouched); a duplicate line id in an imported lines CSV is reported as a rejected row instead of being dropped; the calendar export's release row now matches the header and a baseline_release row carries the baseline date and total. The witness-slot checkbox, once cleared, stays cleared when the start date changes (slots default on only from an empty date). A calendarPlan error is shown as its message instead of "engine without calendarPlan". The PDF sheet's "As of" date is the local date, not UTC.
v3.25.0MINOR
— "DCMOC: five client-ready projects, and a site selection that is a process"
Owner: "provide prefill use case 4-6 scenario yg sudah diisi optimal dan saved, agar saat saya present ke client langsung bisa di show. Improve bagian selection process dan criteria. Perbaiki workflow alur yg nggak sesuai atau nggak nyambung." Audit first (file:line in the work-item plan), then the fixes.
Added
Presentation presets (dcmoc/src/lib/presentationPresets.ts): five complete projects seeded once per browser and listed in Projects with a preset badge — JKT-AI-40 (40 MW liquid GPU campus, Tier III, 2N, Greater Jakarta vs Batam vs Johor), SG-COLO-20 (20 MW in-row colocation, Singapore vs Johor vs Batam), SBY-DR-5 (5 MW air-cooled financial DR, Surabaya vs Jakarta), CYB-HPC-15 (15 MW rear-door HPC, Cyberjaya vs Penang), BTM-HYPER-100 (100 MW phased 25/25/50, Tier IV, 2N+1, Batam vs Johor vs Chonburi). Each fills Requirements 1.1–1.5 (brief, workload mix, growth, SLA, priorities, budget = 1.05–1.10 × the engine CAPEX so the budget rule passes), the capex questionnaire, the architecture profile, the criteria profile and 2–3 candidate sites with screening attributes. Applied through the stores' own setters, snapshotted with the canonical takeSnapshot(), the live configuration backed up and restored, user projects never touched (ids demo_*; skipped when fewer than 5 of the 10 slots are free). The same five are saved as scenarios with engine KPIs, four pre-starred, so Scenario Comparison opens populated. Projects page: Load presentation presets / Reset presets.
Site selection process (SelectionProcessPanel.tsx, lib/site-screening.ts): the Site Intelligence strip is now the process — 1 Criteria → 2 Candidates → 3 Screening → 4 Scoring & Compare → 5 Decision. Criteria = a weight profile (engine DATA.site.weightProfiles: balanced = the previous default, AI training, colocation, enterprise DR, hyperscale — each stated with its rationale) and must-have gates (engine DATA.site.gates: utility capacity ≥ IT load, flood not extreme, PGA ≤ 60 %g, permitting ≤ 24 months, water stress ≤ 3 when the cooling rejects heat through water). Screening = a gate matrix per candidate with the observed value and its source (site attribute or country baseline; unknown never fails). A screened-out candidate keeps its score for reference but leaves the ranking and can never be Overall Best or Recommended. Decision = recommended site, its three largest weighted drivers, weakest factor, runner-up gap, unconfirmed gates, and Adopt — which writes site status = shortlisted and the city into Requirements 1.1 and sets the project country. Criteria and the adopted site persist with the sites and inside a project bundle.
tools/test-dcmoc-presets.mjs — fresh-profile gate: 5 projects + 5 scenarios seeded, live configuration untouched, every scenario OPEX/staff/CAPEX/PUE > 0, each preset opens with Requirements completeness 100, SBY-DR-5 shows Jakarta screened out (flood: extreme), a profile change moves the engine score, Adopt lands in the requirements store, console errors 0, screenshots for the contact sheet.
Tariff model — MRC/NRC defaults by DC business model (rz-engine.jsDATA.tariffs + models.tariffs, dcmoc/src/lib/tariffs.ts). Owner: "improve pre determine MRC dan NRC tariff pada DCMOC … dengan beberapa type business DC misal GaaS … deep research dulu … harus accurate." Six business models — retail colocation, wholesale colocation, hyperscale / build-to-suit lease, high-density GPU colocation, GPU-as-a-Service, enterprise owner-occupied — each with its own NRC, take-or-pay, escalation, typical term and cross-connect add-ons. MRC = market index × block-size factor × model factor: the index is CBRE Global Data Center Trends 2026 (250–500 kW asking; Singapore $403 avg, Tokyo $280, Hong Kong $295, Sydney $188, Frankfurt $235–265, Chicago $200–230, N. Virginia $173, Silicon Valley $200–275, Dallas $160–185, Phoenix $175–215, São Paulo $130–190, Querétaro $250–270, Santiago $190–210; Johor $110–150 and Bangkok $120–165 from secondary trackers, flagged), DATA.markets[*].coloPrice (2025 basis, flagged) for the rest, CBRE NA average $196 as the last fallback; the block-size curve is CBRE North America H1 2026 ($204.69 → $183.75 → $175.00 → $165.31/kW·mo for 250–500 kW, 500 kW–3 MW, 3–10 MW, 10 MW+), cross-checked against Digital Realty's Q2 2026 signed base rent (0–1 MW ≈ $280/kW vs >1 MW ≈ $157/kW). GPU-as-a-Service is priced per GPU-hour (CoreWeave / Lambda / Nebius list prices 2026-11 and Sep-2026 market medians, reserved / on-demand / marketplace bands per SKU: H100, H200, B200, B300, GB300 NVL72) and converted to an IT-kW equivalent through the NVIDIA node power (DGX H100/H200 10.2 kW, DGX B200 14.3 kW, DGX B300 14 kW, GB300 NVL72 142 kW) and the DC-AI fabric/OOB/storage overhead (1.08); the GPU fleet (count, nodes, hardware CAPEX band from 2026 price guides, 4–6 yr life) is written to sim.itHardwareCapexUsd and carried in every DCF on top of the facility CAPEX. Add-ons registered with sources: cross-connect NRC/MRC, remote hands (GSA $157.11/h), TeleGeography 100 GigE transit medians, power overage 1.5–2×. Every number sits in DATA.sources['tariffs*']; nothing is advice.
DCMOC: Requirements 1.5 gains the Revenue model selector (+ GPU SKU / pricing mode / utilisation for GaaS) with the resolved tariff and its basis lines; the Financial module's MRC, NRC and take-or-pay AUTO fields read the same resolution (per-field override kept), a Business model block sits above NRC, and the GaaS fleet line states the hardware CAPEX that the DCF now includes. Presets carry their model: JKT-AI-40 = GaaS (H200 reserved, 70 %), SG-COLO-20 = retail, SBY-DR-5 and CYB-HPC-15 = owner-occupied chargeback, BTM-HYPER-100 = hyperscale lease.
Research memo with every figure and URL: WI-DCMOC-PRESENT-001/TARIFF_RESEARCH.md (rz harness state), summarised in standarization/DCMOC_DCOS_STANDARD.md §7c.
Fixed
Saved scenarios carried OPEX $0 and staff 0 (Shell quick-save, Architecture save) — the Scenarios table and the comparison cost score normalised on a zero. One helper, lib/scenarioSummary.ts, now computes the summary the Dashboard shows (models.opex.fullBreakdown ÷ 12, effective headcount, DATA.pueMatrix, IRR/NPV screening).
Results "Requirements" dimension could never reach 100 %: the intake hardcoded useCase: 'ai' and never sent deadlineMonths, so completeness topped out at 5 of 6 fields whatever the brief said. Both call sites (Results page, value-trace) now send the brief's use case and the months to COD.
Opening a project left the previous country: restoreSnapshot never restored selectedCountry; a Singapore project opened as Indonesia until the capex store reconciled. The country now restores first.
Scenario Comparison lost its ★ set on every reload — the comparison ids persist (scenario payload v2, v1 migrates).
Nav: engine 13 "AI Decision Engine" was an empty group — it is the Results & Decision page; Phased Finance moves from Construction to Financial.
Every Results / Scenario / Executive IRR used $280/kW·mo whatever the project said: finScreening read DATA.decision.revenuePerKwMonth, a key the engine never had, so the ?? 280 fallback always won while the Financial page showed $150–185. The screening now reads the sim revenue basis (the tariff model writes it) and adds the GaaS hardware CAPEX to the capex it discounts. The sim store's revenue clamp rises from 500 to 5,000 $/kW·mo so a GaaS kW-equivalent (≈ $1,000–2,000) is not silently cut.
Measured
Engine: test-rz-engine 814/0 (36 tariff worked examples, proven RED on the previous engine: TypeError: Cannot read properties of undefined (reading 'businessModels')), calibration 19/0, enum coverage 25 clean, reference parity 155/0, value bindings 85/0, catalog regenerated (243 functions, 54 namespaces), DCMOC tsc clean, hardcode audit 0, tooltip coverage no gap.
Worked example (engine): GaaS H100 reserved $3.44/GPU-hr × 70 % × 730 h ÷ 1.377 kW/GPU = $1,276.57/kW·mo against wholesale Jakarta 5 MW $136.76/kW·mo — the GPU fleet ($944 M for 29,048 GPUs at 40 MW) is the difference the client has to see.
v3.24.45PATCH
— "Accent, everywhere it was blue"
Owner: "continue, masih banyak task yang belum kamu lakukan dengan sempurna". The audit-scope rule came from the LTC session: "exempt only what actually has findings, with the finding named."
Changed
Journal hub meta and legal notice (css/rz-article-dark.css, folded in from the LTC session): .jhero p no longer out-specifies .jmeta, so the "Insights · 36 articles · …" line renders at its .78rem mono size (was 18.9 px sans), and .jnotice drops to .8rem/1.55 muted with 12px 14px padding (was 16.6 px) on articles, insights, future-forward and geopolitics.
article-30 cover: a 1600×900 hero (assets/article-30-hero.webp, a frame from the owner's video) as the cover figure; the journal cards and data/journal.json pick it up (image was null), and the OG card is rebuilt from it.
a11y: article-26's privacy line and disclaimer in light mode read 2.5:1 after the hue sweep; now #646a73 (5.4:1). The a11y gate waits for finite entrance animations before axe samples colour — the hub's staggered .keys li fade-in had been sampled mid-fade (#6B7280 read as #A6AAB2, then #CDCFD4), a flake that depended on load.
Rebased onto v3.24.44 (LTC hub tools): where both releases touched a line, the flat graphite (no gradient, no backdrop blur) and the 4 px radius won; 275 token-only conflicts resolved by the normaliser.
Articles: UI blue in 24 articles is now the editorial accent (var(--rz-art-accent), #E8B563 in dark and #92400e in light). This covers buttons, toolbars, widget borders, tab states, links and the TOC. Chart series, map markers, ramps and named --*-blue data series keep their colours; they are categories, not chrome.
Shared links (styles.css): dark links in the article body, article nav, author bio, references, back link, contact block and calculator disclaimer move from #60a5fa/#818cf8/#a5b4fc to amber. Light disclaimer links move from #2563eb/#1d4ed8 to #92400e/#78350f. The cookie banner's accept button is graphite in light and amber in dark, at 4 px radius. The dark footer border loses its navy cast (#1f2b47 to #2a2e36).
Table of contents in articles (css/rz-article-dark.css): muted numbers and a hairline card on every editorial page, with amber on hover.
Second accents removed (contact sheet): cyan chrome on 22 reading and tool pages, green on the ASEAN report and the PUE infographic, the regional hue of each PLN grid page, red and green Login buttons on article-2/3. One accent per theme. Categorical colour (ratings, scale levels, comparison sides, chart series) is kept, and each kept rule is named in the sweep.
Coloured gradient bands on heroes, calculator panels and CTAs (green, teal, red, blue; dark in both themes) are flat graphite. A CTA is flat amber. The PLN heroes lose their regional glows.
Journal hero CTA (articles 14/16/17/18): the 30 px gradient pill with an infinite pulse-glow is a flat amber button at 4 px, with no animation. The dc-market-tracker premium button no longer pulses.
Navy cast → graphite of the same lightness on pages, styles.css, the cockpit instrument sheet and the PLN dashboard scripts.
audit-vibecode scope (AC2): §A27–34 now gate every scanned file. The only exemptions are a named EXEMPT set and a PENDING map pinned per file and per rule: design artefacts, the homepage, the paper document, and rz-calc-editorial.css, which only changelog.html and a plan artefact load. Owning a page is not a reason to exempt it, so clean calculator pages gate. A PENDING entry whose finding has gone fails as stale-exemption; after rebasing on 3.24.42 it flagged exactly capex-calculator.html and js/rz-scenario.js, which are now removed. An aurora-* rule that only says display: none no longer counts as the tell.
Fixed
Calculator button with no visible label (article-9, dark). [data-theme="dark"] .article-body a (0,2,1) outranked .hvac-cta-btn (0,1,0). On-accent text in the 4 affected articles now wins.
Dark instrument panels in the light theme ([data-rz-dark-panel]). These panels stay dark in day mode but took the day accent #92400e, about 2.6:1 on graphite. They keep the night tokens. A page token that aliases the accent on :root (--ig-color: var(--rz-accent-ui)) is resolved at :root, so it is now re-declared on the panel. Without that, the infographic, ASEAN and PLN heroes showed brown numerals on black.
Pale evidence tiles on article-20 (light). The journal sweep marked tiles inside a [data-rz-fact-list] as dark panels, while the editorial register flattens them onto the page. A fact-list tile is never a dark panel, and the existing marks are removed.
article-14 hero stats: rgba(0,0,0,.3) read as a grey box with pale amber in light; it is now a dark panel.
Compare pages: the hero follows the theme, and the theme toggle works (it was a dead button).
Added
tools/audit-render-register.mjs (AC3) renders every sitemap page in {light, dark} × {390, 1440} and counts, from computed style, navy-band backgrounds, neon, serif chrome, and wash: a band at least 50 % of the viewport wide painted by a gradient with a dark chromatic stop outside the accent. Data encodings are exempt. Out-of-scope pages are listed with a reason.
Reproducible sweeps, each with --check wired into test-generator-freshness.py: sweep-navy-register.py, sweep-pln-register.py, sweep-cyan-register.py, sweep-hero-cta.py and sweep-wash-register.py.
audit-vibecodeanimated-cta also matches jump-to/-btn selectors and pulse/glow keyframes. The 3.24.42 detector fails the new test and this one passes it. A P&ID flow arrow and hover feedback stay exempt.
audit-responsive-layout covers sitemap subfolder pages, and a page that fails to load counts as UNMEASURED and fails.
Blue band (computed style, hue 195–250°, 176 pages × both themes): 1,719 → 1,311. Cyan band (180–194°): 3,022 → 1,204. The residue is instrument state colour, charts and data fills, not chrome.
Button / CTA chrome (tools/sweep-cta-accent.py, 14 pages): every off-accent fill, tint and border on a btn|button|cta rule moved onto the one accent (pink article-15, red mode/PDF buttons on 2/6/7, green calculator buttons on 3/12/17, blue mode buttons on 4 and pln-java-grid, mint 404 return button, tinted CTA blocks on FF-2, 8, geopolitics-3, dc-market-tracker); state, legend, share and destructive buttons kept by name. --check is wired into generator freshness.
Accessibility after the sweeps (axe, 8 pages × both themes): the first gate run found 9 contrast rows the sweeps had created — cdu-calculator active buttons white-on-amber in dark, article-26 engineer note forced to day ink inside its dark panel, article-13 disclaimer links on the night amber inside a day-painted disclaimer and a calculator badge whose fill styles.css strips (background:none !important) leaving white text on nothing. All four causes fixed at the source; privacy.html, a dark-only page, lost the day token set the sweep had injected (dark-coverage gate).
AC5 contact sheet: 124 pages × 4 renders reviewed and sent; 40 pages re-shot after the fixes.
v3.24.44PATCH
— "Hub tools on graphite (R2)"
Owner: "continue, polish more, sempurnakan". Release R2a of the hub-tools plan.
Changed
LTC standards workbench (hub + five modules): the inline token block moves from navy (#0a0e1a, #111827) to the hub graphite (#0b0c0e, #121316, #1b1d22); the OK-state green is the site's status green #6aa588 instead of neon #00E07A; links take the text colour with an amber hover underline instead of cyan. Navy fallbacks in js/ltc-app.js, js/ltc-dh01.js and js/ltc-report.js follow.
LTC modelling lab: the navy primary gradient, the radial aurora, 19 two-stop panel gradients, 9 backdrop blurs and 3 hover lifts are gone. Panels take the card colour, data bars keep their hue as a flat fill, the navbar is solid, and four leftover navy surfaces are graphite. The two gradients left are the red-to-green scale (data) and the loading shimmer. css/ltc-lab.css follows; the KPI strip is one framed readout row with hairline dividers and mono values, with no tinted "good" cells.
The three LTC manuals (modelling lab, standards lab, fire pillar) take the manual family's values: neon #00FF88 becomes the status green. They leave the EXCLUDE set of tools/sweep-site-register.py.
Fire system P&ID: the page no longer loads the Tailwind CDN script at runtime; the 35 utility classes it used are a local <style id="fs-utilities"> block, with 4 px radii. Navy symbol fills (#0c1424, #020a16, #04121c) are graphite.
DC market tracker: the Monte Carlo projection uses a seeded generator (mulberry32, seed 2026010), so the same inputs give the same P10/P50/P90 on every load.
Data hall AI: 20 HMI header gradients and 16 backdrop blurs are gone; panels are flat graphite (light: white). Three gradients remain: the heat scale, a hard-stop line and the hatch pattern.
Cockpits: the dc-conventional sidebar loses its navy gradient, dialog backdrops lose blur and the navy tint, and the skip link on all seven cockpits is graphite instead of navy. The ict CCTV cyan glow and the fuel hero gradient are removed. Engine version pins are unchanged. At phone width the dc-conventional subsystem row keeps each item at its label width and scrolls, instead of letting 5 of 7 labels run into each other.
Cx field manual: 40 source-list entries that pointed at private working files (../case_ledger.json, "working files", research/RF-33.md) now name the MSC-1 example campus described on the manual index.
Gates
tools/test-cx-handbook.py C4 also bans those private paths: red on the previous chapter 18 (2 hits), green now.
audit-vibecode §A27–34: neon-green 17 → 14 files site-wide; no LTC, calculator or manual file is flagged for navy or neon.
v3.24.42PATCH
— "Hub explorers"
Owner: "lakukan update dan enhancement utk semua tools di dc hub". Release R1 of the hub-tools plan: the six static hub pages.
Changed
cdu-hub, compare-air-vs-liquid-cooling, compare-raised-floor-vs-slab, pln-sumatra-grid, network-visualization-hub and ai-engineering-maintenance move to the tool register: hub nav and footer, a plain header (eyebrow, title, one lede, a mono basis line, crumbs), link lists and comparisons as tables, no hero cards, gradients, badges, scroll-top or share widgets. A new explorer kind in tools/refit-tool-shell.py does the transform (--check is a no-op after --apply); page rules live in css/rz-explorer.css.
PLN Sumatra grid: the stat tiles are a fact list ("What the dataset holds"), the map keys are drawn as SVG on the register (substation, plant, data centre keep separate marks), and the page now shows its source and dataset version (RUPTL 2025-2034 / PLN AR 2024, version 2026-06-26, read from the data file at load). The Leaflet zoom control is readable again.
compare-raised-floor-vs-slab no longer loads the TCO calculator's script.min.js.
The CAPEX report styles and the account pill use graphite instead of the navy-grey #111827.
Gates
tools/test-tool-register.mjs covers the six pages (19 in all). Red before: 223 failures across the six. T11 skips auth.js's own sign-in modal only while the page is locked; on a calculator that opens the modal the check still fails (4 of 4 views).
The skip-link family in tools/sweep-site-register.py now includes these pages.
v3.24.41PATCH
— "Shared blue, gone"
Owner: "Continue all" and "setelah itu commit and push". The faint mobile menu button was reported by the LTC session, from the new cx-handbook pages.
Changed
Nav avatar (styles.css, styles-index.css): the 2px blue ring, blue shadow and 1.1× hover scale are replaced by a 1px neutral hairline. On hover the hairline turns amber.
Article table of contents (styles.css): flat panel and 1px hairline cards, muted numbers, amber only on hover. This replaces the blue rail, blue numbers, the blue calculator wash and the slide-and-shadow hover. The "Interactive" badge is amber. Article-1, 11 and 18 lose the page-local blue overrides that beat the shared rule.
Info chips (--rz-chip-info): instrument cyan (#0e7490, and neon #00DDFF in dark) is now identity ink (#3a3f47 / #c4c9d1). The token feeds the info badges, callouts and rail tags on about 50 articles. Status colours (good, warn, alert) are unchanged.
Links in the reading column (css/rz-article-dark.css): plain links outside paragraphs and lists, such as references, table cells and notes, take the amber accent instead of each page's own blue. TOC cards read as text. Reference numbers and list markers are muted ink, inline citations are amber, and dotted term underlines are muted.
Dark reference block (styles.css): seven #60a5fa values become muted numbers, amber links and text-coloured headings.
Light footer: graphite #1a1d23 instead of the navy-to-blue gradient. The copyright line moves from white at 40 % opacity (3.77:1, flagged by the accessibility gate) to #969eaa (6.2:1). Footer column headings had no colour of their own and showed the global heading ink at 50 % opacity, nearly invisible on the dark footer. They now read #969eaa.
Fixed
Mobile menu button almost invisible in light theme.js/rz-mobile-nav.js injects the button with an inline style of white bars (#f1f5f9) on a white-tinted chip. That inline style beat every stylesheet rule, so pages without styles.css (the hub, tools, cx-handbook) showed an empty white square at 390 px. The inline style now carries layout only, and colour comes from theme-aware rules: ink bars on white in light, light bars on graphite in dark, 4px corner. The light rules for the button and the open drawer matched only [data-theme="light"]; they now match any theme that is not dark. Screenshots at 390 px show datacenter-solutions in light (empty before, ink bars after) and in dark, tools in light, and article-5 in dark.
Measured
Computed-style scan, 176 pages × both themes, counting visible elements whose text, border, fill or gradient sits in the saturated blue band. From v3.24.39 to this release:
total: 3,413 → 2,072
nav avatar: 136 → 0
TOC numbers: 291 → 41
plain links: 447 → 146
reference numbers: 133 → 0
citations: 86 → 0
term underlines: 128 → 0
footer gradient: 52 → 0
The rest is mostly map markers, data tags and categorical hues, plus pages the LTC session owns.
v3.24.40PATCH
— "Field manual"
Owner: "warna biru ini dan beberapa component itu terlihat ai design slop" and "handbook by sesi lain sudah selesai … jadikan ke web page itu".
Added
Data centre commissioning field manual, R01, as 33 web pages under cx-handbook/: an index and 32 chapters (parts A–G, 384,159 words, 267 drawings). Each chapter has its own contents list, previous/next links and its drawings inline at the section they illustrate, as framed sheets with a stable #F02-1 style anchor and a full-size link. Two chapters of the source (31 and 34) are interview preparation and are not published; employer references in the rest were removed.
tools/build-cx-handbook.py --apply/--check builds the pages from data/cx-handbook/; the freshness gate now covers it (10 generated artefacts).
The Cx calculator's reference section links the manual: index, process and levels (02), integrated systems testing (28), issue library (29), glossary (33).
Changed
Explain popover on the site register (js/rz-explain.js, 101 pages): graphite panel, 1 px hairline, radius 4, no drop shadow. Linked terms are underlined in the text colour and take the amber accent on hover instead of sky blue. Unit, source and related terms are plain rows ("See also …") instead of pill chips. Table text is left-aligned; only numbers use the mono face. The focus ring is amber. On phones the bottom sheet has square corners.
Cx calculator: choosing a scenario applies it at once (the Load button is gone); results compute on load and on every change (the Calculate button is gone); drop-downs use the sans face, numbers stay mono; the number field has no spinner arrows.
Gates
tools/test-cx-handbook.py: page count = chapters + index; every figure placed exactly once (267/267); every prose figure reference resolves (279/279); zero employer or interview terms. Red on a copy with F02-1 removed (266/267, 277/279) and on the unedited sources (4 hits).
v3.24.39PATCH
— "One blue less"
Owner: "continue", "Continue all", then "setelah itu commit and push". The hub slogan removal was agreed with the LTC session after the owner's "sempurnakan jangan ada kata2 filler".
Changed
Hub footer slogan removed.datacenter-solutions.html keeps "Data-center engineering tools and telemetry models." and drops "Mission-critical infrastructure, documented." The 9 calculators lost the same sentence in v3.24.38, so the phrase is gone from every live page. index1.html is an owner mockup and is not swept.
Decorative light blue (#93c5fd) moved to the accent on 13 pages and in styles.css (53 lines). On articles 1, 2, 4, 9, 13, 14, 15, 17, 18 and 23 this affects article navigation links, headings, quotes, formula text, key-insight labels, the TOC and the cookie link. These lines now use var(--rz-art-accent), which is amber in dark (#E8B563) and dark amber in light (#92400e). Unscoped rules that showed pale blue on white in light theme are now readable too. compare-n1-vs-2n (table head, privacy link), compare-ashrae-vs-uptime (cookie link), tools.html (disclaimer link) and the calculator disclaimer link in styles.css take #E8B563 in dark.
Categorical hues stay: the article-14 perspective tabs, article-13 vendor formulas, article-17 scenario and comparison states, the article-27 lever categories, the compare-ashrae A/B sides, the pln province tags and the maturity ramp.
Skip links on 404, privacy and the 5 pln-java-grid-* regional pages are on the register. They were navy in light and bright blue in dark; now they are graphite in light and amber in dark, with a 4px corner, matching styles.css. This is the new skip family in tools/sweep-site-register.py: 14 old values before, 0 after. The LTC session will add its 13 hub-tool pages to the same family.
Measured
A computed-style scan of 176 pages in both themes, run before and after this release, counts visible elements whose own text renders #93c5fd: 94 before, 28 after. The 28 left are the categorical hues above, ltc-system-modelling-lab and pln-java-grid (LTC-owned).
v3.24.38PATCH
— "Last four calculators"
Owner: "audit total dan perbaiki semuanya", then "continue" — plan cryptic-rolling-sonnet, phase 3. All 13 calculators are now on the tool shell.
Changed
cx, fire checklist, RFS readiness and critical spares move to the tool page shell, refit by hand because each is its own app. Each page now has the hub nav and footer and a plain header (eyebrow, title, one lede, a mono basis line, crumbs). KPI tile strips are replaced by fact-list rows. Tabs and actions share one tool bar. The only fixed element is the nav. Behaviour is unchanged: every JS-bound id is kept, and presets, inputs, Free/Full and Save still work.
RFS: the 18-button mode bar becomes a tab row plus four actions. Spares: the aurora, banner and share rail are gone; the 32 module panes are a tab row on desktop and the existing <select> on phones. cx: the 9-button bar becomes a Free/Full segment plus four actions (Share, Reset, Save, Export PDF); the Gantt stays as a figure.
Copy without filler on all 13 calculators (owner: "sempurnakan jangan ada kata2 filler"). The footer slogan is gone. The long disclaimers are now short notices that keep every point: education and planning only, not a basis for decisions, independent research, runs in the browser, Privacy and Terms. Puffery adjectives and intensifiers are cut, and gate T22 keeps them out. Domain names stay: Kraljic "Leverage", the VITAL/ESSENTIAL tiers, critical spares, and the opex "Comprehensive" maintenance tier.
Fire checklist: the red eyebrow, the mono slogan footer and the boxed source tags are replaced by register text.
Fixed
Spares painted readiness values red (65.0 %, 72.5 %). The value now keeps the text colour, and the row's note cell carries the band as a word ("within band", "watch", "out of band"). Only that word takes the colour.
Gates
tools/test-tool-register.mjs T14 now also fails a red value inside a fact list. It was red on the old spares page (#sum_readiness, #sum_hub_read, 4 of 4 views) and is green after the fix.
T22 fails filler words in a calculator's copy (scripts and styles excluded). Red on the previous release: 47 hits across 12 pages. Green now: 0.
ship-gate.sh runs the register gate on all 13 calculators (was 9).
Agent harness
WI-HARNESS-002: the public standard's §3/§4/§8 and its P1–P10 ledger shipped in v3.24.36. This entry closes the work item's last criterion (A15), which is checked by tools/harness-public-sync.py in the harness repo against this site's origin/main.
v3.24.37PATCH
— "The shared controls, on the register"
Owner: "continue" — WI-WEB-ALLPAGES-001. The whole-site render scan for blue and navy found the controls every page shares still in the old palette, after the pages around them had moved to graphite and one amber.
Changed
Keyboard focus ring. It was steel blue on every page; it is now the one accent: 2px #92400e in light, #E8B563 in dark (styles.css, styles-index.css).
Skip link. The blue skip-to-content chip is ink #1a1d23 in light and amber with graphite text in dark.
Theme toggle. It was a 12px glass pill with a navy hairline that grew and cast a shadow on hover. It is now a 4px square: white with a neutral hairline in light, a faint graphite fill in dark, and the hover changes the border only.
Back-to-top button. It was a filled blue disc with a blue glow on the 53 pages that carry it. It is now a neutral 4px square, with amber on hover.
Manual button (js/rz-manual-fab.js). It was instrument-cyan with a shadow and a lift on hover; it is now a neutral outline with amber on hover. The refit calculators no longer carry it and do not get it back.
"Technical manual" links (.rz-manual-link). The slate box with a cyan hover moves to the register on 20 pages, enforced by a new mlink family in tools/sweep-site-register.py: --check reported 68 old values on the previous tree, --apply rewrote 67, and a second apply rewrote 0.
ai-engineering-maintenance.html had its own cyan, slate and navy palette with gradient fills. Its tokens are remapped to graphite with one amber; the diagram fills are flat.
Fixed
The homepage back-to-top button had no style at all.index.html loads styles-index.css since the CSS split, and that file never carried the button's base rule, so it rendered as an unstyled browser button. The rule is now in both stylesheets.
On 41 pages the back-to-top button did nothing. Every article, the pillars, the comparison pages and the homepage carried #scrollTopBtn but no script that shows it after scrolling or scrolls on click; only pages with their own inline wiring (glossary, the hubs) worked. script.js now wires it once: it appears past 500 px of scroll and returns to the top (instantly under reduced motion).
v3.24.36PATCH
— "Calculators, de-slopped"
Owner: "audit total dan perbaiki semuanya … lepas dari AI design slop atau terlihat vibe coding", then "continue, yang terbaik aja" — plan cryptic-rolling-sonnet, phases 0–2 and 4.
Changed
Nine calculators move to one tool page shell: capex, opex, PUE, ROI, tier, carbon, TIA-942, TCO and fire. Each page now has the hub nav and the hub footer. The header is plain: an eyebrow, a title, one lede and a basis line. The toolbar has six controls or fewer. Results are rows, with label, value and note, and one amber figure per panel. Every figure carries a numbered caption. Full analysis is one row with one Sign in button; locked content is hidden, not blurred. A reference section holds the method formulas, the standards each page computes against, and one disclaimer notice.
Removed from those pages: hero photos, badge pills, the KPI tile strips with coloured rails, the floating manual button, scroll-to-top button and watermark, gradient footers, dashed "how to read" boxes and the duplicate cookie banner (TCO). Chart series are amber and greys unless a colour carries meaning (ROI's red loss years).
The refit is reproducible: tools/refit-tool-shell.py with one module per page in tools/refit_shell/. It rebuilds each page from its previous source, and re-running it changes nothing.
pillar-fire-safety.html rewritten. It has eight plain sections: what burns and where, detection, suppression, lithium-ion, standards, inspection, incidents, and tools. Three tables are filled row by row from data/fire/*.csv, and three diagrams come from the diagram engine. Every number cites a dataset row, the fire engine, a standard or an incident dossier.
Fixed
IG-541 residual oxygen was a fixed 12.5 % whatever the concentration. It is now computed from the concentration: 20.9 × (1 − C/100), which gives 13.1 % at the 37.5 % design and moves when the concentration is overridden (js/fire-engine.js, three new cases in test-fire-calc.mjs).
ROI scenarios multiplied the base NPV by a factor, so with a negative base "Bull" came out worse than "Base" (−$62.1M against −$41.4M). Each scenario now re-runs the yearly cashflow (revenue +15 %, −25 % or +50 %, or the discount rate +200 bp). Bull is now −$8.6M at the default inputs.
The explain panel (js/rz-explain.js) covered its own trigger on short viewports. It now opens below the trigger, or above it, or scrolls inside the larger side.
A signed-in phone overflowed: a long account e-mail widened capex to 461 px and pushed the menu button off-screen. On phones the account button now shows only the initial.
The tier page's IT load drove nothing; it now feeds a "Facility assessed" row. Opex free-tier tables were refilled with dollar figures on every re-render, so free users saw them anyway; they are now hidden.
FM-200's GWP of 3,220 is the IPCC AR4 value; the dataset said AR6.
The refit pages linked css/rz-hub.css and css/rz-tool.css without a cache token, so a returning visitor would have kept the old stylesheet. Both links now carry the site's token, the refit tool emits it, and gate T21 fails any untokened local stylesheet or script on a tool page (red on the previous capex page).
audit-mobile-responsive read only inline styles, though its docstring says linked CSS too, so it scored five refit calculators 2/10. It now reads the page's local stylesheets and recognises the hub footer's one-column collapse. Scores only rise (184 pass, 0 fail); tool pages also cap image width at 100 %.
Gates
tools/test-tool-register.mjs:
T10–T18 check the rendered structure: no tiles, nothing floating, a plain header, a toolbar of six or fewer controls, no second hue, no shadows or dashes, hub nav and footer, no logo tile.
T17 checks behaviour: every JS-bound id exists, presets click cleanly, the first input changes the results, the explain panel opens and never covers its trigger, Free/Full works, Save works.
T19 checks that chart series are amber or grey. T20 checks that a signed-in page fits a 360 px phone.
Each check was red on the old pages: 323 failures across 13 pages.
Wired into ship-gate.sh for the nine refit pages.
tools/test-pillar-fire.mjs: exit 1 with 36 findings on the old pillar page, green on the new one. Wired.
The four calculators not refit yet (cx, RFS, spares, fire checklist) keep their old look. Their computed styles were compared element by element before and after this release: no change.
v3.24.35PATCH
— "The whole changelog"
Owner: "continue" — WI-WEB-ALLPAGES-001. Found by a whole-site render scan for blue and navy: changelog.html was the largest offender, and opening it turned up three defects worse than the colour.
Fixed
89 releases were missing from the public changelog, and the page's main script never ran. One CHANGELOG.md line quotes HTML with a double-backtick code span (`<script src>`). The builder only knew single backticks, so it read two empty spans around a bare <script src> and wrote a live script tag into the page. The browser parsed everything after that tag, up to the next closing tag, as script text: every entry older than v1.21.0, and the tag that loads script.min.js. Rendered before: 834 of 923 entries, oldest shown v1.21.0, one stray script, script.min.js never requested. After: 923 of 923, oldest v1.0.0, no stray script, script.min.js loaded. With the script back, the navbar gains its surface when the page scrolls; before, the links sat transparently over the entries in both themes.
The builder now reads a code span as a run of N backticks closed by exactly N, escapes every < in prose (so a code span broken across two lines can no longer open a tag — two <html … and two other stray tags were doing that), and fails the build if the entries contain any tag its own Markdown rules do not write. The old guard only matched a backtick directly before a tag, so it passed this page. Proven red first: the previous builder with the new guard fails on <explain>, <html>, <pue>, <script>, <style>; the new builder passes.
The light theme was half dark. The template wrote <body data-theme="dark">, so every [data-theme="dark"] … rule in the shared stylesheet kept matching inside the body whatever the reader chose: in light, the scrolled navbar was a dark grey bar over a white page. The attribute is gone; the theme lives on <html> as on every other page.
No menu on a phone. The page hid the menu with display:none !important at phone width and did not load js/rz-mobile-nav.js, which every other page with a navbar loads, so the hamburger never showed. It now shows at 390 px and opens the menu (measured: button 44 px wide, menu 498 px tall when open).
Changed
The changelog uses the site register. It ran its own palette: mint version numbers, links and bullets, cyan and emerald tier pills, a green CURRENT badge, slate-navy cards with blurred glass, an animated mint-to-amber gradient title, and three aurora orbs. Now graphite with one amber: neutral version numbers and tier outlines, MAJOR and CURRENT in amber, 4 px panels with a hairline, flat code chips, a solid title, no orbs. The light theme maps onto the same tokens. The duplicate "Latest version" stamp in the hero is gone (the hero line already states it).
v3.24.33PATCH
— "Contents numbered from one"
Owner: "Continue" — WI-WEB-ALLPAGES-001; defect reported by the LTC session.
Fixed
The article table of contents on a phone started at "0." and skipped 1 on article-29 and article-30 ("0. Key numbers", "2. …"), and article-28 printed 1, 2, 2, 4. The shared builder js/rz-article-toc.js took any digit inside a heading id as the entry number, and the "Key numbers" heading carries an automatic id key-numbers-0. The number now comes from the id only for toc-section-N, section-N and secN ids; every other heading is numbered by its position.
Proven on all 36 pages that load the builder, each rendered at 390 px with the previous builder and with this one: exactly three pages change (28, 29, 30), all to 1…N; the other 33, including the pages whose ids run sec0, sec1…, are unchanged.
The navbar's dropdown and journal links were a rainbow. Every item carried its own inline hue — DC Solutions Hub cyan, CAPEX orange, OPEX emerald, ROI blue, DC Market teal, DC MOC red, Global Analysis red, Glossary teal — and no stylesheet overrode an inline colour, so the menu rendered as up to eight colours on 75 pages. The hues are gone; items take the menu's own text colour. Font weight and size stay (the hub entries remain bold), grey items and the amber var(--rz-restricted) marker are untouched.
Done as one more family in tools/sweep-site-register.py (nav): 627 inline hues came off 75 pages (counted against the previous release); --check fails on the old tree, --apply fixes it, a second --apply rewrites nothing. The family also covers the embed widgets' index and the changelog builder's own template — a first pass missed both, and regenerating changelog.html put ten hues straight back. The 13 calculators are skipped — the LTC session's refit replaces their navbar with the hub navigation, which carries no inline hue — and so are the design mockups.
The header link buttons on the two cockpit landings (dc-conventional, datahallAI) — PRD, Manual, FAQ, Portfolio, DC Solutions — were drawn in cyan, with FAQ in sky blue, Portfolio in slate-purple and DC Solutions in green. They are neutral buttons with an amber hover now; Generate Design, the one primary action, keeps its amber. The FAQ button lost its ❓ emoji (an emoji as an icon is banned by the anti-vibecode standard; these pages load no icon font) and, in the light theme, a leftover teal from an a11y rule. Readouts and state colours inside the cockpits are not touched (family cockhead, 15 values).
The glossary's term counter ("Showing 449 of 449") still printed its two numbers in light blue in the dark theme, from an old contrast override that outranked the accent. It now uses the glossary accent, amber in dark (family lang).
Articles on a phone (reported by the LTC session): the first letter of every editorial article rendered 51 px tall and unfloated, followed by a gap — "A YouTube", "F ire" — in both themes. The register's reset lost to the page-local drop-cap rules, which set size and margin with !important at phone width. The reset now wins: the first letter is 18.24 px, the paragraph's own size, margin 0 (measured on article-1, -12 and -30, both themes).
The table-of-contents drawer that opens on a phone kept the old shell: a blue title, blue hover and active links, a 20 px rounded top and a drop shadow. In the editorial register it takes the article panel colour, a hairline top, 4 px corners, a neutral title, and an amber 2 px rail on the active entry.
v3.24.31PATCH
— "Pillars, widgets and the Indonesian pages"
Owner: "polish more", then "Continue, polish more" — the third wave of WI-WEB-ALLPAGES-001. These are pages the first two waves did not reach: three of them (id/) had never been swept at all.
Measured with node tools/audit-vibecode.mjs --json: 143 monitored findings → 119. What remains is index.html (waiting for an owner-approved mockup), design artefacts kept as evidence, and files the parallel LTC session owns.
Changed
The four topic pillars (pillar-cooling, -power, -standards, -sustainability) each had their own identity hue (cyan, orange, slate, emerald), a tinted hero gradient with a glow orb, a gradient hub centre with a glow, and three boxed badges in the hero. They now share one amber accent on a graphite hero. The hero badges are one mono line, and the "by the numbers" tiles are fact rows (value on the left, title and sentence on the right, hairline between rows) instead of four boxed tiles with icon circles. The hero is dark in both themes, so its accent is the warm amber in both; the ink amber read 2.7:1 on that hero in light mode (the warm amber reads 10.2:1) and was caught in a render, not by a gate.
The three Indonesian pages (id/index.html, id/artikel.html, id/glosarium.html) lose the navy-to-cyan hero, the glow orbs, the gradient-text title, the rainbow article thumbnails, the coloured feature-icon washes and the card lifts. The four big cyan numerals in the id/index hero are one mono line, as on the pillars.
The glossary's accent was blue (#1d4ed8 / #60a5fa); it is amber now, in both languages.
The embeddable widgets (embed/pue-widget, capex-widget, carbon-widget, and the embed landing page) move from navy and cyan to graphite and amber. Each widget's four results read as label/value rows. The landing page's guideline sentence described the palette as "#0f172a background, #06b6d4 cyan accents"; it now names the colours the widgets actually use. The PUE rating ramp and the CAPEX category colours are data and stay.
The account dashboard: three boxed counters (saved projects, exports, calculator runs) are one line; the blue/slate accent and its gradient fills are amber, with text colours chosen per theme so the primary button stays readable.
css/rz-finance-suite.css (account, changelog, admin console) — surfaces graphite, identity accent amber; the state colours keep their meaning. The LTC session released this file to this wave.
The admin console's own navy literals and its KPI hover lift; toasts' success colour from neon #00FF88 to the muted status green (js/rz-toast.js, API unchanged).
Fixed
Hub nodes on the pillars jumped on hover..hub-node:hover { transform: scale(1.08) } replaced the translate(-50%) that centres four of the nodes, so those moved sideways when pointed at. The hover now changes only the border colour.
How it was done
All of it goes through tools/sweep-site-register.py, which gains six families (embed, pillar, lang, admin, suite, dash) and support for regex pairs whose body varies per page. Each family failed --check on the old tree before --apply (104, 90, 127, 45, 13 and 10 old values), passes after, and a second --apply rewrites nothing; every regex pattern excludes its own output so the check can go green. Rendered headless at 1440 px in both themes and at 390 px: no page errors, no horizontal scroll, and the fact rows measured as rows (one x position, one y per item) on all four pillars and all three widgets.
v3.24.30PATCH
— "Cockpits on graphite"
Owner: "polish more", then "continue" — the second wave of WI-WEB-ALLPAGES-001. The nine cockpits were the last family still on the navy register.
Measured with node tools/audit-vibecode.mjs --json: site-wide monitored findings 155 → 143, and none of the remaining 143 are on a cockpit.
Changed
The nine cockpits (datahall, datahallAI, dc-conventional, chiller-plant, water-system, fire-system, fuel-system, ict, EPMS_Telemetry) move from navy to graphite. 203 navy values were rewritten to graphite at the same lightness, so contrast did not move: 140 page backgrounds, the modal surfaces built inline in script (#0f1729 / #2b3a52 → #121316 / #2a2d33), the inline status and mode strips (datahall, ict, EPMS), the fire-system zone chips, the HMI drawing canvases in the datahallAI modals, and the datahallAI PDF figure background. Measured in a headless render of each cockpit (dark, 1440 px), counting visible elements whose computed background is navy-cast: before → after: dc-conventional 37 → 0, chiller-plant 32 → 0, fuel-system 53 → 0, datahallAI 33 → 0, ict 24 → 0, water-system 19 → 0, fire-system 16 → 0, EPMS 7 → 1, datahall 70 → 40. The 40 left on datahall are the rack units' standby LEDs (a state colour, kept); the one on EPMS is a 1 px rule drawn by js/rz-explain.js, which the LTC session owns. Shared: the instrument register's fallbacks for the sign-in button and user menu (css/rz-cockpit-instrument.css) and the slate hairline tokens on dc-conventional and ict. Functional state colours (alarm, trip, run, standby) were not touched. These are CSS-only changes: no engine, registry or computed number moved.
The cookie "Accept" button was saturated blue on seven cockpits. It now uses the site's amber with dark text.
Buttons on chiller-plant and the root actions on datahallAI were mono uppercase. They now use IBM Plex Sans at normal case, the same as the rest of the site.
Removed
The KPI cards on dc-conventional no longer rise on hover. Hover now changes only the border colour.
v3.24.29PATCH
— "Graphite"
Owner: "keep improve dan lakukan pembaharuan dan lanjutkan utk semua page, pastikan tidak ada page yang terlewat, lakukan pembaharuan untuk semuanya" — then "polish more". Tracked as harness work item WI-WEB-ALLPAGES-001 (plan by Fable, execution by Opus and Sonnet). This is the first wave; the hub and the journal were already on the register.
Measured with node tools/audit-vibecode.mjs --json before and after: 407 monitored findings in 237 files → 155 in 73 files.
Changed
Shared stylesheets and scripts move to graphite + one amber.styles.css, styles-index.css (shared chrome only — navbar, drawer, dropdown, footer, cookie banner, inputs; the homepage hero and bento wait for an owner-approved mockup), css/rz-dark.css, css/rz-bms-shell.css (state colours kept), css/rz-conv-alarm-workspace.css (alarm states kept), css/rz-documentation-ui.css, css/rz-monitor-instrument.css, js/rz-mobile-nav.js (the drawer was navy in both themes; now graphite in dark, white in light), js/rz-article-chart.js, js/rz-basis-drawer.js, js/network-anim/palette.js (token names kept, neon toned down, canvas graphite), rz-chat.js, the offline page in sw.js. Navy backgrounds were rewritten to graphite at the same lightness, so contrast did not move; aurora orbs deleted; gradient text made solid; card hover lifts and glows replaced by a border-colour change. The light-theme "Demo Account" label in the sign-in modal was mint on cream; it is ink-amber now (auth.js).
The 43 incident dossiers and the incident library page drop the forest-green palette and the Fraunces serif for the journal's graphite + amber and IBM Plex Sans — changed once in tools/build-incidents.py, rebuilt, --check CURRENT.
Technical manuals (39), network explainers (25), PRDs (6), and 42 content pages (compare-, pillar- except fire safety, pln-, infographic-, cdu-*, glossary, tools, 404, privacy, terms, dashboard, achievements and others) move onto the register through one new reproducible tool, tools/sweep-site-register.py (exact-string rewrites per family, --check / --apply, an excluded set with a written reason for each file). --check reported 747 + 148 old values before --apply and passes after; it is wired into tools/test-generator-freshness.py so a regression fails the gate. Cyan chrome accents became amber, neon status green became a muted #6aa588 that keeps its meaning, Fraunces left the font stacks and the Google Fonts URLs, 65 card hover lifts became border changes, and the keyword marquee CSS on glossary and tools was deleted.
Fixed
The fixed navbar covered the breadcrumb on 36 technical manuals (.mn-wrap started 2rem from the top under a ~70 px fixed header). They now clear it the way the 7 manuals that already did (5.5rem desktop, 5rem phone).
The network explainers' header ran its two links together ("Network HubCompare") and stayed transparent while the page scrolled under it — these pages do not load the shared script that adds .scrolled. The links have a gap and the header a solid background in both themes.
Not in this wave (recorded, not forgotten)
The 13 calculators and pillar-fire-safety.html are being refit by the parallel LTC session; the LTC lab pages and manuals are theirs too.
The homepage (index.html hero/bento) waits for a mockup the owner approves; mockup and plan artefacts are kept as evidence for the owner to decide.
The nine cockpits (CSS only, numbers untouched) are the next wave; stat strips on four pillar pages and the dashboard need a design decision.
tools/test-network-anim-determinism.py fails on origin/main too (Cannot read properties of undefined (reading 'width') in its Node canvas shim) and is not wired into the ship gate — a pre-existing defect, not caused here.
v3.24.28PATCH
— "Two cockpits"
Owner, on the DC Solutions hub: "area ini jadi terlalu rame, cukup DC AI dan DC conventional hyperscale aja. yang lain2nya itu kan ada saat setelah akses masing2 dari 2 itu. biar nggak terlihat ramai dan membingkan jadinya saat navigate ini"
Changed
The hub's Cockpits list shows the two cockpits only.datacenter-solutions.html listed nine rows: DC AI (K.01), DC Conventional (K.02) and seven systems that live inside the Conventional cockpit — EPMS telemetry, data hall, chiller plant, fire, fuel, water, ICT (K.03–K.09). Those seven are removed from the hub; each is still linked once from dc-conventional.html (checked: 7 of 7 links present), so no page lost its entry point. Rendered at 390 px dark and 1440 px light: two rows under the ATS figure.
The Conventional row keeps its name. The cockpit does not call itself "hyperscale" anywhere, so the label was not changed to a claim the page does not make.
v3.24.27PATCH
— "Share"
Owner:"continue, improve more, polish more". Fourteen articles carried a share bar nobody could see, and two
shared the wrong article.
Fixed
Share bar invisible on 14 articles.styles.css holds .share-buttons at opacity 0 until a script adds
.visible once the hero scrolls away. Articles 13, 15 and 18–29 carried the bar with no such script, so it stayed
hidden for good (measured in a browser at 390 and 1440 px after scrolling half the page: opacity 0 on all 14).
Each now gets the same hero observer the other articles use; 13 and 15 also gained the shareButtons id. The
browser re-measure shows all 15 bars (with article 30's new one) visible at both widths.
Article 29 and article 30 shared article 28. Their share script hard-coded article 28's URL and title, so a
reader sharing either page sent a link to a different article. Both now read the page's own canonical URL and
og:title.
Article 30 had no share bar at all, only its functions. It now has the same bar as article 29, so the
table-of-contents button on a phone sits in a bar instead of over the text (the last of the three pages left
from v3.24.26).
Gate
tools/test-share-bar.mjs (wired into the ship gate): on every page that loads styles.css, a share bar must be
visible by markup, removed on purpose, or toggled by a script that references that bar; and a literal shareUrl
must equal the page's canonical. On the previous tree it reports 16 defects (the 14 hidden bars, article 29's and
article 30's foreign URL; article 30's bar did not exist yet); now it passes on 36 bars. Its first version accepted
any .classList.add('visible') in the page and so passed article 24 on an unrelated calculator panel; the check
now requires the toggle to sit in a script that names the bar.
v3.24.26PATCH
— "Real Phone"
Owner:"continue, improve more, polish more". The responsive gate emulated a narrow desktop window, so
five pages that load zoomed out on a real phone passed it. The table-of-contents button covered body text.
Fixed
Five pages loaded wider than the phone. On a phone the layout viewport grows to fit overflowing content,
even where body{overflow-x:hidden} keeps scrollX at 0. Measured at 360 px: article-1 601 px wide (the
menu button off-screen), tier-advisor and tia-942 382, capex and opex 378. Causes:
article-1: 56 legacy .tooltip::after tooltips, opacity 0 but white-space:nowrap, still laid out past
the right edge. They now take no space until hovered and wrap at 260 px.
capex, opex, tier-advisor: the footer is an inline 2fr 1fr 1fr grid with no class, so no .footer-grid
phone rule reached it. A scoped rule collapses it to one column under 768 px.
tier-advisor, tia-942: the phone chart grid was 1fr (min auto). Now minmax(0,1fr).
capex: the simulator and input grids were 1fr, and a long <select> option set the track width. Now
minmax(0,1fr); the refrigerant <select> sized itself to its longest option (327 px in a 278 px
column) and now takes the column width.
Nested fact lists on articles 1, 2, 3 and 5.tools/sweep-journal-register.py matched
evidence-card inside evidence-card-value, so each card's own value, label and sub line made the card a
"parent of tiles" and a fact list inside a fact list. The tile word must now end the class token, and a tile
is never itself marked. The old sweep re-adds 20 attributes on 4 pages; the new one 0.
TOC button over the text on phones. It sits at the left end of the bottom share bar (whose icons centre
in the space to its right) instead of 90 px above it.
Gate
tools/audit-responsive-layout.mjs now emulates a phone (isMobile, touch) at 360 and 390 px and also fails
a layout viewport wider than the screen. On the v3.24.25 tree it fails 5 pages (article-1 +241, tia-942 and
tier-advisor +22, capex and opex +18) where the previous version reported clean; on this tree it passes.
v3.24.25PATCH
— "Narrow Column"
Owner:"continue, improve more, polish more" — the two article-30 figures v3.24.23 left scrolling at 360 px,
and a clipped-label case its check could not see.
Fixed
Every article-30 figure now fits a 360 px phone. The phone composition may be at most 355 units wide
(336 px column ÷ the 0.945 legibility floor; it was 368, sized for 375 px). D2 "Observe, act, verify" 356 → 352
(left margin and right padding retuned); D5 "Drafter and checker" 364 → 268: the right-lane label "error text
back" pushed it wide, so on the phone that dashed line carries its meaning in the legend ("Error text back to
the drafter") and the dashed validate-to-test line moved to the left lane. Measured in a browser at 360, 375 and 390 px: every
figure's scroll width equals its box (D2 and D5 scrolled 1 and 8 px before), page scrollX 0, no page errors.
Arrow-label panels clipped at the left edge. The clipped check read the occupancy map, which holds an arrow
label's text but not the panel behind it, so D4's "fix" label panel started at x = −1.8 and passed. The builder
now also fails any drawn rectangle left of or above the origin (RED on D4 before its fix, the only case across
all articles); D4's phone column moves right 6 units.
Checked, no change
The mobile Table of Contents sheet is closed on load (its top sits at the viewport bottom at 390 × 800); it only
appeared in element screenshots because a tall element capture enlarges the viewport.
v3.24.24PATCH
— "Hub Tools"
Owner:"lakukan update dan penyempurnaan semua tools yang ada di hub ini. literally semuanya" — all 33 tools
linked from datacenter-solutions.html were loaded in a headless browser at 390 px (dark) and 1440 px (light),
every visible input was changed, and their standards claims were checked against primary sources.
Fixed — tools that rendered nothing or hid their data
The TIA-942 checklist rendered no items since July. Chart.js loads with defer, so a top-level
Chart.register threw "Chart is not defined" while the inline script parsed, and the DOMContentLoaded init
registered below it never ran: 0 items, score stuck at 0 %, no charts. The plugins now register on the first
draw. Live before: 0 items and the page error; after: 62 items at the default filter, ticking 5 moves the score
0 % → 9 %, 2 charts drawn.
The PLN Java grid map showed 2 of its 181 power plants. The tier-2 rule of the "Show all 150 kV" substation
button also applied to plants, and 179 plants carry tier 2, so the Coal / Hydro / Solar / Geo / Biomass / Wind
toggles had nothing to hide. On the map a plant now follows its fuel toggle only (the schematic keeps its rule):
181 drawn, Coal off leaves 151. The provenance cards no longer run 9 px past a 390 px screen.
The fire-protection checklist lost every tick on reload, and an item's text did not tick its box (the
25 checkboxes had no label). Each item is a label now, each list shows "n of N checked", ticks and the
service-record fields are kept in the browser, and Clear empties both.
Tier Advisor: "Number of floors" and "Building age" were read nowhere. Tier depends on topology, not on either,
so they now appear in the report's facility parameters instead of being dropped silently.
Changed — data brought up to date, one source per number
Market tracker: H1 2026. 74 values in the shared engine (DATA.markets): operational and under-construction
MW and vacancy for all 25 markets from CBRE North America Data Center Trends H1 2026 and Cushman & Wakefield
Global Data Center Market Comparison H1 2026 (Nairobi and Johannesburg vacancy: Knight Frank). Every value carries
a verbatim quote from its source; five were re-read live before use. Largest moves: Dublin 400 → 1,253 MW
(C&W counts hyperscale self-build), Northern Virginia under construction 1,100 → 2,420 MW, Dubai vacancy
3.0 → 26.2 %. Colo price stays on the 2025 basis — CBRE publishes wholesale 250–500 kW asking rent, a different
measure. The page said "Data: Q1 2025 estimates · Updated quarterly" and "Real-time"; the data date is now read
from DATA.sources.markets and the hero carries 2026 figures (7.5 GW under construction in North America, CBRE;
+57 % data center capex in 2025, Dell'Oro; 1,360 hyperscale sites, Synergy). Four 2026 references added. The
manual's 25-row table is regenerated from the engine; it had already drifted on 9 markets.
Industry-average PUE: 1.52 (Uptime Institute Global Data Center Survey 2026, read in the report). The PUE and
carbon calculators hard-coded Uptime 2024's 1.58 in 14 places while the engine used 1.56; both now read
DATA.pueBenchmark, and the ROI preset and four manuals follow. The carbon manual's worked example is recomputed
(5,878 → 5,668 tCO₂e/yr, 0.671 → 0.647 kgCO₂e/kWh, grade D unchanged) and the page prints 5,668 with defaults.
TIA-942 checklist → revision C (2024). Three requirements from TIA's own TIA-942-C white paper: cabinets
≥ 800 mm in MDA/HDA/IDA, Category 6A to high-throughput access points, ASHRAE class H1. The immersion item cited
a clause the old revision does not have; it now points to the -C annex, and the page says its other clause
numbers follow -B and are indicative. BICSI 002-2019 → 002-2024. The manual said 79 items (the page had 85) and
a worked example of "48–55 %, grade D"; the page's own formula gives 43.2 %, grade F — all its figures are now
computed from the item array.
Fixed — standards claims checked against primary sources
DC AI: IEEE 519 "caps TDD at 5 % for I_sc/I_L 100–1,000" — Table 2 gives 15 % for that row (5 % is below 20).
DC Conventional: NFPA 2001 halocarbon discharge "30 s" → 10 s; "ASHRAE recommended 40–60 % RH" → dew point
−9 to 15 °C and ≤ 60 % RH (also in the TIA checklist, whose allowable range was wrong too).
Tier Advisor, OPEX, CAPEX, CX, N+1 vs 2N: the 99.671–99.995 % availability figures were presented as the Uptime
Tier Standard; Uptime removed them in 2009. They are labelled as the legacy estimate, OPEX's Tier IV 99.99 % is
99.995 %, CX's N+1 99.982 % (a Tier III figure) is 99.741–99.982 %, and "Tier IV+" (which does not exist) is gone.
Fire checklist: HFC-227ea GWP 3,220 was labelled IPCC AR6; it is the AR4 value the site uses throughout.
Carbon calculator: its grid-factor source line now names the engine reference (IEA 2026 + Ember), not IEA 2023/2024.
Gates (each proven red on the pre-fix files)
tools/test-hub-tools.mjs (browser): TIA renders and scores, every PLN plant on the map and the fuel toggle
removes its plants, fire-checklist ticks survive reload, tracker date = engine, PUE pages read the engine.
Red on v3.24.22: 0 items + "Chart is not defined", 2 of 181 plants, no labels.
tools/test-standard-claims.mjs: 7 rules over 246 pages — red with 8 findings before.
tools/test-tia942-manual.mjs: every manual count and the worked score from the item array; no top-level
Chart.* call — red 10/10 before.
tools/test-rz-engine.mjs (now in ship-gate; it was not): market totals recomputed, vacancy band 0–30 % (Dubai
26.2 % is a sourced value), tracker and manual read the engine, PUE literals — red 13 before.
v3.24.23PATCH
— "Plain Record"
Changed
Article 30 rewritten in plain engineering prose. Owner, on the live page: *"style tulisannya
dan arrangement dll itu terlalu ai design slop", "'The claim, in his words#' ini kan aneh to the
point aja stright forward. dan bener2 accrrata dan comprehensive"*.
Every heading now says what the section contains ("The 14 claims tested", "How a language model
produces text and calls tools", "Can a language model reason or do arithmetic?", "Investment,
revenue and the IPO claim", "Checklist: connecting an agent to a BMS"); framing, rhetoric and
remarks about the article itself are cut.
Facts kept: every citation, number, quotation and source id was compared mechanically between
the old and new fragments before the rewrite was accepted.
The source notes for [S8] and [S23] were shortened (they render as the "Taken:" line in the
source list), and the leaked-financials row names its route ("Reuters, via CNBC; prospectus not
public").
Title: "Is 'Agentic AI' a Fraud? 14 Claims Tested Against the Evidence"; the OG card was
regenerated with it.
More charts and tables, less list text. Four sourced charts added (run success by step count,
cost of one run by step count, 2026 capex guidance ranges, grades for E1–E14), all carrying
source + basisTag; three lists became tables (3.4 the case against, 4.3 old versus new, 7.4).
Charts now draw their y-axis title (yLabel was never read), use whole-number ticks when every
value is an integer, and space the x-axis evenly.
Fixed
Figures cut off. Owner, with a screenshot of the restaurant-demo diagram: *"terpotong dan aneh
sekali jelek ini chartnya"*.
Desktop: all eight figures were recomposed to fit the 736 px column; the builder now FAILS any
article-30 figure wider than the column (FIT_TO_COLUMN_PAGES). Measured at 1440: 8 of 8 fit
the 744 px container, 0 scroll.
Phone: a figure that fits the desktop column still scrolled sideways in a 351–366 px phone
column (508–732 px drawings). Each figure now carries a phone composition of the same nodes,
edges and zones in one column, drawn at 1:1 and shown below 800 px. It may shrink only to the
8.5 px label floor (scale 0.945). The builder fails a phone drawing wider than 368 units, any
overlap, and any label placed outside the drawing (a new check: fit() grows the canvas right
and down only, so a label left of the origin was silently clipped — it caught five on the first
run). Measured at 375 px: 8 of 8 figures fit, page scroll-x 0, 0 page errors. At 360 px two
still scroll inside their own track, by 1 px (D2) and 8 px (D5).
The safety-system box in the chiller-plant figure renders in the site's alert red
(faultRed), so it no longer reads as the same thing as a probabilistic box.
standarization/AGENT_HARNESS_STANDARD.md "Last updated" restamped to the release date.
Not done
Lucide icons: none are used on this page; standard IEC/ISA symbols appear where the figure draws
equipment (BMS, chiller). Replacing icons elsewhere was deferred by the owner ("skip dulu ini").
17 figures on other journal pages are still wider than the column (listed by the builder); they
scroll inside their own track.
v3.24.22PATCH
— "Flow Layer"
Changed
The hero's flow dashes paint on their own layer. The eight animated dashes were drawn inside
the plant figure, so every frame repainted the whole drawing (46 boxes, 46 labels, 12 symbols).
They now sit on a thin SVG with the same viewBox, laid over the figure on its own compositor
layer. Traced A/B on the hero at rest (headless, 4 s, two runs each): paint time 339 / 338 ms
became 195 / 203 ms, and raster was unchanged. The dashes land on their wires to 0 px at 1440, 900
and 390 px. They draw in with the figure, and hover still lights them.
Tried and dropped: stepping the dash animation to 30 fps (steps(42)) did not reduce Chrome's
per-frame style passes (599 vs 601 per 10 s), so it was not shipped.
standarization/AGENT_HARNESS_STANDARD.md "Last updated" restamped to the release date.
v3.24.21PATCH
— "Callouts"
Fixed
The 3D figure's callouts overlapped and ran off the stage. A headless check now places every
visible tag box in all six states (rack, compute tray, switch tray, each normal and exploded) at
1440 and 390 px. Before the fix: a numbered tag on the "40 °C in" readout (compute tray, desktop),
tag 6 on "50 °C out" (switch tray), and on phones the compute tray ran out of the frame with
readouts off the edge. After: 0 overlaps and 0 tags off the stage in all twelve cases, and every
view still shows its tags.
Tags are tested as their real boxes (4 px clear), not a fixed 26 px square. The temperature
readouts and the part in focus are placed first. A readout that would run off the right edge,
or onto another readout, moves to the other side of its point.
The tray views back the camera off on a narrow stage (same line of sight, distance scaled by
the stage aspect), so the whole tray fits on a phone.
New tags start hidden, so they no longer flash at the stage's top-left corner before the first
frame places them.
Registry ids under the spec grid break at "." and "_" instead of mid-word.
v3.24.20PATCH
— "Light Hub"
Fixed
The DC Solutions hub was heavy enough to hang the browser (owner report). Measured A/B
against v3.24.19 in the same headless browser, 10-second idle windows:
3D figure: it redrew every frame for as long as it was on screen (auto-rotate, coolant
flow and fans kept the loop alive, at half rate after 10 s but never stopping). It now renders
on demand. It turns and runs its coolant for 9 s after it first shows (or after Reset), and
while someone drags, hovers or a tween runs. At rest it holds its last frame and requests no
frames. The hover raycast runs at most once a frame instead of on every pointer event.
Off-screen animations: the hero plant and the seven catalogue panels kept their
infinite dash animations running off screen, repainting their SVGs every frame. Panels and the
hero now pause their CSS animations while off screen.
Unused and duplicate scripts: the 509 KB rz-engine.min.js was loaded and parsed on a
page that never calls it, and rz-mobile-nav.js was loaded twice. Both tags are gone.
Result: scrolled to the end of the page, v3.24.19 ran 109 animation frames and 56 s of
script time across threads per 10 s window (WebGL is software-rendered in the harness).
v3.24.20 runs 0 frames and 1 ms. On the hero, busy time fell from 448 % to 268 % of the window.
The registry ids under the rack spec grid were cut off with an ellipsis
(distribution.rack_feed_…). They wrap now, so every id reads in full.
v3.24.19PATCH
— "Whole Graph"
Fixed — Second Brain said "healthy" while 26 of 36 journal pages had no node
Owner:"Continue. Improve more dan pastikan proof", "Setiap pekerjaan harus ada proof".
Apps/second brain/sync-graph.py checked the graph only against itself (dead URLs, orphans, vault notes), so it
reported "Graph is healthy" with the FF series, the geopolitics series and most articles missing. Check 5 now
compares the graph with data/journal.json (generated from the pages on disk): every journal page needs a node.
--strict exits 1 on any issue so a gate can run it. Proven red on the v3.24.18 graph (exit 1, 26 pages listed)
and green on this one (exit 0).
--sync-journal GENERATES the missing nodes and edges from data/journal.json between journal:auto markers
(curated nodes untouched; each page links the journal hub, its series and the next article). Idempotent: a second
run writes the same block. The graph renders 175 nodes (was 149) and 377 edges, no dangling edge, no duplicate id,
rack + tray rendered dark, then switched to light in place, pageerror 0.
v3.24.15PATCH
— "Calculators"
Fixed — article 30's four calculators were empty; tables, formulas, numbers and edges across the journal
Owner:"sempurnakan, continue", then "Continue polish more" and *"Yg terbaik. Harus terbaik walau
ultraeffort"*.
article-30.html shipped without js/rz-calc-utils.js and js/rz-agent-calc.js: the four calculators (agent or
script, run cost and reliability, loop budget, verification ladder) rendered as empty 16 px boxes, with no error.
Cause: tools/assemble-article.py removed the previous article's calculator scripts from the end of the body and
inserted the new ones into <head> before an anchor the head does not carry — a silent no-op. The scripts now go
back where the shell keeps them, and a missing anchor fails the build.
Tables in the reading column had no cell padding on desktop: .ws-table (articles 29, 30) and .ws-lever-table
(28) were never defined and the only padding rule sat in a phone media query, so "E13" and "11:14" read as
"E1311:14". css/rz-article-dark.css now carries a floor at element specificity, so a page's own table styling
still wins.
Justified paragraphs stretched the spaces inside inline code ("p = 0.95" set as "p = 0.95"; 123 code spans
with spaces across the journal). On desktop an inline code span is now one unit to the justifier.
Calculator results in articles 29 and 30 read as a reader expects: $1.84 instead of 1.843 (USD),
9.11 × 10⁻⁵ instead of 9.113e-5, 0.088 instead of 0.08800; labels with two bracketed asides in a row
("Step fails silently Pu (dangerous) (probability)") rewritten. The raw value still goes to the CSV and the
worked-step line. Both calculator tests now assert the format (578 and 396 assertions).
The series bar under every article hero ("Engineering Journal — Article N of 30") still carried the blue gradient
wash, blue border, 12 px radius and blue number chip from styles.css; it is on the graphite register now.
One left edge: the key-number strip under the hero and the series bar started at 23 px and 270 px at 1440 against
176 px for the title and the text — their wrappers carried their own 900–1100 px caps. Both now start where the
title starts, on desktop and on a phone. On a phone the number column narrows from 9.5 to 7 rem, so a label no
longer breaks into two lines.
The chart basis chip under every sourced chart was mint in light ("PUBLISHED") and two different yellows in dark;
published/standard is neutral, modelled/vendor amber, outlined, the same in both themes.
Charts on the editorial register (js/rz-article-chart.js): every series without an explicit accent took the same
amber, so article 30's "50% horizon" and "80% horizon" bars were indistinguishable; the instrument palette also
carried neon #00FF88, cyan #00DDFF and a navy tooltip into articles. On the register the lead series is amber,
the next ones neutral greys, red only for a fault, tooltip on graphite. Calculator and instrument pages keep theirs.
Gates — tools/test-journal.mjs J5f now also fails on: a data-*-calc mount under 60 px tall, a one-line code
span wider than its own text, a table where most cells have no horizontal padding, the series bar's colours, and a
hero strip or series bar whose left edge is off the title's, a chart basis chip in a second hue, and chart series that share a colour or leave amber/neutral/red. Each proven red on the previous files (three empty
calculators per theme; three stretched formulas; tight tables on six pages; 72 off-edge series bars, 18 off-edge
strips) and green after: 137 assertions on 40 pages.
v3.24.14PATCH
— "Hub"
Added — the 3D figure answers the keyboard (js/rz-hub-3d.js, datacenter-solutions.html)
The canvas took focus but ignored every key, so the figure could only be turned with a mouse or a
finger. On the focused canvas: arrows orbit (left/right around, up/down over), + / − zoom
within the view's limits, E toggles the exploded view, C the cutaway, 1–4 the Iso / Front /
Rear / Top presets, 0 resets. The canvas label names the keys for a screen reader, and the hint
line lists them where there is a keyboard. Proven in a browser: two ← moved the camera from
(3.10, 2.30, 3.90) to (2.08, 2.30, 4.53), + closed in, E set the explode button pressed,
3 went to the rear.
Full ship gate on this tree (the hub's releases since v3.24.9 had run only the static set).
v3.24.13PATCH
— "Captions"
Fixed — text glued to a frame on 23 journal pages
Owner:"improve more" after article 30 went live; the figure captions under every diagram read as a boxed
paragraph with the text touching its border.
The editorial callout rule ([class*="-box"], [class*="-note"] in css/rz-article-dark.css) also matched things
that are not callouts: a figure caption's argument (.rz-figcaption-note), one-line source and result notes
(.src-note, .llm-calc-notes, .ws-note, .pfas-chem-note, .iso-mapping-note) and the title inside a
callout (.info-box-title, .formula-box-title, .a22-chart-box-title …). Each got its own panel and border
while another rule zeroed its padding. They are text again: no frame, no panel.
tools/test-journal.mjs J5f now flags any element whose own text sits flush against its own visible border, figure
captions included (they were skipped with the figure). Proven: 46 failures on 23 pages before the fix, 0 after.
v3.24.12PATCH
— "Hub"
Fixed — two things the hub did that nobody could see (js/rz-hub-illus.js, js/rz-hub-3d.js)
The seven group panels shared one requestAnimationFrame loop that never stopped: it ran every
frame for the life of the tab, checking visibility, even with every panel scrolled away or the tab
hidden. It now runs only while a panel is on screen and the tab is visible, and wakes from the
IntersectionObserver and visibilitychange. Measured: a panel's cursor holds still off screen and
advances on screen.
The 3D figure's numbered tags piled on each other on a phone (tags 6 and 7 on the rack). A tag
that would land within 26 px of one already placed is held back unless it is the part in focus;
temperature tags are never held back. At 390 px the rack shows four clean tags of seven; the
rest appear as the view turns or when picked from the list.
v3.24.11PATCH
— "Hub"
Fixed — the plant tooltip could vanish on a desktop, and spilled when it showed (datacenter-solutions.html, js/rz-hub-product.js)
The hover card was hidden by @media (hover: none). Touch-screen laptops and some browsers
report hover: none while a mouse is in use, so the card never appeared there (the headless
check reported it too, which is how this surfaced). The card is now suppressed per event, only
when the pointer that triggered it is a finger (pointerType === 'touch').
When it did show, long values ran past its edge ("4 MW class MV diesel generator", every value
column). The table is fixed-layout with a 48 % label column and values that wrap; the card is
270 px.
Campus: the "GENSET YARD" label sat on the dry-cooler yard's top border; it now centres in the
road between them.
The bin strip and the status row print temperatures the same way ("24 °C"); phones still drop
the unit.
v3.24.10PATCH
— "Agents"
Article 30: Is "Agentic AI" a Fraud? — and the last second hues leave every article
Owner: the brief to research John Elder's "Agentic AI is fraud" video and write article 30 from basic to
advanced with calculators and diagrams; then "improve more".
Added
article-30.html (≈12.7k words, Future Forward · AI, tested against the evidence): eleven sections that steelman
the video, then grade its fourteen claims and the six counter-claims against 63 sources — the mechanism (right, and
exactly what vendors document), the absolutes (the model computes, unreliably, and cannot report how), what an agent
is mechanically, the measured record (METR horizon 50 % vs 80 %, 5.6×; SWE-bench's own audit), failure modes, the
money (big-four 2026 capex ≈ $0.7T, not the video's unsourced $3T), the name, the engineering stance, a ten-point
checklist for "put an agent on the BMS", and a verdict table. A disclosure near the top: researched and drafted with
Claude, and several key sources are Anthropic's own. Eight engine-drawn figures (agentic-d1…d8, zero warnings);
four calculators in js/rz-agent-calc.js (agent or script, run cost and reliability with compounding error, loop
budget, verification-ladder coverage) with tools/test-agent-calc.mjs (557 assertions); 32 new glossary terms and
backlinks on 8 existing ones; 26 RZExplain entries; five social drafts with UTM links.
tools/assemble-article.py: the article assembler, promoted from the article 29 session script and driven by
Article/Article_<N>/build/article.json (shell from the previous article, fragments spliced, sources from the
dossier table — local-file sources listed as such, cited-but-undefined sources fail the build — JSON-LD, FAQ, key
numbers, search-index entry, token restamp).
Fixed
A compounding figure in the prose: 0.995^20 = 0.899 is wrong (it is 0.905); the step probability is
0.95 × 1.04706 = 0.9947 and 0.9947^20 = 0.899 — caught by the figure agent recomputing the number for D6.
The shared cookie banner (navy, blur, shadow, blue Accept), the login button and the signed-in pill (mint, navy
dropdown) on every editorial page now sit on the register, including the banner's "Learn more" link that
per-page legacy rules (#cookieBanner p a { color:#93c5fd !important }) kept blue or cyan.
tools/test-journal.mjs J5f could not see that chrome: J4 declines cookies and the browser's localStorage is
shared across its pages, so the banner never rendered in J5f; and the mint login button's background alpha
(0.08) sat under the 0.3 threshold while its hue lived in text and border. J5f now clears consent per page and
checks text and border colour of banner, login and user pill. Proven: 72 failures on the previous stylesheet,
0 on this one. Its amber band now starts at 20° so the light-theme accent #92400e (23°) is not flagged.
Changed
standarization/ARTICLE_PRODUCTION_STANDARD.md: a new page's OG card needs build-og-images.py --apply --discover.
Journal grids (36 pages), series navigation "of 30", hub/landing OG cards regenerated.
v3.24.9PATCH
— "Register"
Changed — the shared login, account and cookie widgets leave the withdrawn palette on every page (auth.js, auth.min.js, js/rz-cookie-consent.js)
v3.24.8 fixed these three widgets on the hub only, by override. The standards now list the mint
user pill and the navy surfaces as withdrawn site-wide (CLAUDE.md "Withdrawn", §3, §11), so the
fix moves to the source. auth.js: every mint value becomes the signal amber (#FFAA00,
#FFC44D on hover, #7A4800 as light-mode text), the navy dropdown and sign-in modal take the
graphite surfaces (#121316, hairline #2a2d33, inputs #0b0c0e), and the modal's 2 px top
rail is removed. auth.min.js rebuilt with terser (the homepage loads the twin). The cookie
banner loses its blur and shadow, takes opaque graphite in dark and white in light, 4 px buttons,
an amber Accept with dark text, and neutral links instead of blue. The educator badge keeps its
documented cyan (AUTH_STANDARD).
?v= tokens re-stamped on the 195 pages that load auth.js, the 2 that load auth.min.js, and
every page that loads the cookie banner, and the 44 generated incident pages rebuilt so they
carry the same tokens. Full ship gate 128/128 after two fixes on this tree (the version token
re-stamped after the bump; the incident pages regenerated rather than hand-patched); probe 82/82.
v3.24.8PATCH
— "Hub"
Fixed — the last template widgets on the hub (datacenter-solutions.html)
A whole-page computed scan of the hub (every rendered element, both themes: navy surfaces,
gradients, shadows, radius >= 8 px, backdrop blur, neon-green text) found the hub's own content
clean and three shared widgets that were not: the cookie banner (navy in dark mode, 12 px blur,
shadow, 8 px buttons, a blue Accept), the user dropdown (navy) and the mint login pill. Under the
hub register (body.pp) they now take the graphite surfaces, one amber action, hairline borders,
4 px radius, no blur and no shadow. Scoped to this page: the shared rz-cookie-consent.js and
auth.js are untouched, so no other page changes. Re-scan: zero findings in both themes; the one
remaining gradient is the 3D stage's neutral backdrop.
v3.24.7PATCH
— "Graphite"
Every article on the register: no navy, no rails, no glow — measured in the browser
Owner:"Aku bilang no ai design slop. Itu banyak sekali ai design slop." He was right, and the gates had
passed it. Article 29 in a browser: the reading column sat on styles.css's navy [data-theme=dark] .article-content
(#0f172a), tables on #1e293b, related cards and the author bio on rgba(15,23,42,.8) with a 4 px light-blue rail, a
cyan floating contents button with a blue glow, and 2 px amber rails on every callout, figure note, calculator note
and the sources note. Across the 39 editorial pages a new whole-page scan found 53 failures (navy surfaces, rails,
cyan chrome) and later a cyan glow card on FF-3. The source sweep (v3.23.0) could not see them because they come from
styles.css descendant rules, page tokens with arbitrary names and bespoke widget classes, together.
cards, author bio, key-numbers and calculator blocks on graphite tokens; callouts are a panel with ONE hairline
(the 2 px accent rail and the 6 % amber wash are gone — UX law 14); the contents button is a neutral square.
js/rz-article-editorial.jsregisterSurfaces(): on every editorial page and every theme switch, the COMPUTED
colour decides — a navy-band surface (hue 200–250°, s ≥ .25, l ≤ .22) goes to the register's bg/panel token, a lone
left rail ≥ 2 px is removed, a saturated box-shadow glow is removed; legend dots, bar fills and swatches are data
colour and left alone. Marks are cleared and re-measured on each refresh.
tools/test-journal.mjs J5f: every rendered element of all 39 editorial pages in both themes — no navy surface,
no rail, no coloured glow, no saturated non-amber fixed chrome. RED on v3.24.6's stylesheet (6 of 6 sampled
page×theme pairs; 53 of 78 once widened to the family), GREEN now: 134/134. Two contrast regressions the change itself caused (white related-card numbers on the new
light panel; article-26 chem notes given a light panel inside a dark-in-both-themes block) were caught by axe and
fixed before release.
v3.24.6PATCH
— "Hub"
Fixed — the plant's bin strip clipped every label on a phone (datacenter-solutions.html, js/rz-hub-product.js)
At 390 and 360 px each bin button was narrower than "24 °C", so the strip read "24 °( 26 °(".
On phones the bins print as "24°" (the unit stays in the title and returns on desktop), the
selected bin no longer widens with its hours, and all seven fit at 360 px — checked per button.
On phones the status row drops its two constants (Total IT, GPUs, which do not change with the
bin) and keeps the four that do: two lines instead of five.
The gap between the hero copy and the plant tightens from 48 to 40 px.
v3.24.5PATCH
— "Ragged"
Lists stop justifying on a phone
Owner:"improve more". Article 29 at 390 px still justified 147 list items: the phone rule
.article-body li { text-align:left } lost on specificity to the desktop .article-body > :is(ul,ol) > li
justify rule, so lists with code tokens opened wide rivers between words.
Changed
css/rz-article-dark.css: the phone left-align rule carries the matching > :is(ul,ol) > li selector
(all 40 editorial pages).
tools/test-journal.mjs J5e also fails on any justified paragraph or list item at 390 px (147 on v3.24.4).
v3.24.4PATCH
— "Hub"
Fixed — the catalogue packs, the knowledge panel reads, the panels read on a phone (datacenter-solutions.html, js/rz-hub-illus.js)
The seven tool groups sat in a two-column grid, so a short group left a hole under it the height
of its neighbour (Labs beside Knowledge, Strategy beside Engineering). They now flow in two
balanced columns and never split.
The knowledge panel was eight titles in 9.5 px against the left edge of an empty box. It is a
shelf now: one spine per title across the panel; the one being read stands out of the row, turns
amber and its title prints large on the right with its index.
On a phone every panel scaled its 520-unit drawing into 340 px and its labels fell to about 6 px.
The panel type scales by 1.4 under 600 px; the three labels that then collided (the genset line on
the load icon, the PUE readout on the chart, the shelf header on the last spine) moved.
v3.24.3PATCH
— "Calculators"
Article calculators get the register; the headline's second clause is muted
Owner:"improve more". Reading article 29 on a phone: its five calculators rendered as raw inline
labels beside 21 px inputs and plain-text results — the site's calc-* classes were only ever styled inside
article-13's own <style>, so a new article inherited nothing. And the amber italic second clause of every
article headline is the emphasis §A27 withdraws ("a muted second clause is the only emphasis").
Changed
css/rz-article-dark.css CALCULATOR section: one register rule set for every article calculator — label
over input, 44 px mono controls, results as a label/value list, "show working" in muted mono, the error box,
two columns on a phone; fact-list value column fixed at 9.5 rem (7.5 rem on a phone) so labels align.
.article-title .highlight (both themes, all 40 editorial pages): muted, upright, weight 500 instead of
amber italic.
tools/test-journal.mjs J5e: on article-29 at 390 px every calculator has inputs ≥ 40 px, labels above
inputs, no horizontal overflow, and an upright headline clause — RED on v3.24.1's stylesheet
("input 21px", "label beside input", "overflows"), GREEN now (64/64).
v3.24.2PATCH
— "Hub"
Fixed — five things the live hub showed on a desktop and a phone (datacenter-solutions.html, js/rz-hub-product.js)
The bin strip clipped its last button ("36 °C 160 h · cliff" cut at the right edge on desktop).
Each bin now shows its temperature only; the selected bin widens to its content and adds its
hours and "design" / "cliff".
The readout repeated the "hover or tap" hint the line above it already gives; removed.
Phones: the plant drawing is wider than the screen with nothing saying so. A right-edge fade and a
swipe ⟶ chip appear on the drawing and leave once it is swiped (or reaches its end).
Phones and tablets: the j / k keyboard hint is hidden where there is no keyboard
(hover: none / pointer: coarse); the ⌘K chip stays.
Phones: the "Computed against" standards row is a two-column grid instead of a ragged wrap.
v3.24.1PATCH
— "Tokens"
Article 29: How a Large Language Model Actually Works — from one token to a 1 GW campus
Owner: *"saya mau buat article baru terkait AI, LLM, token, inference engine, transformer fungsi math besar,
agentic ai, harness, cara train, uncensored model, frontier, guessing engine semuanya dari awal basic sampai advance
termasuk step2 utk di level apa hardwarenya seperti apa … ada diagram pastikan standard dan super accurat jika ada
simulasi/calculator juga" — and, on the process: "Update standarization etc any doc standard when create article.
Including drafting post for linkedin, x, mastodon etc. Ensure all article have that."*
Added
article-29.html (15,955 words, 69 min, Future Forward · AI infrastructure): sixteen sections from the one-paragraph
answer through tokenizers, the transformer block, the KV cache, mixture of experts, sampling, model families, training,
Workforce & careers) and three series. The hand-written card copies (28 on the hub, 3 + 3 + 9 on the landings) are now
builder output between <!-- journal:grid --> markers; --check is in tools/test-generator-freshness.py.
js/rz-journal-filter.js (ES5, 170 lines) + the JOURNAL section of css/rz-article-dark.css: search (title, summary,
tags; / focuses it), topic chips with live counts (a chip whose count falls to zero is disabled, never hidden), series
chips, sort, clickable card tags, "N of 34 articles", one Clear, an empty state, and the whole state in the URL
(?topic=power&series=global&q=pfas) so a filtered view is a link and the back button undoes a chip. Chips are
<button aria-pressed>; nothing animates but border colour on the motion tokens.
tools/sweep-journal-register.py: the per-page inline CSS of the 39 editorial pages rewritten onto the register by HUE
BAND, never by hex list — 1,405 rewrites: 635 navy-cast backgrounds (incl. --*bg* tokens; a dark-theme rule takes the theme token, an
unscoped rule — an embedded panel dark in both themes — takes a graphite literal and declares light text so what it
contains stays readable), 104 gradient-text → one colour, 153 transition: all → the register's four properties,
61 hover lifts removed, 54 blurs / backdrop-filters removed, 70 serif faces → the sans stack (Fraunces also
dropped from the Google Fonts link), 35 per-page <html style="--rz-art-accent:…"> overrides removed (one accent, not 28),
5 navy literals in exported-SVG JS strings → graphite, 15 already-graphite containers given light text,
34 dark panels marked data-rz-dark-panel (the register keeps their headings light in light mode), 132 slate-grey
texts on dark surfaces lifted to a readable grey (axe), 107 stat-tile strips marked data-rz-fact-list and rendered by the register as one fact per row (mono number,
label, sub-text, hairline). Idempotent (--check), reproducible (pristine HEAD + sweep == the shipped tree), the full
before→after log in standarization/Audit result/journal-sweep.md.
tools/test-journal.mjs: J1 journal.json = page meta and every grid = the builder; J2 the sweep is a no-op; J3 structure
on all 38 family pages (proven RED on a seeded navy literal + ticker); J4 the filter in a real browser at 390 and 1440
(topic chip → exactly the JSON's count, URL written and restored, search, Clear, empty state, aria-pressed); J5 computed
fonts/backgrounds, composited layers ≤ 20, a fact list renders as full-width rows.
Changed
articles.html, insights.html, geopolitics.html, future-forward.html rebuilt on the editorial register: graphite
surfaces, one amber accent, IBM Plex Sans headings, mono only for the metabar, eyebrows, counts and tags. The hub keeps its
legal notice (one hairline panel), the case-library row, the writing principles and the framework as lists, not tiles.
The landings lose the amber stat strips, navy category tiles, "coming soon" tag clouds and their three inline copies of
the old search modal (the shared palette was already loaded). Heroes are a plain H1 under one mono metabar line
css/rz-article-dark.css: --rz-art-serif now resolves to the sans stack (the token stays so 43 pages keep their
selectors), --rz-art-accent2 (the second, green hue) → muted text, the diagram --dg-ok and the derived-chip green →
neutral / amber, the hanging quotation glyph loses Fraunces; the family's navbar surface is graphite when scrolled.
tools/audit-vibecode.mjs: the journal family joins strict scope (JOURNAL_STRICT); stat-strip counts only tiles
outside a data-rz-fact-list parent and outside <script> (a print/PDF template is a paper document), with the vocabulary
widened to evidence|figure|number tiles (the detector had passed article-26's hero strip; the owner's eye had not);
aurora-navy matches a class or id named aurora-*, not the word in prose (Oklo Aurora is a reactor).
tools/test-audit-vibecode.mjs: the monitored fixture moves off article-99.html (now strict) and a new fixture proves
the eight tells GATE on the journal family. Site-wide monitored findings 538 → 407.
assets/article-29-hero.webp, assets/article-29-card.webp, assets/og/article-29.webp: the owner's two renders for the
coming article 29, hero/OG and card; the builder prefers <id>-card.webp for the grid when it exists.
Research staged, not yet published:Article/Article_29/ (how an LLM works, tokens to frontier hardware: 76 sources,
5 calculator specs, 10 diagram specs) and Article/Article_30/ (the "Agentic AI" debate: 63 sources, Elder's 14 claims
graded) and standarization/Audit result/seo-geo-2026-10-16/ (SEO + GEO strategy, 14 new ledger rows). Content, not code.
v3.22.10PATCH
— "Hardware"
Changed — the 3D gets its detail from pixels and runs smooth (js/rz-hub-3d.js)
Owner on the live v3.22.9: "3Dnya masih sangat kurang detail, dan nggak smooth". The rack had
carried every grille, cage, LED, ear and label of all 37 trays as meshes — about a thousand draw
calls under a 2048 shadow map. Each rack-level tray is now two meshes: a body and a faceplate
whose grilles (honeycomb), OSFP cages with port slots, status LEDs, ears with thumbscrews and the
slot label are painted once on a canvas per kind and reused (faceTexture). Power shelves show
six PSU modules with handles and LEDs the same way; the OOB switch its 24 ports. Draw calls fall
roughly ten-fold; the shadow map is 1024 on phones; the pixel ratio caps at 1.25 on desktops.
The superchip PCBs carry a painted board — traces, vias, silkscreen outlines, GB300 · SUPERCHIP
— instead of a flat green. Each cold plate has inlet and outlet nipples and a doubled serpentine
groove; each package sits on a gold substrate rim; capacitor rows run along the boards; the lid
carries vent slots; the rails carry screws; a front fill light makes faceplates read.
Two temperature tags ride the quick disconnects — 40 °C in, 50 °C out, the engine's TCS
planes — and follow the parts through the exploded view.
v3.22.9PATCH
— "Hardware"
Fixed — the 3D floor no longer mirrors the studio (js/rz-hub-3d.js)
The first render with the environment map showed a white glare band across the floor under the
rack and the tray — the floor was half-metal and the studio's light bands were near white. Floor
to a matte dark plane (metalness .2, roughness .78), the environment bands at 55 % and the sky
gradient two stops darker; the aluminium, copper and steel keep their reflections.
v3.22.8PATCH
— "Hardware"
Added — the hub mockup gets the hardware in 3D, an icon per tool, a running panel per group, and the small interactions (datacenter-solutions1.html, noindex)
Owner, with nine Oxide screenshots: *"Bisa nggak kamu buat 3D nya utk B300 servernya dan rack
nvl72 … bisa explode view saat di klick button exploded", "micro interaction yg seperti linear"*,
"antara group of tools dikasih gambar2 ilustrasi animasi", "per tools … ada icon yg unique",
"terlihat cold plate, alur air dan terlihat perpindahan panas", "Server jangan hitam doank".
js/rz-hub-3d.js — section "02 / The hardware". The GB300 NVL72 rack, its compute tray and
its NVLink switch tray built procedurally in WebGL (the site's js/vendor/three-r170.min.js,
loaded on approach) from a declared dimension table. Published counts from DCAI_MODEL.gb300:
18 compute trays, 9 switch trays, 8 power shelves of 6 × 5.5 kW, 2 superchips a tray (Grace +
2 × B300), 4 CX-8 + 1 BF-3, 2 switch ASICs a switch tray, 72 GPUs. Adopted and printed as such:
the 48 OU frame, 1 OU a tray, the stacking order, the placement inside a tray, four front fans for
the air-cooled share. Inside the tray: two PCBs, Grace under a silver lid with LPDDR5X around it,
each B300 die with eight HBM stacks, VRM rows, a copper cold plate per chip with a serpentine
groove and a fin field, the coolant loop as tubes — supply blue, return amber — and the water as
moving dots that warm from blue to amber as they pass the CPU and both GPU plates (TCS 40 → 50 °C,
the engine's p07/p08), GPU dies pulsing with heat, quick disconnects, an 18-slot NVLink block, the
busbar clip, a translucent lid. The rack: aluminium trays with fan grilles and cages on every
faceplate, PSU handles, OOB ports, busbar, supply and return manifolds with water rising and
falling, the cable cartridge. Orbit with damping, idle rotate until touched, hover = part lit in
amber + readout, click pins, numbered tags projected from 3D, a part list with prev/next,
Exploded view tweens trays out of the rack and lid / plates / NICs / rear parts out of the tray,
Reset. A published-spec row under it (GPUs, NVLink, memory, rack IT, installed power, feed
current, TCS) with registry ids. No WebGL → a sentence, the counts stand. Honest boundary printed
on the figure: no public CAD exists; counts published, placements adopted.
js/rz-hub-icons.js — 41 glyphs keyed by href: 15 from RZEngSymbols (the IEC/ISA glyphs the
drawings use), 26 authored in the same 48-box / 2 px style; a 32 px tile tinted green
(cockpits, engineering, labs) or blue (planning, market, grid, knowledge), hues outside the
neon detector band, first column of every row.
js/rz-hub-illus.js — one live panel above each group, ticking only while on screen: the ATS
sequence on the engine's 171 × 4,000 kW pool with its 24 h fuel; the Cx L0–L6 gates; the
build / colo / cloud cumulative shape, labelled "a shape, not a result"; PUE across the engine's
seven bins with a sweeping cursor, red on the cliff bin; the Java–Bali 500 kV chain; electricity
in = heat out (539.05 MW IT → 424.9 MW th liquid + 142.0 MW th air, heat.*); the knowledge
titles stepping. Reduced motion: final frame, no ticking.
Micro-interactions: a sliding ink under the active tab (both tab bars), scale(.98) on
press, the OPEN ↗ arrow nudging on hover, j / k over the 41 rows with a cursor row, click a
hero key to copy its registry id, a ⌘K chip that opens the palette. Not the cursor-following
glow (rejected site-wide).
Fixed
audit-legibility --strict on the mockup (reported by the LTC session): the year strip's
"600 h · design" and "160 h · cliff" were clipped (buttons sized by hours now never go below
their label); the IEC genset glyph's 13-unit "G" rendered 6 px in the plant figure and is drawn
as a stroke path in that figure instead.
v3.22.7PATCH
— "Envelopes"
The two table-only chapters of the engineering report get their figures
Why. After v3.22.2 every other chapter of the LTC report opened with a drawing; Tier topology and Standards cross-check were still a table each, and the Tier page ran half empty. A Tier verdict is a sweep of elements, and a standards check is a position inside a band — both are pictures before they are sentences.
What changed (js/ltc-report.js):
Element-out matrix (tierMatrix), one for the power system and one for cooling: a row per removable element, a column per Tier, each cell split planned | unplanned, green when the load is still carried, red when a group is lost, a dot where that Tier's topology has no such element. It reads straight off the Tier model's own sweep (tierSweep now returns the per-element rows), so the matrix and the summary table cannot disagree. The default plant shows Tier I–II red on TX, UPS, PDU and MCC, Tier III red only in the unplanned half of its active path, Tier IV all green.
ASHRAE envelopes (envelopes): the hall's supply and return air and DH01's hottest rack inlet under every modelled fault against the recommended band and the A1–A4 allowable classes; below it the facility water supply against the liquid classes W17 … W+, the class the FWS supply falls in lit, the TCS loop ends marked. Coincident fault inlets are listed once per value, so nine faults at 24.0 °C are one line, not nine overlapping labels.
Standards cross-check gains the row "DH01 max rack inlet, worst fault" (the default plant: 34.0 °C with four CRAHs lost, beyond the A1 allowable 32 °C), so the figure's worst point is also a sentence in the table.
Gate (tools/test-ltc-workbench.mjs, W12en, 357 assertions): the matrix is read back from its data attributes and compared cell by cell with LTCTier.sweep for all four Tiers of both systems (80 cells); the envelopes' points must equal the engine's supply, return, every fault's max inlet and the FWS/TCS ends, the air bands must be ASHRAE's, and exactly one liquid class is lit — the one the FWS supply falls in — before and after a colder TCS supply moves it. Mutants: swapping the planned/unplanned attributes (2 assertions red) and plotting the return temperature as supply (1 red).
v3.22.6PATCH
— "Plant"
Changed — the hub mockup's hero is the plant, running (datacenter-solutions1.html, noindex)
The cockpit screenshot is gone. The owner, on the v3.22.5 mockup: *"tampilan ss datacenter
nggak usah", "gak perlu ada 2 card itu", "nggak ada buttonnya, atau indication itu clickable"*,
"sangat kurang immersive". The hero is now ONE figure drawn by RZDiagram — the 150 kV intake
through GIS, main transformers, 20 kV board, unit substations and UPS to the racks, then the
liquid path through the CDUs to the dry coolers and the air path through the fan walls to the
chillers — with IEC 60617 / ISA-5.1 glyphs and every count from DCAI_CALC.snapshot.
The design year plays through it. The seven weather bins (bins[0..6], width ∝ hours) sit
above the drawing as a strip; the page advances one bin every 2.6 s while the hero is in view
and the tab is visible. Ambient, PUE, facility draw and chillers-running move between the
engine's bin values (620 ms tween between two published numbers, never a count-up from zero);
at the 36 °C bin the chillers go to 142 of 143, the liquid path reads "on compressors" in the
alert colour and the chiller node switches state. Hover, focus or tap a bin to hold it; the play
resumes after 7 s quiet. prefers-reduced-motion: no autoplay, no tween, no pulses.
The figure draws itself left to right on first paint (each element staggered by its
x-position, 46 ms a step) and every node answers the mouse: hover or focus reads its rows into
the readout below, click pins it, double-click or Enter opens the page where it is computed.
Phones get the same plant as two stacked sheets (power, heat) so the glyphs stay readable.
Every catalogue row carries an Open ↗ control at its right edge and the row's index, name
and control turn amber together on hover and focus. The four cells are placed explicitly on the
grid, so the lock icon auth.js prepends to a gated link lands inside the name cell instead of
taking the first column — which is what broke A.03 and B.03 on the owner's phone.
Rows arrive in a 35 ms stagger as each group scrolls in (IntersectionObserver, off under
reduced motion). The four vignette cards and the separate "The engine" section are removed; the
heatPath; figPower / figHeat for narrow viewports; tween, stagger, yearRender,
keysPaint; leaders, miniPue and the vignettes deleted.
v3.22.5PATCH
The mockup becomes a catalogue: every tool visible, one render, nothing split
Fix (owner, on the live v3.22.4 mockup): *"tools2nya malah tidak terlihat … gambarnya excessive … jangan
ada split screen"*. datacenter-solutions1.html is rebuilt around the hub's actual job — the tool list:
41 rows (9 cockpit pages, 6 cost and planning, 3 strategy, 9 engineering and compliance, 2 grid monitors,
4 labs, 8 knowledge) as dense two-column rows with index, name and one line, nothing behind a click, no
classifying tag. Images cut from seven to one: the cockpit render in the hero, with the keys row above it and
the leader rails in the gap, not across the cockpit's own bar. The strip of four renders, the platform
images, the map card and the 2×2 cards are gone. No two-column layout remains: the hero stacks, and the
FIG. 2 readout moved from a side column to a row of cells under the drawing. FIG. 2 (power chain · heat
path · campus), the weather-bin strip and the four live vignettes stay.
Gates: static set green; browser gates not re-run (only the noindex mockup changed).
v3.22.4PATCH
A product-page mockup of the DC Solutions hub, published noindex for the owner's review
Added:datacenter-solutions1.html (noindex, nofollow) — the third take on the hub after the owner
rejected the v3.22.0 "Stage" page (*"split screen tanpa interaksi mouse, terkunci scroll, gambar nggak
accurate, tag free/gated = slop juga") and set the bar at the live homepage ("10:4, harus surpass"*).
Borrowed from Oxide Computer and Linear, read live: the product itself as the hero render (the real DC AI
cockpit, screenshotted in the authorized audit state, assets/hub/*.webp), a keys list whose leader lines
are measured from the DOM to the chips on the render, FIG. 2 drawn through RZDiagram with the
IEC 60617 / ISA-5.1 glyphs the cockpits use (power chain · heat path · campus to scale), hover/focus
readouts that print the engine's own rows and registry ids, a weather-bin strip (24 / 30 / 34 / 36 °C) that
re-reads S.bins and flags the chillers at the worst bin, a strip of four cockpit renders, four live
vignettes (PUE by bin, temperature planes, capacity, fleet) and a Linear-style live instrument inside the
calculators card (PUE at your ambient, interpolated between bins). No tag classifies an item; nothing follows
the cursor; no sticky stage. js/rz-hub-product.js. The live datacenter-solutions.html is untouched
until the owner approves.
Gates:audit-vibecode --strict adds the mockup and js/rz-hub-product.js to the strict hub family;
static set green (script-tags, js-syntax, version-stamp, css-structure, vibecode, vibecode fixtures,
mobile scorer, crawler, harness parity, changelog, gate-wiring, cache tokens). Browser gates not re-run:
the only page that changed is the new noindex mockup; v3.22.0's full run stands for the live pages.
v3.22.3PATCH
Every verdict that is not "holds" names what moves it — and the plant is re-solved to prove it
Why. v3.21.19 gave the report a verdict per chapter with the line each was judged against. A reader then asked the natural next question of every "review" and "fails": what would fix it, and would it? The answer was scattered across the Options chapter, the fault tables and the Tier blockers.
What changed (js/ltc-report.js, levers()): the verdict table gains a fourth column, "What moves it". For each chapter that is not holding, the report picks one concrete input change, re-solves the whole plant with it — design day, climate bins, DH01 faults, the heat load test, the transients, the Tier fit — re-judges every chapter on the new plant, and writes the result it reaches as "then holds / review / fails" next to the lever:
Liquid path: the better of adiabatic coolers, fluid coolers and TCS supply +3 °C, by trim-chiller hours; stated as trim hours before → after, WUE before → after and the running-cost delta.
Air path: N+1 on whichever of CRAHs or chillers has no spare, with the units added.
Electrical & sizing: CDU feed dual (ATS per CDU).
Redundancy register: one more unit of each short class, through its redundancy input.
Annual performance: the best other configuration or the best single design step from the sensitivity table, whichever reaches further (by WUE when WUE drives the verdict).
DH01 hall & faults and Heat load test plan: the lever each failing single fault implies (a lost feeder → dual feed; a lost CDU, CRAH or chiller → one more of it), with the faults still failing after it.
Transients: double the loop volumes (buffer tanks), with the shortest ride-through before → after.
Standards: supply air to 24 °C. Tier topology states its blockers; the screening layer names its lowest screen.
Chapters that hold say "holds; no change needed". A cost-neutral lever says "running cost unchanged".
On the default dry-cooler plant: adiabatic coolers take the trim chillers from 3,160 h to 0 h a year, −1,031,124 cost/yr, then holds; dual CDU feed clears the DH01 fault and the heat load test, running cost unchanged; doubling the loop volumes takes the shortest ride-through from 188 s to 376 s.
Gate (tools/test-ltc-workbench.mjs, W12lv, 353 assertions): the gate applies each lever's own input change, re-solves and re-judges the plant itself, and every lever must improve its chapter and reach exactly the verdict it claims; the liquid lever's trim hours must be the engine's and fall, the transients lever's ride-through must lengthen, holding chapters carry no lever, and the rendered table carries a cell per row and a "then …" tag per applied lever. Two mutants proven red: a selector that picks the worst candidate, and a lever that proposes the single feed it is meant to replace.
v3.22.2PATCH
The engineering report draws its four table chapters
Why. v3.21.18 made the report a paper document and v3.21.19 gave it verdicts, a glossary and references, but four chapters — Liquid path, Air path, Electrical & sizing, Redundancy register — were still tables only. A reader had to assemble the temperature chain, the loss chain and the one-unit-out margins in their head. The owner's standing brief for this report: "super comprehensive, detail, immersive, clear".
What is drawn (js/ltc-report.js):
Liquid path — a temperature ladder. One bar per loop on a shared °C axis, cold end to hot end: the TCS loop at the cold plates, the facility water loop, the liquid coolers from their sink (dry bulb, or the adiabatic sink) to their leaving temperature, and — when the coolers cannot reach the FWS supply — a red bar for the gap the trim chillers close, with the share of FWS heat they take. The dashed links carry the CDU heat-exchanger approach and the cooler approach. On the default dry-cooler plant the picture shows why 70 % of the heat is on trim chillers at the design hour: the coolers leave at 39 °C against a 32 °C supply.
Air path — a second ladder. Hall air supply to return, the chilled-water loop at the CRAH coils, and the condenser side: wet bulb plus tower approach for a water-cooled chiller, dry bulb for an air-cooled one. The links carry the coil approach and the chiller's minimum lift.
Electrical & sizing — a staircase. From the facility meter (M0) down to the IT input (M3), one step per loss or branch: mechanical (MCC) with rooms and auxiliaries, transformer, distribution, UPS, the TCS pumps when they sit on the UPS, PDU, cords. The UPS input, M1 and M2 are marked where the stair passes them.
Redundancy register — capacity bars. For every equipment class: installed capacity in grey, the design load as the bar, and a tick at the capacity with one unit out. A bar past its tick is short.
Glossary: HX.
Gate (tools/test-ltc-workbench.mjs, W12ld, 349 assertions): each drawing carries its numbers as data attributes, so the check reads the drawing. The ladders must carry the engine's temperatures with the cold end below the hot end, the trim bar must disappear on an adiabatic plant, the condenser side must switch to dry bulb for an air-cooled chiller, the staircase must close (facility − every step = IT, with and without the TCS pumps on the UPS), and the register ticks must sit at (installed − 1) × unit with a short bar appearing when the CRAH spare is removed. Two mutants proven red: a dropped staircase step, a tick at installed × unit.
v3.22.1PATCH
The gate learns the eight tells the owner found on the hub — and its own fixture suite joins the ship gate
Fix:tools/audit-vibecode.mjs gains one detector per ANTI_VIBECODE §A row 27–34: gradient-text,
glass-token (the --glass-* family, not only the blur), stat-strip (≥ 3 stat/metric/kpi tiles outside the
instrument register), hover-lift (any :hover lift ≥ 2px on a decorative block, whichever block holds the
transition — the split that let the hub's tiles through), keyword-marquee, neon-green and aurora-navy
(by HUE BAND, not one hex: neon h 95–170° s ≥ .9; navy h 205–250° s ≥ .3 l ≤ .14 — graphite #0b0c0e is
outside, #0a0e1a and the site's #0f172a inside), serif-display-chrome (a serif display face, or
mono-uppercase on nav/buttons). Proven RED 8/8 against the v3.21.19 datacenter-solutions.html, GREEN on
the v3.22.0 page. Scope: GATING on the hub family (HUB_STRICT), MONITORED with a printed per-rule count
everywhere else, flip condition written in the standard. Monitor rows today: aurora-navy 162 · serif 126 ·
278 files — the honest backlog, every row sampled and real).
Found on the way — the fourth "gate that exists but is not wired":tools/test-audit-vibecode.mjs was never
in ship-gate.sh. Four of its fixtures seeded 8px against the audit's > 8 ceiling (changed with the
register rule in v3.11.12) and had sat red since; nobody ran it. Seeds moved one step above each ceiling
(9px / .625rem / editorial 11px), suite 20/20, and it is now gate 1 of the anti-vibecode pair.
Gates: the static set from the worktree — script-tags, js-syntax, version-stamp (272/279), css-structure, anti-vibecode --strict (0 gating, 539 monitored), anti-vibecode fixtures 20/20, mobile scorer, crawler 36/36, agent-harness parity, changelog --check, gate-wiring, ltc-workbench 344/344, cache tokens. The browser gates were NOT re-run for this ship: it changes tools, standards, the changelog and cache tokens only, and no page markup or stylesheet; v3.22.0's full run (123/123, probe 82/82) stands for the rendered surfaces.
v3.22.0MINOR
"Stage": the DC Solutions hub becomes one drawing and one camera — and the gate learns the tells it was blind to
The owner's complaint, on a screenshot of datacenter-solutions.html: "ini ai design slop", then
"warna backgroundnya masih ai design slop", then "masih ada hijau neon, itu ai design slop juga".
audit-vibecode --strict reported the page clean while it carried gradient text, a pulsing badge,
three CTAs, four glass stat tiles, a 24-item ticker, 20 floating particles, 48 box-shadow, 39
radii ≥ 8px, 20 transition: all and 18 hover lifts in 2,765 lines of inline CSS.
The page now. Register borrowed from Oxide Computer (numbered figures, hash gutter, leader rows,
one accent, a five-word nav) and Linear (one headline, one action, 13px/4px row density); tokens are
RZ. The copy scrolls in one column; in the other, ONE dotted isometric drawing is pinned and a camera
moves to each section's part of it (js/rz-hub-scene.js, css/rz-hub.css): the campus as the DC AI
engine lays it out (site 412 × 231 m, the bands, roads, fence, genset yard; chillers 143, dry coolers
631, gensets 171 — read from DCAI_CALC.snapshot, printed "—" when the engine is absent, never
typed), transmission tiers into the GIS, the power/cooling chain as an inspection panel whose boxes
are links to the calculator rows (hover a row: its box lights; click a box: its calculators), and
the standards stack. Labels appear per camera; flow runs only while a section with flow is in view;
nothing follows the cursor; reduced motion gets cuts instead of moves. Phone: the drawing is a 42 vh
strip pinned above the copy. Surfaces are neutral graphite (#0b0c0e / #121316 / #191b1f), no
aurora, no mesh, no green: one accent, signal amber, with a contrast-safe text variant for light mode.
Hero: eyebrow, one Fraunces headline with a muted second clause, one lead, one solid 44px action and
one text link, a mono metabar (21 calculators · 7 engines · 9 cockpits · 0 signups). Five numbered
sections (Platforms, Grid, Calculators, Knowledge, About) replace eight; five card families become one
row family and one tile family; trust is a list, FAQ is native <details>, the footer is three flat
columns. The reviewed root-card gate (handleRootCardClick, #rootStandardsCard) is carried over
verbatim and the crawler suite (36 tests) still passes on it. 4,230 lines → ~500.
The gate.ANTI_VIBECODE_STANDARD.md §A gains seven tells (27–33: gradient text, glass by token,
stat-tile strip, hover lift, marquee anywhere, neon green, aurora/navy cast — the last two withdrawn
by the owner from §B). documentation/design.md §5 records the graphite decision. Detectors for
27–31 are the next ship; this one records the tells and the pages that carry them (gradient text 39
Found on the way — two audits that scanned nothing.audit-mobile-responsive.py and
audit-script-tags.py excluded any path containing .cache by matching the ABSOLUTE path, so in a
git worktree under ~/.cache they walked zero pages and exited 0 (Pass: 0 Fail: 0); every ship
gate run from that worktree carried two vacuous audits. Both now match the repo-relative path and
FAIL on an empty scan (291 pages scanned, 182 pass, 76 noindex, clean).
Also in this ship:index1.html, a noindex mockup of the homepage in the same register (career
as a single-line diagram, readouts panel, certification plates; js/rz-home-scene.js) for the
owner's review before index.html is touched.
Gates:bash tools/ship-gate.sh 123/123 and probe-accuracy-validation 82/82 from the worktree. The first full run failed one gate, test-mobile-forms (the command-palette input rendered 13.3px at 390px because this page no longer loads styles.css, where the site-wide 16px floor lives); the floor is now repeated in css/rz-hub.css and the gate re-run green. audit-mobile-responsive.py and audit-script-tags.py now report 291 pages scanned, where before this ship they reported 0 from a worktree.
v3.21.19PATCH
"Verdicts": one judgement per chapter, a glossary that covers every abbreviation, numbered references
Why. v3.21.18 made the report a paper document; the owner's standing bar is "detail dan sangat clear". A 27-page report is clear only if a reader knows where to look first and never has to ask what a term means.
Report (js/ltc-report.js):
Verdict by chapter after the key findings: twelve rows (thirteen from the Model page), each a sentence of the engine's own numbers with a status — holds, review, fails — and a link to the chapter. The lines are written into each verdict, not hidden: PUE against 1.25 good practice and the Uptime 2024 survey average 1.56; WUE against 1.0 and the LBNL typical 1.8 L/kWh; ride-through against 5 min (a controls response) and 1 min; DH01 judged on the single-element faults the design basis covers (a CDU, a CRAH, a chiller, a CDU feeder), with the beyond-N+1 cases reported, not judged. On the default plant the single CDU feed shows up three times — electrical, DH01, heat load test — and a dual feed clears all three.
Glossary: 50 entries, every abbreviation the report uses, from A1 to WUE, including the metering points M0–M3 and the liquid classes.
References: nine numbered sources (ASHRAE TC 9.9 thermal and liquid guidelines, Uptime Tier Standard and 2024 survey, ISO/IEC 30134 parts 2/3/6/8/9, NFPA 75 and 2001, ANSI/TIA-942, OCP, LBNL).
Gate (W12v, 344 assertions): each status is checked against a plant that must flip it (dual CDU feed, no CRAH spare, the cheapest configuration); every abbreviation token in the built report must be a glossary term; every verdict links to the chapter it judges; references and glossary are in the contents. Two mutants red: a verdict that ignores the plant, a glossary entry renamed.
v3.21.18PATCH
"Paper": the engineering report is a PDF document, and the Model page exports it too
Why. The owner opened Export Report on the Model page: "export reportnya jelek sekali — harus PDF export dan harus super comprehensive, levelnya bener-bener detail dan sangat immersive dan clear." That button downloaded a plain-text executive pack; the Report tab's "Export PDF (Single)" was a 127-page flat dump of the lab model's tables. The six workbench pages already had the plant's engineering report (js/ltc-report.js, v3.19.0), but the Model page never loaded it, and the report itself was a web page: no cover, no page breaks, no page numbers, figures cut wherever the paper ended.
The report is now a paper document (js/ltc-report.js, per documentation/design.md §11):
Cover: title, plant and climate, twelve headline figures, author, date, engine version, source URL, licence, scope; then a contents page listing every chapter with the question it answers.
A4 pages with a running head (site · report title) and "page n of m" in the margin boxes (CSS @page, Chrome 131+); none on the cover. Chapters that open on a figure start a new page; the short table chapters (liquid path, air path, electrical, redundancy, standards cross-check, assumptions, data) follow on. Table heads repeat across pages; figures, rows and findings never split. DH01 plans three to a row, transients two.
Opens the print dialog itself once the fonts settle (bounded to 2.5 s), where "Save as PDF" writes the file; a Save as PDF button repeats it. Vector throughout: every figure is SVG, every number searchable. The HTML, JSON and CSV downloads stay.
Plant drawings on any page. The hub's plant map is painted inside a hidden frame that carries the exporting page's stylesheets, the hub's sheet, and the site's light-theme colour tokens read from the page — from its root or from the drawing scope the Model page uses. Before, a page without the hub's stylesheet (the Model page) exported the four schematics as black rectangles.
On screen the same file reads as a stack of sheets.
The Model page (ltc-system-modelling-lab.html, js/ltc-lab-ui.js): Export Report in the nav and the new Export PDF report button in the Report tab open this document on the page's own engine result, with a Design screening chapter right after the findings: the weighted screen (radial), the five standards screens with their weights, the four design dimensions against their 60–100 band, risk index, confidence and operating scenario, the model validation table (each check, value, band, red/green marker, source) and every lab input as set. The key findings state the screen. The older flat dumps stay in the Report tab as "Legacy audit tables".
Gate (tools/test-ltc-workbench.mjs, W12pr, 338 assertions): cover and contents on every chapter, the @page rules and page counters, which chapters open a page and which flow, the print hook, the screening chapter present only when opened from the Model page with its scores, validation rows (2 of 3 in band, one red marker), inputs, title in the running head, and the Model page's script order and button wiring. Headless check: 25 pages from the Model page, 23 from the hub, no NaN, schematics in colour on both.
v3.21.17PATCH
"Roads": the campus becomes an engine layout — roads, fence, gates, bands, and a site area that is a product of lengths
Why. The owner looked at v3.21.9 again: "masih kurang sempurna". The site plan was yards floating in a boundary scaled to (building + yards) × 1.3: no roads, no fence, no gate, the fuel farm one road from the oil-filled main transformers, and no feed-B route. The roof clipped two of its bank labels on the plate edge, and the hall level printed its gallery labels across the fan-wall ribbon.
campusRoadFactor 1.3 is retired. New leaves site.roadM 12, site.setbackM 6, site.stpM2 240 (all ASSUMED, sourced in the model).
The arrangement is published: three east–west bands on 12 m roads (north: GIS · main transformers · STP · fuel farm; middle: the building plate · genset yard; south: chiller plant · dry-cooler yard), a perimeter road inside the fence, the setback to the property line. Outputs: genset_yard_width_m, north_band_depth_m, south_band_depth_m, campus_width_m, campus_depth_m, site_width_m, site_depth_m, site_area_m2 = width × depth, roads_m2 = all that is not a block.
Site 412 × 231 m = 95,126 m², of which 33,441 m² is roads, verges and setback (was 79,879 m² from the factor, which left no room for a perimeter road). Ten identities in test-dcai-engine.mjs (588 pass).
Site plan (datahallAI.html):
Property line, setback, fence, road surface and centre lines, two gates (main gate at the west on the road under the entrance face; service gate at the east under the fuel farm) and a gatehouse — page-authored and declared.
The STP stands between the transformers and the diesel. Each main transformer in a bunded bay behind a fire wall. Feed B 20 kV drawn down the east road into switchroom B; feed A into switchroom A.
Dimensions: site width and depth, the three band depths. Schedule gains STP hooked (site.stp_m2) and a "Roads, verges and setback" row (site.roads_m2); the site row prints W × D.
Roof: bank labels inside the bank on an opaque pill above the units. Hall level: gallery labels centred in the free band between the hall edge and the fan-wall ribbon.
"Coast-Down": the fans do not stop at once, and every rack reports when it crossed 27 °C
Why. v3.21.15 solved a CRAH trip from normal operation, but took the tripped fans to stop at once, so the first rack crossed 27 °C at 1.0 s. Real fans coast down over tens of seconds, and the survivors ramp up under pressure control. The owner asked for more accuracy, and an operator wants to know which racks go first.
Changed (js/ltc-cfd.js):
Fan coast-down. At the fault, faultFrom lists every prescribed face that differs between the normal and the faulted solve: a tripped unit's supply and return faces, and the survivors' speed-up. Each one coasts from its normal velocity to the fault's with a 6 s time constant (95 % of the change by 18 s). A coasting inflow face keeps its supply source, so the air it still brings carries its temperature.
Every unit lists its faces. A tripped unit's supply and return faces exist with zero velocity, so the heat report counts a coasting unit's return at its present velocity. During the transient the report reads "removed 3,520 kW (−10 %, the room air is warming)", which is the heat the room stores while it heats.
Per-rack crossings. During a transient the solver records when each rack first crossed 27 °C and 32 °C. report() adds t27, t32 and ramping.
Shown:
Panel: "t = 17 s after the fault · fans coasting · over 27 °C: first rack at 6.0 s, 40 racks by now, half of them by 12 s".
Rack tooltip: a "CFD after the fault" row, "over 27 °C at 8.4 s · over 32 °C at 21.0 s", or "still under 27 °C at 65 s".
Guided tour: the hottest rack's stop says when it crossed 27 °C; the summary gives the first crossing and the time by which half the crossing racks had crossed.
Legend: states the coast-down and its time constant.
Four CRAHs out, 0.6 m, fans stopping at once → coasting:
At once (v3.21.15)
Coasting (this release)
First rack over 27 °C
1.0 s
6.0 s
Hottest inlet at 5 s
31.6 °C
26.1 °C
Racks over 27 °C at 15 s
52
35
Half of the crossing racks over 27 °C by
—
15 s
Settled hottest inlet
34.9 °C
34.7 °C
The steady state is unchanged; only the first half-minute moves.
Gate W12tr now also requires:
a tripped unit's supply face at its normal velocity at t = 0, at e^(−5.8/6) of it after the step ending at 6 s, and under 1 % of it at 40 s; a survivor's face rising the same way;
the coasting inflow keeping its source: no room air below the supply temperature during the first 12 s;
each rack's crossing consistent with the reported first, and no rack over 32 °C before it was over 27 °C;
the legend stating the coast-down time constant.
It was proven red with the ramp removed, with the coasting inflow's source dropped (room air then fell to 0 °C), and with the 32 °C threshold lowered.
v3.21.15PATCH
"Step Response": a sudden fault is solved from normal operation, so the CFD shows how fast the hall heats
Why. The DH01 CFD started every solve from still air at supply temperature. Its convergence curve was a start-up artefact, not the hall's response to the fault. For a CRAH or CDU loss, what an operator needs is how long before racks go over 27 °C. That needs the fault to strike a hall already in normal operation.
Sudden faults (one, two or four CRAHs, a CDU, a CDU feeder) now run in two phases:
1. 15 s of the no-fault solve. The page sends it as pre, a balance solve with no fault.
2. The fault at t = 0. faultFrom carries the settled temperature and velocity fields over, applies the fault's fixed faces and projects the field so it is divergence-free.
Time-averages start 45 s after the fault, once the transient has passed.
Time-driven faults stay the steady field at the replay minute: chilled water, chiller and facility water, where the plant transient already carries the time.
Panel: "normal operation before the fault, t = −8 s", then "t = 32 s after the fault · first rack over 27 °C at 1.0 s". A transient solve is called settled after 45 s.
History chart: negative time is the pre-fault phase. A "fault" marker sits at t = 0, and the axis reads seconds after the fault. The balance label moved below its line, where it no longer collides with the hottest-inlet curve.
What it shows, with four CRAHs out:
Before the fault, every rack inlet is at 24 °C with return at 35.2 °C.
After the trip, the first rack crosses 27 °C at 1.0 s, as hot-aisle exhaust is pulled over the rack tops.
The hottest inlet reaches 31 °C by 5 s and settles at about 35 °C within 40 s, the same steady value as the old solve from rest.
Return air climbs to 39.3 °C.
The fans are taken to stop at once. A real coast-down adds some 10–20 s, and the legend says so.
Gate W12tr, new. On the default plant, with four CRAHs tripping after 15 s of normal operation, it requires:
the hall inside 27 °C before the fault;
the fault starting from exactly the settled field, with max |ΔT| 0;
the projection after the switch at or below 5e-3;
the reported first crossing of 27 °C equal to the crossing measured step by step, within 10 s;
the hottest inlet climbing toward the recorded four-out steady 34.8 °C, within 2 K and without overshooting past it;
the legend stating the pre-fault time the worker uses.
It was proven red with no projection after the switch, with the temperature field not carried over, and with the crossing threshold changed.
v3.21.14PATCH
"Walkthrough": a guided tour of DH01 that reads its captions off the solve
Why. The owner keeps asking for the CFD to be "super rich and immersive". A reader dropped into the 3D view still had to find the problem rack on their own. The tour takes them to it.
Added:
Guided tour button in the 3D camera controls (js/ltc-dh01-3d.js), driven by LTCCFDView.tourPlan (js/ltc-cfd-view.js, pure). It makes six camera stops, each with a caption built from the current solve:
1. The hall. The fault, the hottest rack inlet and which rack it is (CFD when solved, else the balance solve), plus RCI HI and RTI.
2. The cold aisle of the hottest rack. The range of its rack inlets.
3. That rack. Its inlet and its inlet at the bottom, middle and top. Smoke is released automatically at its top front, so you watch the exhaust curl over the top.
4. Its hot aisle. The row's mean exhaust on its way to the plenum.
5. The ceiling plenum. Return and supply temperature at the coils, and the CRAHs out.
6. Summary. Racks over 27 °C in the CFD and in the balance solve. With four CRAHs out, after v3.21.13's recalibration, that reads 68 and 68.
Each stop eases in over 2.6 s and holds 4.2 s, or 7 s at the smoke stop. The tour ends on its own, when you press the button again, or when you drag the view. With reduced motion it cuts instead of flying.
The caption sits at the bottom of the stage and is announced to screen readers.
Gate. W12a now also checks the tour:
six steps;
the named rack is the CFD's hottest, and its inlet is in the first caption;
the smoke point is in room air, outside any rack or contained hot aisle, on the cold-aisle side of the rack face;
no caption carries NaN or undefined, and every camera coordinate is finite.
It was proven red with the coldest rack picked and with the smoke moved to the rack's hot side.
v3.21.13PATCH
"Recalibrated": the DH01 balance solve now puts recirculated air where the CFD does
Why. v3.21.11's rack-by-rack chart showed the balance solve and the CFD disagreeing by 4.4 K RMS per rack with four CRAHs out. The owner approved recalibrating the balance on the CFD. The balance solve drives the plan, the rack tooltips, the HLT verdicts, the sequence of events and the report, so this changes numbers across the workbench.
What was wrong. The balance marched each cold aisle as plug flow from both fan walls, so all of the unmet airflow landed on the racks the supply never reached, at the dead units' end. The CFD, on the 0.3 m grid with the second-order schemes, disagrees in two ways:
Position. The heat sits around where the two walls' supplies meet, with exhaust curling over the rack tops there.
Spread. A large shortfall spreads over every aisle; a small one stays near the dead unit.
The change (solveHall step 5a, js/ltc-dh01.js):
The unmet airflow is shared out before the hot-aisle fixed point.
Up to 75 % of each aisle's shortfall is pooled and handed back to the aisles by demand. The share rises linearly to that 75 % as the hall's shortfall reaches 5 %.
Within each aisle, 25 % follows the old march and the rest follows a hump centred where the two walls' supplies meet: a narrow core (0.4 of the half-length) plus a 30 % wide shoulder.
No rack draws more than its own demand.
Mass is conserved: the recirculated airflow still equals the shortfall to 1e-12 (W12r). Each aisle's mouth flows are rescaled to the supply its racks now draw, so the 3D streams still sum to the solve (W12x).
With no shortfall (no fault, one of N+1 CRAHs out, CDU and feeder faults) nothing changes: max |Δ| 5e-14 °C.
One option was tried and rejected: centring the hump mid-aisle. It gave the same RMS, but with four CRAHs out it put 5 racks over 32 °C against the CFD's 13.
Fit against five CFD cases. Each cell gives the balance before → after, then the CFD.
Case
RMS per rack
Hottest inlet
Racks > 27 °C
Racks > 32 °C
4 CRAHs out (0.3 m)
4.39 → 1.72 K
41.8 → 34.0 · CFD 34.8
29 → 68 · 64
26 → 15 · 13
2 CRAHs out (0.3 m)
1.93 → 0.70 K
37.5 → 28.3 · 27.7
9 → 5 · 9
4 → 0 · 0
N units, 1 out, 100 MW (0.6 m)
1.48 → 0.83 K
36.7 → 28.2 · 27.7
4 → 2 · 5
2 → 0 · 0
N units, 1 out, 300 MW (0.6 m)
1.52 → 1.02 K
37.3 → 29.7 · 29.3
8 → 6 · 26
6 → 0 · 0
N units, 1 out, 37 MW (0.6 m)
2.38 → 0.98 K
37.2 → 27.6 · 28.4
4 → 4 · 11
2 → 0 · 0
Off-fit check against the 0.6 m CFD on plants the fit never saw. The new model is better on every case:
Plant
Fault
RMS per rack, march → new
70 % capture
2 CRAHs out
1.67 → 0.91 K
70 % capture
4 CRAHs out
3.21 → 1.49 K
37 MW hall
2 CRAHs out
4.53 → 2.13 K
37 MW hall
4 CRAHs out
7.97 → 6.81 K
The last case loses four of a small hall's CRAHs and runs to 56 °C, and neither model holds it well.
Still true:
An N-unit hall with one CRAH out puts racks over 27 °C, as the CFD does (W12r).
An N+1 hall holds every rack at or below 27 °C.
What moved for a reader:
On the plan and in the rack tooltips, a CRAH fault's hot racks now sit around the middle of the aisles.
They are spread over more racks and are less extreme: with four out, 34 °C at worst instead of 42 °C.
The plan's red dashed box keeps marking where the walls' supply runs out; its tooltip now says the exhaust is pulled back over the racks around it.
Gate W12cal, new, reads the five cases from tools/fixtures/dh01-cfd-reference.json. For each it requires:
a per-rack RMS within 0.25 K of the fit;
the hottest inlet within 1.5 K of the CFD;
racks over 27 °C whenever the CFD has them;
racks over 32 °C within 4 of the CFD's count.
It was proven red on the old march, on the share without its ramp, and on the mid-aisle hump.
Gate change. In tools/test-ltc-plant.mjs, P7's check "unmet air sits only between where the two supplies end" encoded the plug-flow assumption that the CFD disproved. It is replaced by each aisle's own balance: the supply its racks draw plus the air they pull back equals its demand, and that supply equals what its mouths deliver.
v3.21.12PATCH
"Yard": the campus blocks open their own inspector
Why. v3.21.9 drew the campus, but three of its blocks had no inspector of their own. The 150 kV GIS hall and the genset yard borrowed the 20 kV substation payload, which titled them "MV customer substation". The fuel farm could not be clicked at all.
Added (js/datahall-ai/hmi-payloads.js, datahallAI.html site plan):
site-gis — grid circuits A/B from the electrical scenario. Engine values: intake current and voltage, duty and installed circuits, circuit rating, line and transformer bays, GIS hall area. SF6 density is a simulated reading, declared.
site-genset-yard — generator pool state from the scenario. Engine values: installed, duty and standby units, unit rating, yard area, fuel autonomy. Jacket-water keep-warm is simulated, declared.
site-fuel-farm — engine values: stored diesel, autonomy and bunded area. Tank level is simulated, declared.
Dependencies link the three blocks: grid, then GIS, then main transformers; fuel farm, then genset yard, then the 20 kV boards.
Gates. hmi-payloads 6/6; inspector-runtime PASS; coverage --strict --modals CLEAN, including the three new inspector:site-* rows.
v3.21.11PATCH
"Second Opinion": DH01 CFD convergence history and a rack-by-rack CFD vs balance chart
Why. The panel gave one line of CFD against the balance solve, the hottest inlet and the count over 27 °C. It did not show where the two disagree or whether the solve had settled. Two charts now sit under the 3D stage.
Added (js/ltc-cfd-view.jshistorySvg / scatterSvg / scatterStats, wired in js/ltc-dh01-3d.js):
Convergence history. The hottest and mean rack inlet and the CRAH return, against simulated time. The balance solve's hottest inlet and the 27 °C limit are drawn as references. You can watch the solve settle, and at 0.3 m the eddies show as wobble on the hottest inlet. The history is kept to 400 points and redrawn at most twice a second.
Rack by rack. Each rack's time-averaged CFD inlet against the balance solve's, with a 1 : 1 diagonal and the 27 °C limit on both axes. The chart prints the RMS difference and the racks over 27 °C in both models, in the CFD only and in the balance only. Hover a dot for that rack's full tooltip.
Both charts are drawn at the container's width, so the labels stay readable on a phone.
What the chart shows (four CRAHs out, 0.6 m):
The two models differ by an RMS of 4.8 K per rack.
The balance solve marches each cold aisle as plug flow. Racks reached by supply get exactly 25 °C, and the ones past the end of the supply get up to 42 °C.
The CFD mixes more:
racks the balance calls clean read 25–33 °C;
racks the balance calls starved read 25–35 °C.
Where the heat lands differs. By aisle thirds (west · middle · east), the CFD's hot spot sits in the middle of each aisle: aisle 1 reads 26.9 · 31.6 · 26.1 °C. That is where the jets from the two fan walls meet and exhaust curls over the rack tops. The balance puts it at the west end, by the dead units: 34.2 · 32.2 · 25.2 °C.
The CFD spreads the shortfall over every aisle. Aisle 3 reads 26.8 °C in the CFD against 25.2 °C in the balance.
The mean inlet agrees: 27.45 °C in the CFD, 27.86 °C in the balance, so the heat balance is consistent and only the distribution differs.
With two CRAHs out, the RMS difference is 2.1 K. With one out, or with a CDU or CDU-feeder fault, the two agree within 0.1 K.
Smoothing the balance's supply coverage along each aisle brings the four-CRAH RMS from 4.8 down to 3.2 K, and no further, because it cannot move the hot spot to the aisle middle. A proper correction is a change to the balance model. It was not made in this release.
The CFD agrees with its own 0.3 m grid (v3.21.8).
Gate. W12a now also checks the charts:
one dot per rack, carrying the rack's index;
the printed RMS and the over-27 counts equal a hand computation;
three history lines, with no NaN.
It was proven red with a dot dropped and with the RMS divided by n + 1.
v3.21.10PATCH
"Air Management": DH01 CFD reads like a data-hall survey, with smoke, inlet profiles and RCI / RTI / SHI
Why. After v3.21.8 the owner asked for more: "improve more, super rich and immersive". The CFD showed the field, but not the numbers a data-hall air survey reports, and you could not drop smoke into the room.
Smoke. Click anywhere in the hall to release a puff of 600 particles: on the slice, a rack face or the floor (1 m up).
The solved velocity carries the puff (RK2). A small random walk (eddy diffusivity 0.01 m²/s) stands in for the eddies the grid cannot carry; the legend says so.
The puff grows and fades over about 30 s, tinted by the air it is in.
It vanishes when it is drawn into a rack or a coil.
Inlet by height. Each rack's front door is now three bands, bottom, middle and top third. Each band is coloured by the CFD inlet at that height, so exhaust curling over the rack tops shows as a red top band.
Rack tooltip. Hovering a rack adds:
the CFD inlet, time-averaged;
the inlet by height;
a note when the top runs more than 3 K hotter than the bottom.
Air-management indices on the panel, from the CFD rack inlets. Hover each for its formula, the inputs this solve fed it and its rating.
RCI HI / LO (Herrlin): the rack cooling index, on ASHRAE A1. Recommended is 18–27 °C and allowable is 15–32 °C. 100 % is ideal, ≥ 96 % good and ≤ 90 % poor.
RTI: the return temperature index, return − supply over the racks' airflow-weighted rise. Above 100 % means recirculation; below means bypass.
SHI / RHI (Sharma): the share of heat picked up before the inlets.
With four CRAHs out at 0.6 m: RCI HI 75 %, RTI 120 %, SHI 0.17.
Solver (js/ltc-cfd.js):
Each step now lists the moving energy faces once, with the upwind side resolved, and the sub-steps loop over that list. The result is bit-identical to v3.21.8 (max |ΔT| 0, max |Δu| 0 at 0.6 m and 0.3 m) and slightly faster.
report() adds inletMean, the time-averaged rack inlets, plus profile (each rack's inlet by height), idx (the indices) and nsub.
Tried and rejected: warm-starting the 0.3 m grid from a settled 0.6 m field. With four CRAHs out, 300 fine steps, rack inlets averaged over the last 200, compared with the 900-step 0.3 m reference:
Start
Hottest inlet
Racks > 27 °C
RMS vs reference
Cold
34.4 °C
64
0.35 K
Seeded from 0.6 m
33.4 °C
73
0.70 K
The seed carries the coarse grid's bias, and the fine grid takes longer to forget it than to settle from rest, so the code was removed. A cold 0.3 m start matches the reference within 0.35 K after about 60 s of simulated time.
RCI HI / LO, RTI and SHI equal their formulas on hand-made inlets;
on crah4, each rack's inlet profile averages back to its inlet within 0.05 K;
the solve reads as recirculation: RTI > 105 %, SHI > 0.05 and RCI HI < 96 %.
W12z now also checks that a smoke puff drifts with the flow, spreads, and never sits inside a solid.
Each new check was proven red on mutated code:
the RCI denominator changed;
the RTI rise offset;
the top inlet row dropped from the profile;
the smoke's solid check removed;
the smoke's advection removed.
v3.21.9PATCH
"Campus": the plant the engine sizes, on a site that can hold it
Why. The owner asked for the layout to be perfected. The building views were the last drawings still on the retired 72 × 48 m grid, and the engine's own counts show that grid could never have worked:
The engine sizes 631 dry coolers, 171 gensets of 4 MW, 143 chillers, 16 main transformers of 100 MVA and a 20-bay 150 kV switchyard.
The ground floor drew six gensets with a "171×" label, and the roof drew eight dry coolers.
At honest per-unit allowances, the dry coolers alone want about 4.6 roofs and the gensets about 2.2 ground floors.
Each hall's 2N electrical rooms need about 2,855 m², while the level plan gave them a 20 m schematic wing of 620 m².
The owner chose a campus site plan over captioning the shortfall.
Engine 1.5.0, new site branch (js/dcai-model.js, js/dcai-engine.js):
Per-unit floor allowances, each including service clearance and each ASSUMED:
Chiller 45 m² × plant factor 1.4, main transformer 300 m², GIS bay 30 m².
Tank farm 0.4 m²/m³, campus road factor 1.3.
Fuel: 24 h on site (ADOPTED) at 0.25 L/kWh (ASSUMED).
Outputs:
The electrical wing per hall: 2,855 m² of rooms, rounded up to 93 m.
The building plate: 163 × 68 m, 6 levels.
The roof share of the dry coolers (172) and the yard share (459).
The yard areas, the fuel volume (4,044 m³), the plant outside the building (50,361 m²) and the site area (79,879 m²).
Each output has an identity in tools/test-dcai-engine.mjs: 568 pass.
The registry has 315 parameters, R7 and R8 strict.
Room Layout:
Ground floor → site plan.
Eight numbered blocks to one scale: STP, GIS, main transformers, fuel farm, building, gensets, chiller plant and dry-cooler yard.
Every genset, chiller, dry cooler and main transformer is drawn as a unit. The main transformers carry the SLD inspector hooks sld-main-tx:dh0Nx.
The links between blocks are drawn: 20 kV, CHW/CDW, HTW and diesel.
Under the plan, a schedule gives every count and area from the engine, and a ground-floor inset at the hall-level scale shows the rooms inside the building.
Hall levels. The core, spine and electrical wing now come from the engine. Each wing holds unit substations and UPS + battery for feed A and feed B, plus the RPP/busway riser.
Roof. Four banks carry the 172 roof units over the two hall wings, with the HTW headers down the spine and the support AHU on the core.
Building view. It is redrawn as one exploded stack of opaque plates, each zoned like its own plan. The labels sit in one column, beside a facility card with the footprint, the yards and the site. It replaces 72 × 48 m translucent slabs carrying ~90 floating tags.
Gates:
test-dcai-roof-plan R1 now closes the roof share against the engine: roof + yard = installed.
test-datahall-ai-authority built its request-mismatch fixture from a hard-coded ?v=1.4.0. After the engine bump the replace matched nothing and the fixture silently became the healthy page. It now uses the pinned constant and fails if the page stops loading it.
test-cockpit-reveal existed but was never wired into the ship gate, so it went stale for five releases: it still expected four floor views after the building became six levels. It now counts the levels from the engine (site.building_levels) and is wired.
test-rz-diagram-engine extracts the placement block up to renderOverview. The retired slab and box helpers are deleted, along with the old floor-plan helpers (rm, eq, rmLive, eqLive, rmTx, stair) this change orphaned.
Honest boundary:
Every allowance is a planning allowance, not a vendor footprint.
The block arrangement, the 12 m roads, the STP footprint and the room zoning are page-authored and declared on the drawing.
The site area depends on the 1.3 road factor more than on any single allowance.
v3.21.8PATCH
"Smoke Test": DH01 CFD as a volume, with trails, vortices and a probe, and a grid-checked solver
Why. The owner looked at v3.21.6 and said the CFD was still ugly: "I can see the gradient, the flow... Where the turbulence? Where's the temperature? I need more accurate." The streaks read as hair, the heat showed only as sprites above 27 °C, and the turbulence was a slice option. This release shows the whole field in 3D and checks the solver against a finer grid.
What the view shows (new js/ltc-cfd-view.js):
Volume. The solved field is ray-marched on the GPU, one texel per cell, interpolated between cells. It has five modes:
Heat: room air above 27 °C glows. The contained hot aisles and plenum stay faint, so escaped hot air stands out. Fast supply air shows as a cool haze.
Isotherms: the 27 °C and 32 °C surfaces of the room air, lit by the temperature gradient.
Speed.
Turbulence intensity.
Vortices: |∇×u|, showing the shear layers and eddies the grid resolves.
A ray stops at a rack or CDU, so what is behind a rack stays hidden. The ceiling slab is see-through, like its glass.
Flow trails. 2,400 particles move with the solved velocity. They start at the fan walls, the rack exhausts and anywhere in the room, and each has a one-second tail coloured by the air it passed through. You can watch air leave a fan wall cold, cross a rack and reach the plenum hot.
Probe. Hover the slice to read the field at that point:
temperature, time-averaged and instantaneous, against the ASHRAE envelope;
speed and its three components;
turbulence intensity and vorticity;
position and cell size.
Slice. It can now show vorticity too.
Panel. It adds the room's median turbulence intensity and the 99th-percentile vorticity.
Other controls. Auto-orbit, and a colour scale for whatever the volume shows.
Phones. The stage is taller. The panel keeps its first line, and the rest sits under the stage.
Solver (js/ltc-cfd.js):
Energy is now second-order upwind with a van Leer limiter (TVD). It is still conservative, so the heat balance closes, and it creates no new extremes. Each cell's outflow per sub-step is held under half its volume.
Momentum uses clamped MacCormack (second order) on the 0.3 m grid and stays semi-Lagrangian on the 0.6 m grid. On 0.6 m cells, MacCormack over-mixed: with four CRAHs out, the hottest inlet came out at 31.9 °C, against 34.8 °C on 0.3 m. A rack row is only two cells deep at 0.6 m.
The energy faces are listed once per grid. With the old schemes selected, the result is bit-identical to v3.21.6.
Grid check: four west-wall CRAHs out, rack inlets averaged:
Solve
Hottest inlet
Racks > 27 °C
Racks > 32 °C
0.3 m, second-order schemes (reference)
34.8 °C
64
13
0.3 m, v3.21.6 schemes
35.9 °C
65
17
0.6 m, this release (default)
35.3 °C
68
16
0.6 m, v3.21.6 schemes
34.3 °C
70
14
0.6 m, MacCormack + TVD (rejected)
31.9 °C
82
0
With two CRAHs out, the 0.6 m default puts 9 racks over 27 °C. The 0.3 m reference and the balance solve both give 9; v3.21.6 gave 5. The 0.6 m solve runs at about 35 ms per step.
W12g: the 0.6 m solve must stay within 1 K, 6 racks over 27 °C and 4 racks over 32 °C of the recorded 0.3 m reference.
W12z: five checks on the view and the energy scheme:
the vorticity of a solid-body rotation is 2Ω;
the probe reads cell centres and interpolates between them;
the volume texels carry temperature and cell kind;
trails never enter a solid;
the energy scheme keeps the air inside the temperatures its sources carried.
Each new check was proven red on mutated code: MacCormack forced on 0.6 m, a non-TVD (Fromm) energy scheme, a sign flip in the vorticity, and trails allowed into racks.
Vendor. The three.js r170 bundle adds Data3DTexture, DataTexture, GLSL3 and the texture constants (533 KB, still lazy-loaded).
v3.21.7PATCH
"Fan Wall": the DC AI hall laid out like a real GB300 liquid hall
Why. The owner reviewed the hall plan and said the CDU and CRAH positions are "not like that in the real world", and that the arrangement made the hall a joke. They were right:
The plan put 90 CRAHs in six lines in the middle of the room and stood the CDUs in an east gallery.
Cold air had to travel 31 m down a cross aisle.
The engine's own comment called the air plant a "perimeter gallery" while the drawing put it mid-room.
The text above the plan was faint prose, the floor plans in Room Layout were the retired 72 × 48 m grid with "CRAH ×6" at 4 px, and the hover effects read as gimmicks.
The layout follows the GB300 reference designs (Vertiv 360AI RD0019: perimeter chilled-water fan walls and liquid-to-liquid CDUs; Schneider GB300 NVL72 hall). The owner chose it over keeping the old engine geometry. From north to south:
Fan-wall gallery N, 3 m. 45 CHW fan-wall modules discharge straight into the cold aisles.
Rack field. 20 rows of 22 racks run north–south at 3.1 m pitch, the full 62 m length. Hot aisles are contained and rise to the ceiling plenum, which returns to both fan walls.
Service corridor, 4.6 m. TCS supply and return headers run overhead, with the busway feeders, the leak rope and the egress doors at both ends.
CDU line, 3.9 m. The 55 CDUs stand at the row ends, one group under each containment pair. Their secondaries rise to the corridor headers and their primaries sit on the HTW header behind them.
Service aisle, 3.3 m. This is the floor spare, 226 m².
Fan-wall gallery S, 3 m. 45 modules, one of them standby.
Engine (js/dcai-model.js, js/dcai-engine.js):
geometry.banks changes from 2 to 1.
New leaves cduLineDepthM (3.9 m) and fanWallGalleries (2). crahServiceM2 is retired.
The air-plant floor is now galleries × length × depth = 372 m², and its frontage is the two long walls.
New outputs: fan_wall_modules_per_gallery (45), fan_wall_module_pitch_m (1.38), width_stack_m (27.7 of 31 m) and length_stack_m (62 of 62 m), each with a fits flag.
The fan-wall face check still holds: 535 m² needed against 682 m² available at 2.5 m/s.
New output fan_wall_face_velocity_ms publishes the 2.5 m/s face velocity, so the plan prints it from the engine instead of as a literal.
The registry is regenerated: 292 parameters, R7 and R8 strict.
The hall plan:
Three header cards give the IT load, the liquid path and the air path, each value hooked to the engine.
Under them, the floor budget term by term, and the width, length and fan-wall-face checks.
Every band of the plan is labelled down the west side.
Captions sit on solid backgrounds, so pipes run behind them, not through them.
Room Layout: Floors 2–5 now draw both halls of a level at true scale on a 90 × 68 m plate:
A west core with stairs, lift, risers and MDF.
A spine corridor between the wings.
An east electrical wing per hall: UPS feed A and B, and the RPP and busway riser.
The core, spine and wing are page-authored allowances and are declared as such. The wing is schematic and is not sized to the UPS frame count.
The level totals (IT, racks, CDU, fan-wall modules) are two halls × the engine's per-hall value. They are declared as such, not hooked to the per-hall parameter they double.
Interaction:
Hover a rack and its row, its contained hot aisle and that aisle's sensors stay lit.
Hover a CDU and the liquid path stays lit; hover a fan wall and the air path stays lit. Layer chips and key entries light their layer.
A clicked unit keeps a selection ring after its HMI closes.
A hall switch crossfades.
The cursor coordinate readout from the first cut was removed as a gimmick.
State is attributes only, so the drawing never refits on hover (js/datahall-ai/hall-interact.js).
Other changes:
Labels: the Go-to chips read "CDU line" and "Fan walls"; the floor-budget panel reads "Service corridor", "CDU line" and "Fan-wall galleries"; the CRAH HMI now supplies "the cold aisles".
Rack inspector: it no longer names a bank. Its CDU dependency is the first CDU at its own row pair, and its fire zone is spread across the four rack-row zones.
Fire zones Z01–Z04 are renamed "CDU line 1–4".
Gates:
tools/test-datahall-ai-hall-plan.mjs is rewritten: every CDU must stand in the CDU line, past the corridor, and every air unit must be a fan-wall module in the N or S gallery, split evenly. Proven RED on v3.21.6 (CDUs east of the field; 0 of 90 units in a gallery).
New tools/test-datahall-ai-hall-interact.mjs covers the interaction contract. Proven RED on v3.21.4 with 9 findings.
tools/test-dcai-engine.mjs gains the gallery, pitch and stack identities.
Still open: the ground floor, the roof and the building isometric remain on the retired 72 × 48 m grid, captioned on the drawing. (Closed in v3.21.9.)
v3.21.6PATCH
"Wind Tunnel": DH01 in 3D, with a CFD of its airflow
Why. The DH01 view was a plan and a mass-and-energy balance: it showed where a fault leaves each rack, not how the air gets there. The owner asked for a hall to turn around in, with the airflow, the heat, the turbulence and the water all moving, and accurate. The hub's DH01 tab now has a 3D · orbit view and, behind it, a real CFD solve of the hall.
The CFD (js/ltc-cfd.js, a background Web Worker):
Equations: incompressible, buoyant airflow on a staggered grid: semi-Lagrangian momentum, Boussinesq buoyancy, and a pressure projection solved by multigrid-preconditioned conjugate gradients, so mass is conserved in every cell. Energy is a conservative finite-volume upwind scheme, so the hall's heat balance closes.
Racks: black boxes, as DC CFD codes model IT. Each draws its solved airflow from the cold aisle and returns it at the drawn air's mean temperature plus its heat.
CRAH fan walls: supply below 2.6 m and return through the plenum wall and roof, at the solved unit flow. The coil holds set-point until its fitted effectiveness cannot.
Geometry: racks, CDUs, ceiling, hot-aisle chimneys and end doors are solid. A short row is closed with blanking panels. The containment leaks through a stated 5 cm slot between rack tops and chimney panels.
Grids: a 0.6 m grid settles in seconds. A 0.3 m grid is on offer and takes minutes.
No turbulence model: eddies the grid resolves are resolved, smaller ones are lost to numerical diffusion. This is for training and for comparison with the balance solve, not a design CFD.
What the view shows:
Streaks advected through the solved velocity: jets off the fan walls, the turn into each aisle, recirculation over the racks.
A slice at any height, or down a cold aisle. It can show temperature, speed or resolved turbulence intensity, interpolated between cells on the GPU.
Glowing volumes of air above 27 °C outside the contained hot aisles.
Rack doors coloured by the CFD inlet and exhaust.
The TCS, FWS and CHW pipes with their flows.
A panel with the heat in versus heat removed, the mass residual, and the hottest inlet and racks over 27 °C, each against the balance solve.
Camera presets: orbit, top, cold-aisle walk-through, hot aisle, plenum. Every rack and unit carries the plan's tooltip.
Default 100 MW plant, CFD vs balance:
Case
CFD
Balance
Normal operation
hottest inlet 24.0 °C
24.0 °C
Two CRAHs lost
9 racks over 27 °C
9 racks over 27 °C
Four CRAHs lost
71 racks over 27 °C, hottest 34 °C
29 racks over 27 °C, hottest 41.9 °C
With four CRAHs lost, the CFD spreads the recirculation over more racks while supply and return agree within 0.4 K. Heat removed equals rack heat within 0.1 % in every case.
Found on the way:
The last row holds 12 of 20 racks, which left its hot aisle open across the empty slots. The first CFD poured the whole air shortfall through that gap, giving a 56 °C hotspot. It is now blanked, as on site, and gated.
A plain SOR projection on the 0.3 m grid left 14 m/s spurious jets. It was replaced by MGPCG.
W12x: the 3D balance streams equal the solve. That covers supply per aisle mouth and wall, recirculation, every rack's inlet and exhaust, TCS per CDU, and FWS and CHW stopping with their fault, for 8 plants × every fault × both headers.
W12y: the CFD's containment is closed except for its declared openings. Heat balance is within 1 %, mass residual is below 5e-3, and supply and return are within 0.5 K and 1 K of the balance. Normal operation stays inside 27 °C. A west-wall fault heats the west half and puts racks over 27 °C when the balance does.
Mutations fail them: the wall split fixed at ½, rack heat at 90 % ("heat removed 3,516 of 3,907 kW"), and the blanking removed ("40 open faces, 16 declared").
three.js r170 (MIT) is vendored as a tree-shaken bundle (js/vendor/three-r170.min.js, tools/three-vendor-entry.js). It is loaded only when 3D is first opened.
v3.21.5PATCH
"Alarm List": DH01's sequence of events, and a replay of the fault
Why. The DH01 view showed where a fault leaves the hall, not how it gets there. An operator meets a fault as an alarm list: what tripped, what crossed which threshold, and when. The workbench now writes that list for every fault, and every time on it is solved, not scripted.
Sequence of events. A new list sits under the DH01 plan on the hub and in the report. Each entry has a time (T+mm:ss, or steady once fans and valves settle), a severity (ok / info / warn / alarm / trip) and the event.
CDU feeder, one CDU: the loss at T+0. If cold-plate flow is short, the racks throttle at T+00:10 with the IT shed from the hall solve; otherwise the CDUs left carry the duty. The steady air consequences follow.
CRAH faults: fan speed of the units left, coil set-point, exhaust pulled back into the cold aisles, racks above 27 °C / 32 °C.
Facility water lost: TCS return halfway to the throttle, then the throttle itself. Both are read from the loop transient.
Chiller trip, chilled water lost: the hall is re-solved along the CHW transient and each threshold is bisected to the second: coils lose set-point, first rack above 27 °C, first above 32 °C, then the state at 30 min.
At full load, the default 100 MW plant with chilled water lost reads: coils lose set-point at T+00:22, first rack above 27 °C at T+03:26, first above 32 °C at T+08:39. Facility water lost throttles the GPUs at T+03:40.
Replay. Facility water, chiller and chilled-water faults get a ▶ Replay button. It runs the fault from T+0 in half-minute steps, and the plan, the heat chain, the transient chart and the alarm list move together. Events still ahead are dimmed, and the transient chart marks each event on its time axis. Reduced motion jumps straight to the end.
W12v: 12 plants × every fault. Each timed threshold is crossed at its stated second and not one second earlier. The FWS throttle is the transient's own crossing, the feeder trip sheds what the hall solve sheds, and the list runs in time order with steady state last. Mutations fail it: bisecting to 30 s instead of 1 s ("coils lose not crossed"), and the FWS throttle moved 20 s ("throttle at 240, transient 220").
W12w (per plant): the report carries every event of every fault. Dropping the section fails it on all 8 plants.
v3.21.4PATCH
"As Built": the Tier this plant reaches, and what blocks the next one
Why. The Tier sweep builds each Tier's own topology on the plant's unit counts. So the report could say "Tier III — concurrently maintainable" about a plant whose CDUs sit on single feeders: maintaining feeder A drops half of them. The workbench now also answers the other question: what does the plant, as configured on the strip, reach?
Tier page — new tab This plant. Each requirement has a status:
met: the plant inputs satisfy it. Failures are read from the engine's own fault model and the DH01 solver.
gap: they do not, with the number that shows it.
verify: the plant model does not describe it.
The requirements per Tier:
Tier II: N+1 generators, transformers, UPS modules, CDUs, liquid coolers, CRAHs and chillers; pumps are verify.
Tier III: a second distribution path (met with UPS 2N, otherwise verify); CDU power maintainable with one feeder off; MCC A/B each carrying the mechanical load (verify); a second, valved FWS and CHW path (verify).
Tier IV: two active power paths; a CDU feeder fault; one CDU, one CRAH (DH01 solver) and one chiller failing; continuous cooling (CDU pumps on UPS); compartmentalisation (verify).
The plant reaches the highest Tier with no gap at or below it. The blockers for the next Tier are named.
Default 100 MW plant reaches Tier II. Tier III is blocked by the CDU feed: 5 of 10 CDUs lose power and 7,142 kW of 17,142 kW liquid duty is uncovered. Changing the CDU feed:
Dual feed: Tier III.
Dual feed plus UPS 2N: Tier IV.
Without CDU pumps on UPS: back to Tier III.
Report. The Tier topology section gains "This plant, as configured, against each Tier", and the key findings state the Tier reached and what the next one needs. The Tier page now loads the DH01 solver, so "One CRAH fails" reads the hall's max rack inlet.
This is a screening against the Tier objectives, not a Tier Certification.
W12t: 64 configurations (presets × CDU feed × UPS N+1/2N × CDU pumps on UPS × CRAH N/N+1). Each reaches the Tier worked out independently in the test, and each one short of Tier IV names its blocker. A mutation that marks the CDU feed always met fails it ("Tier 3, want 2").
W12u (per plant): the report lists every requirement.
v3.21.3PATCH
Eight halls everywhere the page names a hall: switcher, building, floor plans, single-lines
The engine has run 8 halls × 440 racks since 2026-09-09 (four halls of 880 did not fit their own floor or pass their own air). The page still offered DH-1..DH-4, drew a 4-storey building with two hall floors, and had four per-hall single-lines. The owner asked for the best option, so the page now follows the engine everywhere it names a hall.
Fixed — one source for where a hall sits.window.HALL_LAYOUT holds the building's stacking rule in one place: two halls per level, North wing then South wing, from Floor 2 up. The rule is ADOPTED and declared on the drawing. It is applied to facility.halls, so 8 halls = Floors 2–5. Everything below reads it, and none of it types a hall list or a floor name any more.
Hall switcher: DH-1..DH-8, one button per engine hall. The sidebar location comes from the rule. It now says DH-1 is in the North wing, as every floor plan has always drawn it; the sidebar used to say South.
Building elevation: one slab per hall level, so 6 levels (ground + 4 hall levels + roof).
Floor plans: renderHallLevel draws any hall level. It replaces renderFloor2 and renderFloor3, the same drawing typed twice. Floors 4 and 5 had no plan at all. UPS and battery room letters follow the hall (A/B serve DH-1/DH-2, … G/H serve DH-7/DH-8).
Electrical: eight per-hall SLD tabs, cloned from the DH-01 template. Each sheet is drawn the first time its tab opens, instead of eight 2,400-unit drawings built up front. The facility overview carries 16 feeder breakers (8 per bus), 8 hall columns in two rows of four, and 8 KPI cards. Clicking a column opens that hall's sheet.
The Tech Spec and the BMS tag note print the engine hall count where they said "4 halls".
Fixed — halls 5–8 were read as hall 1.electrical-visual-map.js matched scopes and line ids on [1-4], so dh05..dh08 fell back to dh01. The patterns now accept halls 1–9 (module 2.1.0). test-datahall-ai-electrical-visual-map.mjs asserts that dh05..dh08 are halls; the assertion fails on the old patterns.
Gates: the cockpit tab set gains floor plans for levels 4–5 and the DH-05..DH-08 single-lines, so geometry, legibility and coverage now walk them.
Retracted from v3.21.0: "CRAH airflow 3.3 % above heat / (ρ·cp·ΔT)" was my error. The engine computes crah_air_m3s with ρ = 1.16 kg/m³ (air at the ~30 °C return class). My check used the 1.2 kg/m³ standard-air leaf that the fan-wall sizing uses at the supply plane. Both are declared, and the engine is consistent with itself.
Open, not changed here: the building footprint is still the retired 72 × 48 m grid, captioned on the drawing. Two 62 × 31 m halls on one level need at least 62 × 62 m before any corridor. The site re-grid (and the ground floor and roof plans that sit on that grid) is its own change.
v3.21.2PATCH
DH01 at every heat-load-test step; one CRAH out as a fault you can pick
DH01 thermal view — hall load. New control Hall load · HLT step: 25 / 50 / 75 / 100 % of design. It sets the load each fault is tried at, the same steps as the heat load test.
What scales with load:
Rack heat and non-IT air heat.
The CDU shortfall. The survivors' capacity is fixed, so at part load less duty is missing.
The FWS and chilled-water ride-through. The rejection left is a share of design capacity, so it covers a part load more easily.
Default 100 MW plant:
CDU feeder A lost: 8.3 MW IT throttles at 100 %, 3.3 MW at 75 %, nothing at 50 %. The note says so: "At 50 % hall load the CDUs left carry all of it."
Facility water lost: 220 s to the TCS throttle point at 100 %, 877 s at 25 %.
Where it shows: the header shows "10.0 MW of 20.0 (50 %)". The TCS chip and CDU hover show the part-load return (constant flow).
One CRAH lost. Added to the DH01 fault list beside 2 and 4 CRAHs. On an N+1 hall every rack stays at 24 °C; with N units the same fault reaches 37.4 °C.
Heat load test plan. The failure table gains Holds up to: the highest HLT step at which each single failure still holds. Default plant: CDU feeder A holds to 50 %. The test team knows the step at which the feeder test will throttle racks. The hub's L4 inspector says the same.
Engine.LTCPlant.transient(r, kind, { load }): heat in scales with load; rejection left is divided by it. LTCDH01.solveHall(…, load); LTCDH01.LOADS.
CDU feeder A holds exactly to the step the surviving CDUs can carry (checked against the engine's own shortfall).
FWS ride-through at 25 % is 4× full load.
Proven red: a mutation that keeps the design shortfall at part load fails W12s on 3 plants ("feeder holds to 0 %, survivors carry 50 %").
v3.21.1PATCH
"Proven at Load": the heat load test written for this plant; one CRAH out no longer overheats an N+1 hall
Report and hub — heat load test plan. The L4 heat load test (HLT) is now written from the plant on the strip, not as generic text:
Load steps (25 / 50 / 75 / 100 %). Each step lists DH01's liquid and air load banks, split by capture (100 MW default: 17,000 + 3,000 kW), plus kW per duty CDU. It gives TCS flow held at design (407.3 L/s, 45.3 per CDU) and the expected TCS ΔT and return (2.5 to 10 K, 37.5 to 45 °C). The accept band is the measured ṁ·cp·ΔT at ±5 % of the liquid banks.
Failures at 100 %. One CDU, one CRAH and CDU feeder A are each solved on the DH01 model, showing max rack inlet, racks throttled and IT throttled. The default plant holds the first two and fails the feeder (8,333 kW throttles on a single feed). The report says so before the test does.
Where it appears. The report gains section "Heat load test plan". The hub's L4 inspector shows the same numbers. The DH01 solver accepts crah1 (one unit) beside crah2 / crah4.
Fixed — DH01 air plenum wasted spare air. Each CRAH wall split its air by aisle demand, discounted near a dead unit. What an aisle could not take was bypassed, never offered to the aisle that was short. With one CRAH out of an N+1 hall (280.8 m³/s for 274.2 needed), the aisle facing the dead unit starved: 3 racks over 27 °C, max inlet 36.4 °C. The plenum now re-offers spare air to the aisles still short, from either wall, until the air or the need runs out. Air pulled back out of the hot aisle now equals the hall's real shortfall: 2 CRAHs out gives 11.0 m³/s for a 10.9 shortfall, was 16.9.
One CRAH out of N+1 holds every rack at 24 °C.
With N CRAHs, one out still fails, at 37.4 °C.
The DH01 thermal view and the report's fault plans read the corrected model: 2 CRAHs lost now peaks at 37.5 °C (was 38.0).
W12q (per plant): load banks split the hall load by capture; ΔT scales with load and reaches the design ΔT at 100 %; each single failure holds exactly when the design has that redundancy (CDU N+1, CRAH N+1, dual CDU feed).
W12r: over 24 plants (4 presets × 37/100/300 MW × 70/85 % capture), recirculated air equals the airflow shortfall within ±1 %. N+1 CRAHs keep every rack at or below 27 °C with one out; N CRAHs do not.
Both gates were proven red on the v3.21.0 allocation: 9 failures, including "2 racks over 27 °C with N+1 CRAHs".
v3.21.0MINOR
The data hall drawn as the room it is: whole hall, one scale, both heat paths, honest modals
Owner, on the phone: "Jelek sekali dan nggak sesuai aneh cdunya mana, crah utk residual air gimana. Designnya sangat kurang immersive", then "Harus super immersive, intuitive dan keren tampilannya. Modal2 saat click harus tetap ada dan improved accurate". Layout chosen by the owner: follow the engine.
Fixed — the plan drew half the room, in the wrong places. The engine budgets the 62 × 31 m hall as two 13.2 m banks plus the 4.6 m cross aisle (the full 31 m width) and ten 3.1 m rows (31 m of the 62 m length). The other 818 m² holds the CDU gallery (220 m²), the air plant (450 m²) and the spare (148 m²). The old plan drew only the rack field, put the CRAH units on the walls the rows already touch, and put the CDUs in the egress aisle. Now #hSvg draws the whole room at 17 px/m on both axes:
the rack field (440 racks, cyan front face and amber rear face, contained hot aisles with doors);
the cross aisle, with TCS headers for both banks, the leak rope and the cold supply;
the CDU gallery: 55 × CHx1000 at their plan footprint, with an HX and pump symbol, standby dashed, HTW entering from the east wall;
the air plant: 90 CRAH with spinning fans on duty units, coils on CHW;
the spare floor.
Fixed — it could not be read at its own size. Labels of 2.5–6.5 units forced a 2,700 px render, so a 1440 px screen showed a third of the hall and a phone a sliver. Every label is now ≥ 9 units in a 1200 viewBox. The whole hall fits a laptop at 100 %, and a phone pans a 1,240 px drawing.
Added — both heat paths, end to end, and the controls to read them.
The header states LIQUID kWth → 55 CDU and RESIDUAL AIR kWth → 90 CRAH from the engine, plus each path written out: cold plates → manifold → TCS → CDU HX → HTW → dry coolers, and CRAH → cross aisle → cold aisles → hot aisle → ceiling plenum → CRAH coil.
The heat map colours every aisle segment from its own sensor.
Layer chips (Thermal, Airflow, Liquid, Leak, Overhead) and "Go to" chips (Rack field, CDU gallery, Air plant) are added; the Go to chips matter most on a phone.
Fixed — the modals said things the engine does not.
CRAH HMI: duty is the engine per-unit duty (199 kW, was 114 from a simulated CHW flow); CHW flow is derived from it; return air is the engine plane 36 °C (was 31 °C); "raised floor" became the cross aisle (the hall is slab, no raised floor); "residual 15 %" became the computed 26 % of hall IT.
CDU HMI: a CoolIT CHx1000 has no fan, so the "Fan Speed" tile is now pump speed. A standby CDU says STANDBY, not "Online" at full flow.
Payloads (inspector + HMI): standby CDU and CRAH units carry zero flow, duty and fan or pump speed, with the reason declared.
Rack manifold HMI: TCS 35/45 °C literals now read the engine planes (40/50 °C).
Hover: CDU, CRAH and HTW tooltips are built from the same payload as the HMI. They printed GB200 planes (TCS 35/45, FWS 12/22, CRAH return 28.5–30 °C). The rack tooltip printed 182.05128205128204 L/min.
Gates
tools/test-datahall-ai-hall-plan.mjs rewritten to the engine layout: CDUs in the gallery east of the rack field and clear of the cross aisle; one CRAH glyph per installed unit, east of the CDUs; rack glyphs at the engine's depth/width ratio. It now reads geometry from SVG attributes: it never enters the authorized audit state, so the hidden panel measured every box as 0 and the old "CDU in the cross aisle" check was vacuously true. Fails on v3.20.3 with 3 findings, passes now.
test-dcai-coverage.mjs --strict --modals: hSvg 34/34 numerals hooked; every hall inspector and modal row CLEAN.
The aisle sensor readouts are rotated to fit a 1.2 m aisle, so a reading's on-screen height is its width. A bare "—" placeholder measured 7–8 px and failed test-conv-geometry.mjs (G5) before the first tick. The placeholder now carries its unit ("— °C").
Found, not changed here: the engine runs 8 halls, while the building drawing shows 4 on two floors and the hall switcher offers DH-1 to DH-4. design.flows.crah_air_m3s carries 3.3 % over heat ÷ (ρ·cp·ΔT); the CRAH modal prints the engine plane rather than a third number.
v3.20.4PATCH
Plant map: options you can apply, a live sensitivity tornado; cooler label fits on phones at 100+ units
Hub — options on the plant map. The "same site, same load, four plants" table now holds three things, each row a full engine solve:
The four configurations, with design and annual PUE, WUE, water, trim hours, energy + water cost a year and the difference against this plant. A Use button puts a configuration on the shared plant, and every workbench page follows.
The same plant in each of the three climates, each with its own Use.
A tornado of annual PUE for seven design inputs moved one step each way, widest swing on top. Selecting a bar end puts that value on the plant.
The numbers come from the same LTCReport.options the exported report prints.
Fixed — cooler block overflowed on phones. At a cold TCS supply (28–32 °C) a 100–300 MW plant needs 100+ liquid coolers, and the sublabel "346 units · 30.6 °C out" no longer fitted the phone layout of the overview and liquid views. A sweep of 2,880 diagrams found 32 such renders. From 100 units the leaving temperature is shown in whole degrees: "346 units · 31 °C out". The model lab's overview gets the same change.
Gates — tools/test-ltc-workbench.mjs:
W13 now draws 1,320 diagrams over 132 plants. The new plants are configurations × TCS supply 28/32/45 °C × climates, at 100 and 300 MW: the plants the options and tornado can put on the map.
On the v3.20.3 hub, W13 fails with the overflow.
v3.20.3PATCH
Search works below the site root; the 404 page works at every depth
Fixed — search found nothing on all 40 manual pages.js/rz-command-palette.js fetched search-index.json relative to the page, so on /manual/*.html it requested /manual/search-index.json, which does not exist (404 on the live site). The palette now resolves every fetch, result link, section link, command link and preview image against its own script URL (at()), so it works at any depth.
Fixed — the 404 page broke on any nested path. GitHub Pages serves 404.html for a missing URL at any depth. Its URLs were page-relative, so on /manual/x.html the illustration, the favicon and three scripts all 404'd, and every "popular page" link pointed at another missing page. Every URL on the page is now root-absolute.
Fixed — the search icon filled the modal. The palette's own injected markup, and articles 19–28, wrapped the input in .search-input-wrapper; the stylesheets style .search-input-wrap. The magnifier was drawn unsized: 359 px wide on a 390 px phone, on top of the input. Fixed in the palette, in the ten articles, and in standarization/article prompt/ARTICLE_CREATION_PROMPT.md, which was copying the bug into every new article.
Changed — 404 on the design system, with search (UIUX audit 2026-02-17, "404 Page Lacks Search Functionality"). The page gains a search button (opens the shared palette, "/" works too) beside Return to Base Camp. The aurora blur, gradient text, film grain, glowing pill button, hover lift and glassmorphism links are replaced with a flat surface, solid type, 4px controls, 1px hairlines and named 160 ms transitions. The illustration and the copy are kept.
Gates
New tools/test-site-root-paths.mjs (ship gate): P1 every URL in 404.html is absolute; P2/P3 every palette fetch and every link it writes goes through at(); P4 no search modal uses .search-input-wrapper. Fails on v3.20.2 with 29 findings (P1 ×11, P2 ×3, P3 ×4, P4 ×11).
tools/test-asset-cache-tokens.mjs now resolves root-absolute /js/… paths against the site root. Before, it skipped them, so tokens on 404.html were never checked. Proven by a bogus token on 404.html, which the gate now fails.
Verified headless on /manual/does-not-exist.html at 390 and 1280 px: 0 failed requests, all links resolve, a search for "chiller" returns results whose first link returns 200; the palette icon is 20 px on article-1, article-23, manual/capex and 404.
v3.20.2PATCH
Plant report compares the options; air-cooled trim chillers rated as air-cooled
Report — Options & sensitivity. New section: the same IT load re-solved under each of the four plant configurations and in each of the three climates, and a tornado of annual PUE for seven design inputs moved one practical step each way: liquid capture ±5 pts, TCS supply ±3 °C, utilisation ±0.1, CHW supply ±2 °C, chiller COP ±1, CRAH ΔT ±2 K, UPS efficiency ±1 pt. Each row shows design and annual PUE, WUE, facility MWh, water, energy + water cost, the cost difference against this plant, and tCO₂. Key findings name the cheapest configuration and the input that moves PUE most. At the 100 MW default, adiabatic coolers would run about 1.03 M a year cheaper; TCS supply moves annual PUE most, from 1.215 to 1.176.
Fixed — air-cooled trim chiller rated like a water-cooled one. The main chiller's rated COP is capped at 5.5 when air-cooled; the trim chiller kept the water-cooled 7. So the air-cooled plant came out more efficient on the design day (PUE 1.290) than the water-cooled one with towers (1.298). The trim chiller now takes the same cap. Air-cooled design PUE: 1.323. Water-cooled plants are unchanged.
Fixed — climate comparison. A re-solve that changed only the climate kept the current plant's bins and design conditions, so every climate gave the same answer. The report's re-solve now takes the climate's own values.
W12o: on 8 plants, every configuration and climate is re-solved; this plant's row equals the result; each climate differs; each step brackets the input.
W12p: more capture, a warmer TCS, a better UPS and a better chiller each lower annual PUE; the air-cooled plant draws more on the design day than the water-cooled one and uses no water.
Both fail on the v3.20.1 tree (9 assertions).
v3.20.1PATCH
Plant report draws the plant and tests every Tier; the plant map holds on every plant size
Report — plant schematics. The exported report now opens with the four plant-map views (overview, power, liquid heat, air & rooms), drawn by the hub's own builder on the reader's plant. Each drawing is written into the file with its colours and type resolved under the light theme, so it reads correctly in a standalone file and when exported from a page in dark mode. Wide views scroll sideways instead of shrinking their labels below 80 % of their drawn size.
Report — Tier topology. New section: the Tier sweep for Tiers I–IV, power and cooling, on this plant's unit counts. Every element is taken out once, planned and unplanned. The table shows whether each Tier is concurrently maintainable and fault tolerant, and what breaks it, by name ("UPS A — the whole board → IT power lost"). The Tier page keeps the system it was showing. The Tier and hub models now load on any workbench page for the report; their page boot runs only where their figure exists. Sections number themselves, and the contents list is built from them.
Fixed — plant map labels missing on most plant sizes. The report's first sweep of the hub drew every view on 60 plants (5–300 MW, 50–98 % capture, four presets), desktop and phone. It found engine warnings, each logged as a console error on the live page, that no gate looked at (every other gate draws the 100 MW default):
On the overview and liquid views, the liquid-heat arrow label was unplaced on 102 of 240 plant/view pairs. A label longer than "85 MW" did not fit the 40 px column gap.
On phones, the "Engine generators" block overflowed on the power view at every plant size, the default included.
On phones, the cooler block overflowed on the hub and in the lab overview; and on desktop, at 300 MW, the cooling-tower block did.
Fixes:
The column gap is 48 px.
Arrow labels read whole MW from 10 MW up.
The generator, cooler and tower sublabels are shorter: "MV bus · 391 MW", "65 units · 39 °C out", "256 MW · 120 L/s make-up".
W13: 600 diagrams over 60 plants (four hub views and the lab overview, each at desktop and phone width), zero engine warnings. On the v3.19.0 tree it fails with eight kinds of defect.
W12m: the report carries the Tier sweep, each Tier reaching its own objective.
W12n: every plant-map view renders for the report on the result it is given.
v3.20.0MINOR
Every article block diagram drawn by the engine, with IEC symbols on the equipment
Twelve article figures were already engine-built. Seven articles still drew theirs by hand, and every gate was green, because build-article-diagrams.mjs --check only sees the figures it owns. Those hand-drawn diagrams were:
rows of <span class="flow-box"> joined by "→" (article-6 ×2, article-13 ×4);
a fixed-coordinate SVG pipeline (article-28);
three hand-drawn living diagrams (article-9, -13, -25);
a scrollytelling canvas (article-23).
All eleven are now drawn through the RZ diagram engine (tools/article-diagrams-converted.mjs), with the same boxes, words, numbers and connections. Equipment carries its IEC 60617 / ISA-5.1 glyph from js/rz-eng-symbols.js:
utility, switchgear and transformer;
UPS, PDU, battery and rack;
dry cooler, CDU, CRAH and cooling tower;
generator.
Concepts get no glyph. A control plane, a pipeline stage and a coal-gas-nuclear fleet are not one piece of equipment.
Living diagrams keep their scenarios. The three living diagrams keep their buttons, ticking values and fault injections. Each configuration is copied verbatim into the figure definition. The runtime now reads the engine's own hooks: a connector is data-rzd-edge with a bead overlay, a node is data-rzd-node, and a value is text[data-pv]. A scenario alarm on a reading's key now colours that reading, because the engine draws no separate instrument ring.
Article-23 scrolls by stage. The Colossus canvas still reveals turbines and battery, then rack rows, then go-live as the reader scrolls, through a new ctx.layer(n). The DAY / MW / GPUS counters are engine node values that js/rz-scrolly.js animates.
Two corrections the redraw forced, both stated in the captions:
AWS in-rack BBU. The original put the battery in series between the 48 V busbar and the load. A battery backup unit sits on the busbar in parallel with the load, and that is what is drawn now.
Article-13's 2N UPS pair. It sits in the section on AWS's distributed UPS. Its caption now says it is the conventional centralised baseline, which the AWS architecture below replaces with a battery in every rack.
Figures that already used the engine gained glyphs where the node is equipment:
cooling tower on the article-10 water loop;
CRAH and CDU on the article-18 heat paths.
Engine (js/rz-diagram.js), additive and byte-identical when unused:
valueKey / valuePrefix on nodes and edges;
ctx.layer(n) for reveal layers.
The incident figures (43 dossiers) and the existing article figures rebuild unchanged.
Gate.tools/test-article-diagrams-engine.mjs (ship gate) lists every article from the filesystem and fails on:
hand-built flow-box markup;
any static SVG with 4+ labels outside an engine figure;
any living diagram not drawn by the engine;
any empty figure placeholder.
On v3.19.0 it reports 10 findings; now 0. The article-16 scrollytelling bar chart is exempt by name with its reason, because a chart has no blocks or connectors. Orphaned .flow-box / .architecture-diagram CSS was removed from article-6 and -13.
v3.19.1PATCH
Login dialog on the design system, plaintext admin accounts retired, CDN scripts pinned by hash
Three findings from the February security and UI/UX audits were still open. Each was re-checked against today's code before it was fixed.
Login button, dropdown and dialog (154 pages).auth.js injects its own stylesheet from a JS string array, and tools/audit-vibecode.mjs reads static CSS only — it says so in its own output. So the one dialog on almost every page kept 16 px radii, two-stop gradients on the button, the avatar and the PRO badge, a hover lift with a glow, resting drop shadows, a blurred dropdown and transition: all. Now: 4 px radius, 1 px hairline with a 2 px mint top rail on the dialog, flat mint fills, no resting shadow, a 1 px outline for input focus, named-property transitions at 160 ms. The close button is 44 px and the inputs are 16 px so iOS does not zoom. Avatar text was white on mint (fails contrast); it is now dark.
Manual accounts kept the password in plain text. The admin page's legacy "Create Account" wrote {email, password, …} to localStorage.rz_manual_accounts, and auth.js accepted those passwords at login. The account only ever worked in the admin's own browser, so the feature never did what it claimed and left a readable password behind. Login no longer reads that key, both auth.js and the admin page delete it on load, and the admin "Create Account" button now opens the Supabase form (hashed, server-side). The demo account (demo@, published on purpose) is unchanged.
Subresource Integrity. 0 of the CDN script and stylesheet tags carried integrity. Every version-pinned one now does — 171 tags on 79 pages, plus the incident hub's Leaflet (tools/build-incidents.py) and the command palette's lazy Fuse.js load. Each hash was fetched twice and was stable, and Leaflet's matches the SHA-256 Leaflet publishes. A CDN that serves different bytes for that version is now refused instead of run. Not pinnable, so not hashed: Google Fonts (CSS varies by browser), cdn.tailwindcss.com, an unversioned chart.js, and gtag. ltc-system-modelling-lab.html belongs to another session's in-flight work and is left for it.
Gate.tools/test-auth-surface.mjs (ship gate) evaluates the same stylesheet array the browser gets, hover states included: radius ≤ 4 px, no gradient, no box-shadow, no hover transform, no transition: all, blur only on the dim layer; and no read or write of rz_manual_accounts. On v3.19.0 it reports 36 findings; now 0.
Still open, and not fixable on this host: no CSP, HSTS or Permissions-Policy header (GitHub Pages sets none and allows none), and client-side session and Pro gating (a static site has no server to enforce them; the client role was never an authorization boundary).
v3.19.0MINOR
LTC Workbench: export the plant as an engineering report
Export report — a new button on the plant strip, on all six workbench pages (js/ltc-report.js). It opens a standalone engineering report of the plant in a new tab. Every number in it comes from the same LTCPlant result the pages draw. It has 16 sections:
Overview: cover KPIs and key findings written from the numbers (trim-chiller hours, worst DH01 fault, CDU-feeder exposure, facility-water ride-through, heat-balance closure).
Heat balance: a Sankey diagram whose two sides sum to the same kW, and a facility-power breakdown whose items sum to the engine's facility total. The report prints both checks.
Metering: the M0→M3 cascade with PUE categories 1, 2 and 3.
Plant detail: liquid path, air path, and electrical sizing.
Redundancy register: need, installed, label (N+1 or 2N), capacity, margin, and whether the plant still carries the load with one unit out.
Annual performance: PUE per climate bin, column width ∝ hours, bins where the trim chillers run in red.
DH01 in plan under all eight faults: racks shaded by inlet temperature, lost CDUs and CRAHs marked, and a fault table.
Transients: four curves, each with its limit line and the time to reach it.
Reference: a standards cross-check (ASHRAE, ISO/IEC 30134, Uptime capacity, NFPA, TIA), every calculation trace with its formula and inputs, and every engine input with its name, unit, default and auto / yours state, plus assumptions & limits.
The DH01 solver loads on demand on pages that do not carry it.
Data inside the file. The full engine result is embedded in the report. Its toolbar downloads it as JSON or CSV (every scalar, one row each), saves the report as HTML, or prints it to PDF (print stylesheet). On a phone, wide charts scroll sideways instead of shrinking their labels to 5 px.
Fixed — ISO readout spilling out of its box. On the ISO page at desktop width, "0.856 kgCO₂/kWh" ran 10 px past its readout cell, because the readout never wrapped. The number still never breaks; a long unit now drops under it.
Gate — tools/test-ltc-workbench.mjs W12 (+127 assertions, 274 total). On eight plants (default, 37 MW, dual CDU feed, 70 % capture at 40 °C, and the four presets), it checks that:
the Sankey closes, and the power items sum to the facility total, both within 0.5 kW;
the report states the engine's PUE (design, annual, category 1), M0, CUE and trim hours, plus every bin PUE, every trace, every input, every fault and every transient;
no NaN, undefined or [object Object] appears in the rendered report;
the embedded data is the engine result, and the CSV carries every scalar.
The first render printed "[object Object]" for the per-cooler-type inputs; W12e now fails that.
v3.18.5PATCH
Plant map and model lab on a phone: two lanes, and the lab diagram no longer runs off-screen
Model lab, broken on phones. A CSS floor (min-width: 480px on .ltc-schematic svg) made the overview 480 px wide on a 390 px screen. The block borders on the right and every connector routed down the right-hand side channel ran past the screen edge, cut off, unless the reader scrolled sideways. The geometry gates passed, because the geometry was correct; the drawing was simply off-screen. The floor is removed.
Two-lane phone layout. Phones now get the two-lane layout the Tier diagrams use (v3.18.4), on both the lab overview and the hub's four plant views (overview, power, liquid heat, air & rooms):
Power or liquid runs on the left lane, air on the right.
Blocks alone in their row span both lanes, so long sublabels fit.
Connectors run straight down or straight across. One side channel is left (electrical rooms → chillers on the overview). Before, these views stacked every block in one column, with skip connectors in 3–4 nested channels.
Gates — tools/test-rzd-ports.mjs adds two rules:
R7: a phone view shows the same connectors as the desktop view, so a narrow layout cannot silently drop a flow. A mutant missing cdu>reject fails it.
R8: at 390 px no diagram is wider than the screen. It fails the v3.18.4 tree ("lab: a diagram 480 px wide on a 390 px screen").
v3.18.4PATCH
Tier diagrams on a phone: two lanes, no crossing lines, and 2N where it is 2N
Tier on a phone (390 px) stacked all nine blocks in one column. The connectors that skip a block (TX → MCC, PDU → IT, the generator feeds) then ran down five or six nested side channels and crossed one another 60 times across the eight views. Each connector met the routing rules, but the flow could not be followed.
Layout. Narrow panes now use a two-lane layout:
Power: path A on the left, path B on the right. The generators and IT load span both lanes, and the MCC sits between them.
Cooling: the liquid sink on the left, the air sink on the right.
Every connector now runs straight down or as one L. Only the A/B loop diamond in Tier III/IV cooling needs a side channel, one on each side of its lane. Desktop layouts are unchanged. On narrow panes the sublabels drop "up" so they fit the lane.
Redundancy labels. Tier IV generators read "18/18 up · need 9 (N+9)". Twice the need or more now reads 2N, in both the diagram and the inspector, which shared no rule before. Tier IV generators, the MCC pair and the Tier IV chiller plant now read 2N.
Gates
tools/test-rzd-ports.mjs adds R6: no two connectors cross on any Tier diagram, at 1440 or 390. It finds 60 crossings on the v3.18.3 tree and none now.
tools/test-ltc-workbench.mjs W9h checks the redundancy label: 18 for 9 is 2N, 17 for 16 is N+1.
v3.18.3PATCH
DH01: every CDU is on the map, and its loop never crosses a rack
Found in review of the v3.18.2 map:
CDU 1 and CDU 6 were missing. The hall geometry put each CDU at the row end of its first rack. Blocks of 16 racks in rows of 19 (or 20) meant two CDUs started on the same row end, and one was drawn on top of the other.
Secondary-loop lines ran through the middle of the racks. The feed from each CDU box ran along the row centre line, and a block that spilled into the next row was joined straight across.
Fixed in the plant model (js/ltc-plant.js, hall geometry):
Serpentine blocks. Racks are assigned to CDUs serpentine: odd rows run east to west. A block that spills into the next row turns at the row end, as the pipe does.
Balanced blocks. Block sizes differ by at most one rack (15 or 16 at 100 MW, instead of 16 × 9 + 8).
One spot per CDU. Each CDU stands in the end zone at a row end its own block reaches, choosing the row holding most of its racks. No two CDUs share a spot. A block wholly inside a row takes the nearer end, on a second slot.
Fixed in the drawing (js/ltc-dh01.js):
Each loop runs along the hot-aisle edge of every row it serves.
A turn within a row pair stays inside the shared hot aisle. A turn across a cold aisle runs 1.1 m out, clear of the row labels.
The feed leaves the CDU box straight to the hot-aisle edge.
Gate — tools/test-ltc-plant.mjs P6 checks, across four plants (100 MW, 37 MW, 13 racks per row, 1.5 MW CDUs):
each CDU is in an end zone, at a row it serves;
each CDU is at the row end its racks reach;
no two CDUs share a spot;
each block covers adjacent rows only;
block sizes differ by at most one.
On the v3.18.2 tree it fails all four plants ("CDU 2 sits on another CDU"; "CDU 3 is not at the row end its own racks reach").
v3.18.2PATCH
DH01 thermal map: a mass and energy balance, drawn as one continuous air loop
Owner review: the flow lines and arrows on the DH01 map were not right and the simulation did not hang
together. The old solver assigned each CRAH to its nearest aisle and spread any deficit with a fitted
"middle of the aisle" shape; the plan drew short arrows from units that did not line up with the aisles,
and nothing showed where the air went after the rack.
Changed — solver (js/ltc-dh01.jssolveHall)
Rack airflow from its heat at the design ΔT.
CRAH fans share one speed up to their rated flow, 5 % over demand under pressure control. Normal: 97 %. Four units out: survivors at 100 %.
Each fan wall is one plenum feeding the aisle mouths in proportion to demand, discounted where its units are out.
Each cold aisle is marched from both ends until its air is used up.
Unmet demand is pulled from the hot aisle and plenum return, solved to a fixed point.
The CRAH coil is an effectiveness model fitted at design. With four units out, supply air rises to 25.2 °C, the set-point is not held, and 29 racks run above 27 °C, all on the west side.
Changed — plan
One continuous loop: fan-wall plenum → end zone → cold aisle → through the rack → contained hot aisle (⊙ rising) → ceiling plenum (dashed) → fan walls.
Supply bands are as wide as their m³/s, narrow as racks draw air, and end where the air runs out; the stretch no air reaches is boxed in red.
Red arrows mark exhaust pulled back to the inlet.
Each CDU's secondary loop is drawn along the racks it serves: red dashed when dead, amber when a common header is short. Throttled racks are hatched grey.
Mutating the return or the bypass split fails seven assertions.
Connectors leave and enter their blocks properly (all LTC diagrams)
Owner screenshot of the Tier power diagram:
Generator → transformer ran along the transformer's left border, with its arrowhead pointing up at that border.
TX → MCC ran along TX's bottom edge and down MCC's left edge.
The router (js/rz-diagram-layout.js) knew nothing about which side of a block a connector meets, and excluded both end blocks from its obstacles.
Changed
RZDiagram.edge reads the side each endpoint lies on (or takes fromSide/toSide). It leaves and enters perpendicular through a short stub, and routes the middle with both end blocks as obstacles.
The hub and model-lab "level with the rack block" adjustment no longer moves an endpoint off the block when the other end lies outside the block's height. This was the 3 connectors the new gate caught that started or ended in empty space.
Live solid connectors stay solid. They used to be dashed by the flow animation, which read as the legend's "Alternate path". Their motion now rides on a bead overlay that follows hover tracing.
Two flows no longer share a stroke:
hub and lab overview — rack → CRAH leaves the rack block's bottom instead of riding rack → CDU's line;
phone stacks — each connector that skips blocks gets its own side channel and its own port, shorter spans nested inside longer ones (the skip % 3 offsets repeated, so two TX B connectors used one port).
fit() now counts connector channels, so they are no longer clipped off the right edge on phones.
Tier blocks are 156 wide with 20 gaps, so the power diagram renders at 0.96 scale instead of 0.81.
Gate — tools/test-rzd-ports.mjs (wired in the ship gate) renders the hub's four views, all 16 Tier diagrams, ISO, ASHRAE, TIA, NFPA and the model lab at 1440 and 390. On all 256 connectors it requires:
a perpendicular first and last segment, pointing into the block;
no segment along an end block's border;
no segment through another block;
no two connectors on one line.
On the old tree it fails 47 connectors, including the owner's tx>mcc. The shared-line rule, added afterwards, found 106 more cases in the 390 px stacks of the new routing; all are fixed.
v3.18.1PATCH
Fixed — every Leaflet map was drawing "API KEY REQUIRED"
CARTO's keyless basemaps (dark_all, light_all, rastertiles/voyager) now answer every tile request with an
image reading API KEY REQUIRED. v3.18.0 moved the incident hub off them; this release moves the rest:
js/rz-map.js — the shared wrapper behind the six pln-java-grid* maps. The tile keys keep their names
(carto-dark, carto-light) so no page changes; they now resolve to Esri's keyless Canvas
World_Dark_Gray_Base / World_Light_Gray_Base.
dc-market-tracker.html, pln-sumatra-grid.html (voyager → light gray), and the dashboard map's dark and
light layers.
Esri serves these basemaps to zoom 16; maxNativeZoom: 16 lets Leaflet upscale beyond that instead of
requesting empty tiles. Attribution changed to Esri / HERE / Garmin / OpenStreetMap as Esri requires.
Verified headless: java-grid 18, sumatra 24, market-tracker 12 tiles loaded from server.arcgisonline.com,
all HTTP 200. No CARTO URL remains in the site source.
v3.18.0MINOR
Incident library: who did what, who should, and two figures the diagram engine draws
Owner: *"sampai level action misal first responder ngapain, on scene commander dll semua roles terlihat
fmeanya bener2 comprehensive", "tambahkan blok diagram, fishbone"*, no filler, no AI-design slop.
Added — response by role on every dossier (tools/incidents_response.py). Two layers that make
different claims and are never mixed:
Recorded — what a role actually did. Each of the 263 actions points at a sequence-of-events entry and
carries a quote that must be a verbatim substring of it; if the source does not say who acted, nothing is
credited. A role the dossier declares but the record never shows acting is shown as silent.
Playbook — analyst guidance per failure mode: 1,524 actions, ≥4 per mode from ≥3 roles across detect,
contain and recover, each citing a named practice (NIMS/ICS, NFPA 1561/75/72/855/70E/110, EN 50110-1,
ITIL 4, change management) or the dossier's own COE / improvement / lesson. No clause numbers.
Rendered as a role × phase matrix, a role list with recorded / playbook per role, a "Who acts" block in
every FMEA row and role tags on the sequence of events. The hub adds the library view: an incident commander is declared in 43 dossiers and seen acting in 1 — and that one is the Dutch public safety region at GRIP 2, not the operator; a first responder in 41 and seen acting in 12 — across 43 public post-mortems, the person in charge of the response is almost never on the record.
Added — two engine figures per dossier (tools/build-incident-diagrams.mjs, RZ diagram engine, IEC/ISA
glyphs): a failure chain — FMEA modes in the order the record first shows them, latent conditions apart,
ending in the outcome (a one-column variant replaces it under 640 px) — and a fishbone of contributing
factors by category, each label pointing at the factor or mode it stands for. Any engine warning fails
the build; a zone-label overflow check was added because the engine sizes nodes but not zone eyebrows.
Changed — page structure. Sticky section nav with scroll-spy; a six-field fact sheet; the root
cause opens with its lead sentences and keeps the full narrative one click away; analysis and the
technical deep-dive are collapsed; COE is a table.
Removed — filler. The generic cascade box (Trigger → operator → services) and the SOE dot strip,
which was clipped at the column edge and repeated the list beneath it.
Fixed — the hub map showed "API KEY REQUIRED" on every tile. CARTO now answers keyless requests with
that image; tiles moved to Esri World Dark Gray Canvas. (dc-market-tracker.html uses the same CARTO URL
and is not in this release.) KPI tiles lost their top stripe and hover lift; recent-incident titles are no
longer cut at 52 characters.
Gatestools/test-incidents-response.py --strict --require-all and
node tools/build-incident-diagrams.mjs --check, both in the ship gate.
v3.17.3PATCH
Incident magnitude notes agree with the measured durations
Since v3.13.6 the duration sub-score is derived from the measured duration by fixed bands, but the
magnitude notes on the incident pages still argued for the judged scores they replaced and quoted
withdrawn durations — Rogers "down ~15h" (the dossier itself withdrew that figure; the measured window is
~26 h), CrowdStrike "Duration is scored 8" (it is 9, ~10.8 days to 99 % of sensors), Dyn "Duration (7) …
~11 hours" (5, ~5.8 h to the second wave's mitigation), Optus "~14h", Flexential "duration (8)".
Fixed — 15 notes rewritten clause by clause: each duration statement now matches the measured
duration and its derived score; other measures (full-restoration tails, end-to-end windows) stay where the
dossier supports them, named as what they are. The Virginia note written in v3.13.5 said duration 2; the
band rule gives 1.
Gatetest-incidents-corpus.mjs: a magnitude note that states a duration score different from
magnitude.durationScore fails (it found 6 in the old notes and the Virginia slip in the new one).
v3.17.2PATCH
The LTC manuals describe the engine the tools run
The model-lab manual still explained the retired computeModel (heuristic multipliers, a 12 MW example)
a release after the lab moved to the shared plant engine; the workbench manual did not mention the plant,
the plant strip, the heat load test or the DH01 map.
Not taken: the same chiller curve's tail (COP ≈ 50 at 10 % load — drive, motor and fan losses do not
shrink with load; clamped at 30 % PLR); a uniform-random 75–100 % IT profile (mean 87.5 %, σ 7.2 % —
identical to the posted figures, i.e. plain noise, not an AI workload shape); a plant report's pump
power (6.1 MW for 775 L/s implies ≈ 600 m of head); the KL vs JB numbers (arithmetic consistent, but
unverifiable without the EPW files — the model takes site bins instead of hard-coding either site).
Fixed — incident hub titles wrap instead of ending in an ellipsis (8 were clipped by 28–183 px; audit-legibility failed on them).
Gates.tools/test-ltc-plant.mjs (balance on 202 plants, 100 MW hand checks, physics direction,
faults, the practitioner claims recomputed). test-ltc-workbench.mjs W7 (+17): the new blocks, ≤ 9 per
view, traceable connector ids, a script at every gate, HLT at L4. Seven mandatory audits that CLAUDE.md
named but ship-gate.sh never ran are now wired (CSS structure, motion, flatten ratchet, incident
builder, incident FMEA, FMEA simulator), and report-register-wiring is declared out with its reason.
v3.13.6PATCH
The duration sub-score follows the duration; long outages read in days
Changed — magnitude.durationScore derived by fixed bands (owner decision). It was a 0–10
judgement made against the durations v3.13.5 corrected, so a 47-minute and a 10-day outage could
carry the same score. Now: <1 h 3 · 1–4 h 4 · 4–12 h 5 · 12–24 h 6 · 1–3 d 7 · 3–7 d 8 · 7–30 d 9 · >30 d
10; no customer downtime 1; where the dossier records no restoration time the judged score stays.
30 of 43 change, and with them the ranking (duration is 20% of magnitude): UniSuper #14 → #4 (≈ 9 days
dark), GCP Service Control 2025 #4 → #8 (≈ 1 h 39 m to substantial restoration), AWS us-east-1 2021
#8 → #13. The band table is in the sub-score's tooltip.
Changed — durations of 48 h or more read as days: CrowdStrike "259h 51m" → "10d 20h".
Gatetools/test-incidents-corpus.mjs: durationScore must equal its band (seen to fail on one
drifted score).
v3.13.5PATCH
Incident durations mean one thing; the Virginia dossier says what actually happened
Two owner decisions from data/incidents/_review-queue.md.
Changed — durationMin, one definition across the library. Customer-impact onset → substantial
restoration, taken from the dossier's own sequence of events and metrics. Before this it meant five
different things: a fire's knock-down time, a vendor's source-side exposure window, a media headline,
a straggler tail, and in some cases nothing the dossier could support. 31 of 43 changed.
Every measured duration now carries durationEvidence: two pointers (onset, restored) into the
dossier whose quoted text is checked verbatim. The header reads "Xh to substantial restoration";
Impact & scale states the basis ("Customer impact 04:58 EDT 8 Jul → 07:00 EDT 9 Jul (CRTC window)").
Where the dossier cannot support a number, none is shown and the page says why: *no customer
downtime recorded (Council Bluffs, Virginia) or restoration time not established* (Azure China
North 3, BA Heathrow, Digital Realty LA and Singapore, NIRS, OVHcloud, PSN, Red Sea).
Largest corrections: CrowdStrike 78 min → ~10.8 days (78 min was CrowdStrike's own exposure window;
hosts needed manual recovery to ~99% by 29 Jul); AWS EBS 2011 ~90 h → ~36 h (97.8% of volumes back,
straggler tail excluded); Rogers 15 h → ~26 h (the dossier had already withdrawn 15 h as
unsupported); X Oregon 48 h → ~24 h; Azure West US 2 22 h → ~7.6 h (50% at 06:15, 95% at 12:00).
Changed — virginia-dc-alley-heatdome-2022 re-framed. It was titled and categorised as a July 2022
heat-dome cooling failure; its own root cause and SOE say there was none. Now: *Transmission
Constraint and Mass Data-Center Load Loss (2022–2026)*, category power, dated to Dominion's 29 Jul 2022
constraint notice, with the 2024 (~1,500 MW, ~60 data centers) and 2026 (~3 GW) load-loss events that
rode through on backup. Magnitude re-scored (users 2, duration 2 — no customer downtime), so it drops
in the ranking. Slug unchanged to keep the URL.
Gatetools/test-incidents-corpus.mjs: every dossier carries durationStatus (measured |
no-outage | not-established) and durationBasis; a measured duration must resolve both evidence
quotes verbatim; a non-measured one may not carry a number. Seen to fail on one altered quote.
Alsoincidents-chunks.jsonl regenerated — it still carried the four header dates corrected in
v3.13.0. Still open:magnitude.durationScore was judged against the old durations and is not
re-scored here; listed in the review queue.
v3.13.4PATCH
The LTC manual describes the workbench that exists
> Repairs commit 58d4e25, which was pushed from a tree based on v3.12.1 and so reverted v3.13.0–v3.13.3. This commit restores that work in full and carries only the manual change on top.
manual/standards-ltc-lab.html — the page every workbench module links as "Manual" — still described
the pre-v3.12 hub: evidence checkboxes, an items_passed / total × 100 compliance score and five
"engines". None of that exists any more. It is rewritten as the workbench's methods reference: for each
module, the equations the tools run (Magnus psychrometrics and pressure at altitude, the Tier
maintenance/failure test and k-of-n availability, the TIA-568.3-D loss sum against IEEE 802.3 budgets,
the NFPA 2001 flooding equations with the NOAEL check, ISO/IEC 30134 KPIs and the ISO 50006 regression
with Guideline 14 limits), the limit tables they use, and where each stops being valid. HowTo and FAQ
structured data were rewritten to match, and the manual index card renamed.
ISO 14644 misattribution fixed. The old manual grouped "ISO 50001 / 14644" as energy governance, and
the model manual cited ISO 14644 as the source of its WUE metric. ISO 14644 is the cleanroom
air-cleanliness standard; WUE and PUE are defined in ISO/IEC 30134-9 and 30134-2. Both manuals now cite
those.
v3.13.3PATCH
Two independent reviews of the FMEA layer, and every finding acted on
A UI review against the anti-vibe standard and a code review (XSS, a11y, state) were run on
v3.13.1–v3.13.2. 23 of 24 findings fixed; one declined (below).
Fixed — design (UX law #14, one accent per view)
Criticality matrix: the amber/red cell wash is gone; zones are 1px hairline boundaries drawn
wherever two neighbouring cells fall in different zones.
Coloured left stripes removed from every incident page: .chip (4 variants), .ca-block,
.mag-trace — state stays in the text colour, grouping is the panel and its hairline.
"Most risk still standing" bars neutral, not amber. Facts list drops the residual row that
repeated the metabar. Dimmed (not-yet-in-play) rows at 0.45, readable.
Type scale collapsed from 12 one-off sizes to five steps. Slider value sits beside its thumb.
Tap targets 44px under 640px. Replay ticks mark new modes by height, not a decorative dot.
SOE timeline (every incident page): one 16-character time label ran ~96px into a 92px step
and overprinted its neighbour on 28 pages. Labels now wrap by word into two ≤13-character
lines, and the strip has 46px margins so the end labels are not cut at the SVG edge.
Fixed — code
select(id, noScroll) scrolled anyway (a && b || c precedence) — deep links jumped back up.
A malformed #hash threw URIError on load and on every hashchange.
Bars inside role=list became role=button; the container is now a group. The scenario
control used role=radio without roving focus; now aria-pressed toggle buttons.
Keyboard selection moves focus to the what-if panel; focus survives a drag that moves the
marker to another cell. "Resume" only after a pause.
A what-if above the as-found axis clipped at full width; the axis now widens.
Dossier URLs: only http(s) becomes a link. Embedded JSON escapes every <, closing the
<!--<script parser state as well as </script>.
Gates
test-fmea-sim.mjs exits 1 on any finding (it printed FAIL and exited 0 without
--strict). New checks: JS priority colour equals the builder's for every mode, matrix markers
equal occupied cells, replay lights the data's count at the first and last event, a what-if drag
must change the RPN, zero pages fails. Seen to fail: a perturbed threshold and replay offset gave
88 + 64 findings.
incidents-visual-qa.mjs: SOE timeline labels may not overprint or run past the SVG edge.
Against the v3.13.2 pages it flags 28 of 43.
Declined — "the metabar is a hero-metric strip": it is one sentence-line, the form the
standard prescribes in place of stat tiles.
v3.13.2PATCH
FMEA across the library: what 43 retrospectives say together
The incident hub (dc-incidents.html) gets a library-wide FMEA view, summed from the per-incident
FMEAs — nothing is scored at hub level.
Added
Headline line: 284 failure modes in 43 incidents; 213 (75%) caught only after impact (D ≥ 8);
197 (69%) with no published action; published actions cut total RPN by 17%.
Where detection sat — modes per detection score; each row opens to the modes behind it, worst
RPN first, each linking straight to that mode on its incident page. Scores between rubric anchors
now read "Between 6 and 8" instead of borrowing a neighbour's description.
What the published actions moved — 87 of 284 modes have one; 46 lower occurrence, 29 detection,
27 severity.
By incident category — the split the corpus had never shown: fire incidents have 96% of modes
with no published action and actions cut 2% of their RPN; software 39% and 34%; network 72% and 15%.
Most risk still standing — the ten incidents with the highest RPN left after every published
action (NIRS Daejeon 2025 first: 2,906 of 2,906, all 8 modes residual).
Deep links: incident-…html#FM-03 opens that worksheet row and loads the mode into the what-if.
Gatetools/test-fmea-sim.mjs now also checks the hub: its mode count equals the sum over the
incident pages, every mode link resolves to a worksheet row (seen to fail on a renamed anchor), and
the section fits 390 px with no sideways table.
v3.13.1PATCH
The FMEA becomes something you can run: replay, scenarios, what-if
Owner: *"buat lebih detail dg data2 dan simulasi dan sangat intuitive dan immersive. Filler2
hilangkan."* Every one of the 43 incident FMEAs now carries an interactive layer
(js/rz-fmea-sim.js) over the static bars, matrix and worksheet, which still stand alone without JS.
Added
Replay — step or play through the sequence of events; each failure mode lights at the first
event its evidence cites, with a running "in play: n of N modes · RPN x of y". 257 of 284 modes cite a
timed event; the other 27 are labelled latent and light at the end rather than at a guessed time.
Scenarios — As found · Published actions · What-if: detect at onset (every D capped at 6) ·
What-if: guard the cause (every O capped at 2). The two what-ifs are labelled hypothetical; they
show which lever carries an incident's risk. On AWS EBS 2011: actions −37%, detection −25%,
guarding the cause −56%.
What-if per mode — select any bar, matrix marker or worksheet row; drag S, O and D. Under each
slider: the rubric line for the dragged value and the analyst's written rationale at the scored
value. The matrix draws the selected mode's move; priority recomputes with the same rule as the
builder.
Facts list per incident: modes caught only after impact (D ≥ 8), how many published actions cut
The 6,000-row worked-model dataset disclosed the choice honestly — `"320 used as low-end
illustrative"` — but its own cell in the same row stated the range as 370–575, which 320 sits
below. A conservative figure is defensible. A conservative figure borrowed from a different fluid
and printed below the range you yourself quote is not.
Replaced with the primary reference, not a vendor range
IPCC AR6 WGI Table 7.SM.7 tabulates these compounds directly:
fluid
AR6 designation
lifetime
GWP100
was
Novec 7000
HFE-347mcc3 · CH₃OCF₂CF₂CF₃
5.1 yr
576
320 in the article, 370–575 in the data
Novec 7100
HFE-449s1 · C₄F₉OCH₃
4.8 yr
460
320 (3M literature)
Fluorinert FC-72
PFC-51-14 · n-C₆F₁₄
3100 yr
8620
7910 — the AR5 value, in a row that cited AR6
AR6 also lists HFE-7100's isomers separately, n- at 544 and i- at 437, which brackets the 460
mixture value and confirms the row matched is the right one rather than a same-formula coincidence.
Novec 649 is absent from the table, consistent with its five-day lifetime; its GWP of ~1 rests
on the lifetime rather than on a tabulated figure, and the dataset now says that instead of
implying a citation. Opteon 2P50, R-1234yf and R-1233zd(E) already agreed with AR6 and now carry
their AR6 designation so the next check is trivial.
What moved
The worked example goes from 61 to 111 t CO₂e/yr for a single tank at field loss — the same
192 kg of fluid, correctly weighted. The chart's CO₂e series scales by 1.8×. The live calculator's
two-phase-pfas constant goes 320 → 576. And the fluid-choice swing between a Novec-class HFE and
a Galden-class PFPE is ~17×, not the ~30× the understated figure produced.
Every mass figure is unchanged. Only the weighting was wrong, and it was wrong in the direction of
making the problem look smaller.
v3.11.3PATCH
The dataset gate now covers every dataset, not the one that broke
v3.11.2 added audit-fluid-properties.py after four rows of one CSV turned out to mis-parse. That
scope was wrong, and the reason is in the defect itself: unquoted commas are a property of
hand-edited delimited text, not of chemistry. The four broken rows happened to be chemical names.
The next one will not be.
Renamed to tools/audit-published-datasets.py and widened to **41,469 rows across all 51
published CSVs**, each parsed the way a consumer will parse it. All currently pass — the audit was
run across every file before the gate was written, so this is a verified floor rather than a hope.
Added — relations that hold by definition, so they need no standard in hand
data/fire/clean-agent-properties.csv is a fire-safety table, and two of its columns constrain
each other regardless of which code is being followed: a design concentration cannot sit below the
concentration that extinguishes, and a no-observed-adverse-effect level cannot exceed the
lowest-observed one. Eight such relations across five agents, all holding.
Blank cells are skipped, not read as zero. CO2 carries no NOAEL because it is lethal at design
concentration and its own source cell says so; coercing that blank to 0 would manufacture a passing
comparison out of missing data.
Deliberately not gated: the 1.2× class-A safety factor and the occupied-space design ≤ NOAEL
rule. Both are NFPA 2001 provisions with exceptions, and asserting a compliance rule from memory is
the same mistake the whole gate exists because of.
Proven against four seeded defects before being trusted
Structure, on a file unrelated to the original bug (spares-oems.csv, an injected comma). NOAEL and
LOAEL swapped. A design concentration pushed below its extinguishing concentration. And the original
transposed vapour pressure, at ×0.37. All four exit 1 under --strict; clean exits 0.
v3.11.2PATCH
Twenty-eight pages had navigation links too small to tap
Found while verifying v3.11.1 on the live site rather than locally. The protocol explainers show no
hamburger at phone width, which is a valid design — their navigation is two always-visible links.
Measured, those links render 17px tall at 390px, on 28 pages: all 25 network/** explainers
plus all-in-one-dashboard, network-compare and network-visualization-hub.
WCAG 2.2 SC 2.5.8 (Target Size, Minimum, AA) asks for 24×24 CSS px. Its "inline" exception covers
targets inside a sentence; a navigation bar is not a sentence, and on those pages those links
are the only way out.
The cause is that .nav-right renders its links as bare inline anchors, so the target box is the
TEXT box. The fix is a floor, not a redesign: min-height: 24px with inline-flex grows the box
without moving the text, in the shared stylesheet all 28 already load. The 44 other .nav-right
pages whose targets already clear 24px are unaffected, because a floor beneath them changes
nothing.
Added
tools/test-tap-targets.mjs, wired into ship-gate.sh. Every page with a .nav-right bar, at
390px, must render each visible link at least 24px tall. **Proven RED against shipped v3.11.1:
28 pages. 0 after.**
Why the hamburger gate could not see this
test-mobile-nav.mjs enumerates pages that have both a navbar AND a menu container, then asserts
one working toggle. These pages have no toggle and no menu container — they were never in its
population, and they were right not to be. A gate that asserts "the hamburger works" cannot notice
a page that correctly has no hamburger and fails for a different reason. That is an argument for
checking the rendered page against a different question, not for widening the first gate.
The downloadable dataset was corrupt for every parser, and nobody could see it
v3.11.1 fixed a vapour pressure in article-26 that belonged to a different fluid. Auditing the
rest of the table it came from turned up a second defect of a different kind, in the same file.
data/article-26/fluid-properties.csv is offered to readers as a download, so a stranger loads it
into pandas and trusts the result. Four of its eleven rows mis-parsed, because chemical names
like cis-1,1,1,4,4,4-hexafluoro-2-butene and 2,3,3,3-tetrafluoropropene carry unquoted commas.
Those rows silently shifted columns: R-1234yf reported C3H2F4 where its boiling point belongs,
and R-1233zd(E) reported a CAS number as a vapour pressure. Opened in a text editor the file looks
perfect. One of the four was introduced by the v3.11.1 fix itself — writing
3M TDS (GWP, vapor pressure) into the source cell added a comma and split the Novec 7000 row.
All eleven rows are quoted properly now and parse to fifteen columns.
Added — tools/audit-fluid-properties.py, a gate the eye cannot be
Two defects in one week, neither visible by reading the file, is a gate's job description.
Structure: every row must parse to the header's column count. A row that does not is broken for
every consumer, full stop.
Physics: boiling point and vapour pressure at 25 °C are not independent. Trouton's rule gives
the enthalpy of vaporisation from the boiling point alone (ΔS_vap ≈ 88 J/(mol·K)), and
Clausius-Clapeyron turns that into a predicted p_sat(25 °C). **Needing only the boiling point is
what makes it an independent check on the pressure cell** — exactly what the Novec 7000 defect
lacked.
Run against the repaired table, all seven pure compounds agree: Novec 7000 ×0.88, Novec 7100 ×0.96,
Novec 649 ×0.93, Opteon 2P50 ×1.20, FC-72 ×0.89, R-1234yf ×0.87, R-1233zd(E) ×0.92. Run against the
unfixed table it fails both original bugs — the transposed pressure at ×0.37, and the unquoted
comma by column count — and exits 1 under --strict. A gate that has never been shown to fail is
not a gate.
Mixtures are exempt by name, with the reason written down, never by pattern. Galden HT55/HT70
are polydisperse PFPE cuts whose "55" and "70" are nominal mid-cuts rather than normal boiling
points; FC-40 is a blend; mineral oil gives ">300" and "negligible", neither of which is a number.
The split turns out to be exact — every pure compound passes and every failure is a mixture — and
that pattern is the evidence the method works, which a blanket "skip anything that looks like a
mixture" would have hidden.
Wired into the ship-audit list in CLAUDE.md, because a gate that exists and is not wired is a gate
that does not exist.
v3.11.1PATCH
The drawer opened and the page showed straight through it
v3.10.21 put the hamburger back on all 138 navbar pages. Opening it on
datacenter-solutions.html still printed the hero headline over the menu — a screenshot of the
live site shows "Plan, Cost, and Commission Your Data Center" legible across the open drawer.
The drawer's background was not the problem: it computed rgba(15,23,42,0.97) and its gutter
pixels sampled (23, 30, 49). **A z-index on the drawer cannot fix this, because the drawer lives
inside .navbar and can only ever paint as high as the navbar's own stacking context.** Lifting
the NAVBAR while the drawer is open is what makes the page disappear behind it. The background is
fully opaque now for the same reason — 3% of a bright gradient headline is still legible.
Found by screenshot, fixed by screenshot. elementsFromPoint reported the drawer ABOVE the H1, and
that reading was useless: the pixels disagreed.
Added — N5, the check a link count cannot make
tools/test-mobile-nav.mjs counted tappable links and passed a drawer the page printed through.
N5 samples six points down the drawer's own area and requires each to land on the drawer rather
than on the page beneath. It skips root-gated pages, where a full-screen gate overlay covers the
drawer legitimately and N5 has nothing to say.
The gate was flaky, and the flake was the same one dark-coverage had
Three consecutive runs of the UNCHANGED gate returned 4, then 3, then 1 finding — a shifting set,
which is the signature. It reported drawers as closed while simultaneously counting 22, 19 and 17
visible links inside them; measured alone, every one of those reported opacity: 1.
A drawer fades and slides in, and getComputedStyle during that returns the interpolated value.
Under the CPU load of a 138-page sweep the animation clock runs behind and the gate reads a
half-opened drawer. tools/audit-dark-coverage.mjs hit the identical flake reading body colour
mid-theme-flip. Same remedy: **suppress transitions before measuring, and assert the settled state
— which is exactly what the animation was delaying.** Three consecutive clean runs after.
Fixed — the gate crashed on its own viewport switch
Flipping isMobile on a live page makes Puppeteer reload it, and doing that once per page killed
the run with a protocol error. Mobile and desktop are now two passes with the viewport set once in
each; the number of navigations is identical. A page that fails to render is now reported as a
finding instead of being skipped by a bare catch.
Fixed — article-26 quoted another fluid's vapour pressure
article-26 gave Novec 7000's vapour pressure as 270 hPa at 25 °C. The 3M datasheet gives
65 kPa — 650 hPa. 270 hPa at 25 °C is Novec 7100's figure, and the article's own
data/article-26/fluid-properties.csv carried the proof one row apart: the 7100 row reads
~27 kPa. One row's value had been written onto another's.
Verified three independent ways before anything was touched: the datasheet (65 kPa), a vendor spec
quoted as 9.5 psi at room temperature (65.5 kPa), and Clausius-Clapeyron anchored on the 34 °C
boiling point and the 142 kJ/kg heat of vaporisation — which uses neither of the other two and
lands at 72.4 kPa. The article's own water comparison of 32 hPa is correct, so the units were
consistent and the gap was real.
The correction makes the argument stronger, not weaker. The number was being used to show that
opening a two-phase system releases significant vapour, and the true figure is 2.4× larger. Liquid
equivalent per m³ of headspace goes from 1.56 L to 3.75 L.
A first count found nine occurrences. A sweep for the DERIVED claims found four more, because the
error travelled without its own number attached: "eight times that of water" carried it in prose,
and one passage credited the 8.4× to industry documentation when it came from the article's own
division of 270 by 32. That ratio is now stated as a vapour-pressure ratio of about 20×, not an
evaporation-rate ratio — a pressure quotient is not a rate, and since this molecule is ~11× heavier
than water, calling it one understates the mass anyway.
The calculator's 0.008 coefficient is deliberately unchanged, and settling that did not need
the original author. Its comment claimed a basis of "vapor pressure 270 hPa, open surface ~2m², avg
open time 4hr", but the implemented term scales with fluid volume while a surface-evaporation
model does not scale with volume at all — so one cannot be derived from the other, and numerically
the surface model over-predicts it by an order of magnitude at any tank size. 0.008 is an assumed
per-event fraction, about three headspace exchanges plus film and drag-out. The comment now says
so. Output is byte-identical. **Compare the shape of two models before asking which number came
from which.**
Scope beyond the article: glossary.html, the downloadable dataset, and **seven unpublished social
drafts** that would have spread it to LinkedIn, Medium, Quora, Mastodon, TikTok and Facebook.
Generated artefacts rebuilt from source and the explain-db token rotated on the 108 pages that load
it.
v3.11.0MINOR
Eleven figures, fourteen reasons
Explanatory figures for the article corpus, built rather than hand-drawn, and a written
verdict for every article that did not get one.
The corpus survey corrected an assumption first: counting <svg> per article returns 19–52
and looks like a corpus rich in diagrams, but almost all of it is interface — share buttons,
nav chevrons, icon glyphs. Counting only SVGs larger than 200 units gives thirteen real
figures across twenty-nine articles, and eight articles carrying no visual at all.
Added
tools/build-article-diagrams.mjs — figures rendered at BUILD time, so they survive
the PDF export, the Markdown extraction into llms-full.txt, a reader with JavaScript
off, and the first paint. The author places an empty
<figure data-rz-figure="id"></figure> where the figure belongs; the tool fills it and
never inserts a placeholder itself, because where a figure belongs in an argument is an
editorial decision. Every definition records a because naming what it adds that the
prose cannot.
Eleven figures, each earning its place by showing a shape a table could not:
article‑4 and 6/8 (what the calculator actually does — the output is a distribution
and the ranking is derived, not authored), 10 (the cooling loop §8.1 assumes but never
draws), 12 (the tariff gap, as a distance rather than a subtraction), 17 (Jevons as the
reinforcing loop it is, not three independent rows), 18 (two tables never connected — the
density trajectory crossed the air ceiling, so the heat path changed shape), 19 (the
corridor is one system, not a choice between two sites), 21 (the causal chain that makes a
20‑year PPA legible as a replacement for one link), 22 (drawn at the real 24:1 trace
ratio — the retimers exist only to pay for the distance), 26 (the escape routes bypass the
meter), 27 (two of three levers never reach the pool).
standarization/ARTICLE_FIGURE_AUDIT.md — every article, and the reason for each
verdict including the fourteen against building. Article 25 nearly received a static
duplicate of the 6 GW gap its own living diagram already animates; article 1 was excluded
by reading its code, which scores dimensions and runs no Monte Carlo at all.
Changed
The diagram engine gained the guards listed in DIAGRAM_ENGINE_STANDARD §3e. Colour may
never be the only channel (WCAG 1.4.1) — which caught hot and cold streams separated by
hue, and three node classes using ISA alarm hues as categories. A legend may not name
two things with an identical swatch, which found three such pairs already shipped. Ports,
node height, unknown options and the node budget all now report instead of failing quietly.
Figure size bounds travel with each figure, from its own viewBox. One blanket
min-width:640px had produced two opposite failures: wide figures shrank until their type
fell under the 8.5 px floor, narrow ones stretched until they shouted.
Fixed
tools/test-asset-cache-tokens.mjs --fix <asset> repairs the tokens it reports, for the
assets named explicitly. The manual sweep was forgotten twice in one session; a control
that has to be remembered is only a rule again.
v3.10.21PATCH
v3.10.20 fixed "two hamburgers" by shipping "no hamburger"
Owner, on the same page, after that release: "Malah nggak ada navbar. Tolol. Audit fix yg bener."
He was right. Wiring the page's own button, the code replaced its <i class="fas fa-bars"> with
three bare <span>s — and those are styled by .rz-nav-burger span in styles.min.css, which
these pages do not load. The button rendered 0px wide. An existing icon is already a hamburger;
the span-morph is cosmetic and must never destroy content that is doing the job.
Then the sweep he asked for found far more than the seven pages v3.10.20 touched
Every page the sitemap publishes that has a navbar and a menu — 138 of them, not a sample:
fault
pages
no visible toggle at all
24 — article-1…18, FF-1…3, geopolitics-1…3
a 4px-wide toggle that opened the menu and could not be tapped
1 — spares-readiness-calculator
menu opened showing 1 of its 3 links
1 — rfs-readiness-workbench
The 24 had been broken all along, for a reason worth writing down.findNavbar() used a single
querySelector with a comma list, and that returns the first match in **document order, not
selector priority**. On article pages nav.toc-sidebar sits above nav.navbar in the markup, so
the last fallback in the list — body > nav:first-of-type — claimed the table-of-contents sidebar.
That sidebar is display:none on a phone, so the burger was injected into a hidden element and no
reader ever saw a menu button.
Fixed, all in js/rz-mobile-nav.js
Never destroy an existing button's content. Spans are added only when the button is empty.
Selectors are asked in priority order, and a candidate must have a rendered box — a hidden
shell is not the navbar a reader uses.
The fallback styles the burger too, and all four menu shapes (.nav-menu, .nav-links,
.cx-nav-links, .rfs-nav-links). The inject branch set burger styles inline; the wire branch
had nothing, which is how a 4px tap target opened a working menu.
Drawer links are forced visible.rfs-readiness-workbench hid two of its three links at
mobile width, from a time when there was no drawer to put them in. Inside an open drawer there is
room, and hiding them there serves nobody.
The gate now covers every navbar page, because a sample told me about the sample
tools/test-mobile-nav.mjs sampled seven "navbar shapes" and passed while 25 pages were broken. It
now walks all 138. Proven RED against shipped v3.10.20: 28 findings on 28 pages — 26 with the
wrong number of visible toggles, 2 opening a menu with nothing tappable. 0 after.
I had to correct my own test twice, and both corrections are the same lesson
Height is not "open". A correctly CLOSED slide-in drawer measures 774px tall, parked at
left:-601px. Judging by height called two working pages broken.
A centre hit-test is not "visible". It fails on a root-gated page because the gate overlay
sits above the menu, and on another because the centre landed on a nested input.
It now counts what a reader actually needs: links they can see and tap. Same lesson as v3.10.18 —
measure the thing the metric is a proxy for, not the metric.
v3.10.20PATCH
Two hamburgers, neither working
Owner, on datacenter-solutions.html at phone width: *"Hamburger button utk menu ada 2 dan nggak
working saat di klik. Ini critical."* Two buttons in one header, and neither did anything. Seven
pages: datacenter-solutions.html and the six pln-java-grid* pages.
js/rz-mobile-nav.js states in its own header that it exists to prevent exactly this — it detects
an existing toggle and wires it up "instead of double-injecting". **Three compounding faults, and
none of them is visible by reading the file.**
The detector's selector list did not include .mobile-nav-toggle, which is how those seven
pages mark their button up. The query missed, the code fell through to the INJECT branch, and
the reader got a second hamburger beside the first.
Those pages do not load styles.min.css, so none of its 45 body.rz-nav-open rules reach
them. The surviving burger set the class correctly and nothing listened.
They DO ship their own drawer — .navbar.menu-open .nav-links, designed in their own
palette, var(--pjg-bg) and all — but 89 pages carry a mobile block with
.nav-menu, .nav-links { display: none !important }, and importance beats specificity. A
finished design sat behind a rule it could never win against, with nothing setting its class.
Fixed
.mobile-nav-toggle and #mobileNavToggle added to the detector. One hamburger, not two.
menu-open is now toggled on the navbar alongside body.rz-nav-open, so a page whose own drawer
rule CAN win gets to use it.
A minimal drawer is injected only where no body.rz-nav-open rule exists. The coverage test
deliberately does not count .menu-open: it is not !important, so it cannot beat the mobile
hide, and counting it suppressed the fallback on exactly the seven pages that needed one.
The CSS trap that cost the most time
The first fallback used position: fixed; top: 56px; bottom: 0 and rendered 48px tall. No rule
set a height and max-height was none. The cause: **.navbar carries `backdrop-filter:
blur(20px)`, and a filtered ancestor becomes the containing block for its fixed descendants** — so
the drawer resolved against a 64px bar instead of the screen. The navbar is itself position: fixed
on these pages, so the fallback now hangs off it with top: 100%, which is both correct and simpler
than fighting the containing block.
Added
tools/test-mobile-nav.mjs, wired into ship-gate.sh. At 390px on seven navbar shapes it
asserts exactly one VISIBLE toggle, that clicking it opens a menu of real size, that clicking
again closes it, and that no toggle is visible at 1280px. Proven RED against the shipped code,
where it reports precisely the owner's two symptoms: found 2 toggles, and menu 0x0px.
Reading the source would not have caught any of this. Only a rendered page at a phone width,
clicked, answers the question the owner asked — which is why the gate drives a real browser.
Verified not to regress the 82 pages that already worked
index.html, articles.html, glossary.html and tools.html each still show one toggle, open to
a full drawer and close again.
v3.10.19PATCH
The token followed the clock, and the clock moved
v3.10.13 made every ?v= token content-derived, <yyyymmdd>-<first 8 of sha256>. The digest half
is stable and is the half the gate compares. **The date half was minted from date.today() by a
sweep script retyped by hand at each ship**, and that is a defect with two faces:
Midnight churn. Re-running the sweep the day after a release rewrites every token on every
page — 253 pages, for assets whose bytes never changed. Measured this morning: index.html
carries seven tokens, all dated 20260920, every digest current. A clock-derived sweep would
have restamped all of them 20260921 for nothing.
A split token, introduced while fixing the first. Regenerating ONE page with today's date
leaves it as the only page carrying a different token for a shared asset — precisely the "one
asset served under several tokens" defect tools/test-asset-cache-tokens.mjs exists to catch.
The first face was found and fixed in tools/build-changelog-html.py (ed5eb0f1): the ship gate
failed on a generated artifact nobody had touched, and the obvious remedy was the trap above.
Added
tools/normalize-cache-tokens.mjs, wired into ship-gate.sh. **The token follows the SITE, not
the clock:** it adopts the date the rest of the site already carries for this exact digest, and
mints today's date only when the digest shows the asset genuinely changed. A content change still
busts the cache, because the digest changes; a rebuild of an unchanged file does nothing.
Both halves proven by injection: with the tree clean it reports nothing to normalise; after a
real edit to styles-index.css it mints ?v=20260921-0409f1bd for the one changed asset and
leaves the other six alone.
It uses the local date, not UTC. build-changelog-html.py mints with Python's date.today(),
which is local; this box is UTC+7, so the first cut of this tool minted 20260920 at 00:30 local
— two dates for one digest, which is the split it exists to prevent.
Why a tool and not a snippet
This rule lived in an ad-hoc Python block pasted at each ship, which is exactly how the clock got
into the token: a rule that lives in the author's head is applied when the author remembers it.
test-asset-cache-tokens.mjs checks the digest; this writes it. Version pins are left alone — a tag
carrying a data-*-authority attribute declares a CONTRACT version, and hashing those blanks the
cockpit fail-closed.
Also in this release
Dark-coverage measures the settled colour (17b031ca). The body carries
transition: background-color 0.3s, so a theme flip ANIMATES, and getComputedStyle during the
transition returns the interpolated value — pure white under CPU contention. That is the
body-lum=255 light-block=has-share-buttons false positive that has cost three investigations,
shifts to a different page set every run, and names pages that render correctly in isolation.
Independently confirmed here before the fix landed: the four pages the v3.10.18 battery and a
solo re-run flagged — article-2, article-25, article-7, FF-2 — all measure
rgb(14, 15, 18), luminance 15, when probed directly. Transitions are now suppressed before
the flip, so the reading no longer depends on how busy the machine is.
Fixed while shipping this
Two type tiers in the diagram engine were below the legibility floor.article-10's new
loop figure rendered ten labels at 8px — "DOES THE WORK", "RAISE SETPOINTS", "RAISE CYCLES",
"DELIBERATE" and six more — and the legibility gate caught it on the release battery, having
reached main in bbd8ce4e.
My first fix was wrong and a second gate caught it. I edited the font-size attributes in
article-10.html directly; the article-diagrams gate immediately reported the page out of date
against its generator. The figure is GENERATED, and hand-editing generated output is the exact
mistake this repository's freshness gates exist to prevent — committed by the person who spent
the previous release writing them.
The real cause is js/rz-diagram.js: eyebrow and arrow-label are the smallest type the
engine emits and were size: 8, so **every figure carrying an eyebrow or an arrow label failed
the floor the moment it shipped.** Raised to 9 in the engine and all four figures rebuilt with
their own tool. The type ramp (from documentation/design.md) and the 8.5px legibility floor are
both this repository's standards and they disagreed; the floor wins, because it is the one
measured against a rendered page. 9 is already sublabel's size, so the ramp keeps three
distinct steps below datum. Engine gates: 113/113 assertions, 0 collisions after the reflow.
AGENT_HARNESS_STANDARD.md's Last updated: now matches the release date. The harness gate
asserts that parity and failed because this release crossed midnight — the same clock that put
this entry in the changelog.
v3.10.18PATCH
Four of my own audit findings were measurement artifacts
Working the SEO ledger down, four rows did not survive being re-measured. They are recorded because
the mistakes are the same class the site's gates exist to prevent — a count taken without asking
what it counts:
claimed
truth
what was wrong
7 orphan pages
0
hrefs compared as raw strings instead of resolved against the linking page's directory, and sitemap URLs in directory form (manual/) excluded from the source list — which is where all of them are linked from
41 pages with multiple <h1>
0
<h1> inside <script> string literals counted as markup; one calculator holds 11 report templates for its PDF export
94 titles out of band
2
a generic 30–60 char rule applied to titles that deliberately carry `\
ResistanceZero`. The question that matters is whether the SUBJECT survives the ~60-char display cut: 174 of 176 do
96 descriptions out of band
90, cosmetic
length is not the defect, truncation is. 86 end on a complete clause; the rest are keyword lists cut mid-item, which loses no meaning
A count is not a finding. Each needed a question about what the number means.
Added
tools/test-page-metadata.mjs, wired into ship-gate.sh — the gate the ledger carried as OPEN-7
because v3.10.17 brought metadata to 100% with ad-hoc scripts and left nothing holding it there.
It requires eleven signals of every page sitemap.xml publishes, parses every JSON-LD block
rather than counting them, checks each advertised og:image is on disk, and requires a
publication date on every article-type node. Proven RED against the pre-v3.10.17 tree: **193
missing signals across 34 pages**.
Fixed
Every published page now has exactly one <h1>. Five cockpits had none — chiller-plant,
datahall, ict, water-system, EPMS_Telemetry — because their identity sits in instrument
chrome. Each page's single .title/.brand element was promoted to <h1>, keeping its class and
therefore its styling. Their mobile hero rules (h1 { font-size: 1.6rem !important }) are written
for article headings and would have jumped an instrument header to 1.6rem on a phone, so those
rules are now scoped h1:not(.brand). a11y audit clean after, both themes.
111 article-type nodes carry datePublished; 76 pages had none. The date is the commit that
ADDED the page — when it was published here, which git records exactly. That is a different claim
from sitemap <lastmod>, which tools/build-sitemap.py:53 refuses because a commit does not
prove a CONTENT update. First appearance is not an update; it is a publication. Spot-checked
against known history: manual/pue.html 2026-07-20 matches the manuals programme,
network/industrial-ot/modbus-tcp.html 2026-05-24 matches its own asset token.
dateModified is deliberately absent — nothing in this repository can prove one.
Score
78 → 88 / 100, with the same caveat as before: these are static source measurements, not
Lighthouse runs and not Search Console data. Core Web Vitals are inferred from page weight, never
measured, and an SEO score is a proxy for discoverability, never a ranking prediction.
Also measured and found clean, so no row was opened
0 pages carry more than one <title> in <head> (the apparent duplicates are PDF-export templates
inside JS strings, well past </head>), and 0 titles are duplicated across pages.
The two title rows, judged rather than left hanging
geopolitics-1.html is one character over the display cut and article-28.html's colon-led
hook — "The Compression Horizon:", 24 characters — survives it intact. Shortening either would cost
the question the article is about, so this is closed as judged, not left implying pending work.
Still open, and each says why
96 descriptions truncate mid-clause, and re-measuring says it cannot be swept: **70 are a single
list with no sentence boundary inside the window** — there is nothing to trim to — and the other
26 could end at a boundary only by discarding the text after it. Both paths delete authored content
to fix a cosmetic SERP issue, so it stays open. Plus datahallAI.html at 1,375 KB against a 55 KB
median, and the commercial artefacts in a public repository where deleting the files would not
remove them from git history. Ledger: standarization/Audit result/SEO_AUDIT_LEDGER.md.
v3.10.17PATCH
A gate that walks one directory reports a clean site by not looking at it
Two findings from the SEO audit, and they are the same bug twice: a tool that enumerates pages with
readdir(ROOT) or glob("*.html") sees the repository root and nothing else. Half this site's
public surface lives in network/, manual/, id/ and prd/.
A — 79 of 90 sub-directory pages were serving a shared asset under a stale or legacy token.
tools/test-asset-cache-tokens.mjs has been green on every ship while never opening those
directories:
asset
pages carrying a stale token
js/rz-version.js
78
styles.min.css
76 — still ?v=20260908-editorial, twelve days and many rebuilds old
js/rz-mobile-nav.js
53
js/rz-cookie-consent.js
53
auth.js
43
js/rz-command-palette.js
40
auth.js on 43 pages is the same class as the stale auth.min.js that once made the HOMEPAGE run
old code.
The gate now enumerates git ls-files '*.html' and **keys every asset by its path from the
repository root**, because a sub-directory page writes ../../styles.min.css for the file a root
page calls styles.min.css — treating those as two assets would have split every pin and every
token check down the middle even after the walk was fixed. 266 pages scanned, up from 179;
118 pages retokened; version pins still exempt, since hashing those blanks the cockpits fail-closed.
B — the 25 network/** protocol explainers had no social or structured metadata at all.
Modbus TCP and RTU, BACnet/IP and MS/TP, OPC UA, DNP3, PROFINET, EtherCAT, EtherNet/IP, TLS, mTLS,
was almost entirely hidden), network-visualization-hub 5 px, network-compare 4 px; the last two
had been raised by that sweep and still did not clear the bar.
And two rules were reporting the browser working correctly. missing-image raised 1,255 findings
across 69 pages, every one a loading="lazy" image below the fold — not fetched because it is not
needed yet. Proof rather than argument: 20 incomplete images on articles.html before scrolling,
zero after. prose-font-size counted site chrome — the cookie-consent sentence and the footer
copyright line appear on every page at their own size on purpose, so every page carried a finding it
could never legitimately fix.
Not acted on, and why: large-blank-gap findings on article pages sit inside
[data-rz-scrolly] (article-16's root spans y 7577–10453; the reported gaps start at 8416) — that
runway is the mechanism, and only the full runner's verifyScrollingGaps() can judge it;
document-overflow is entirely archive and build output, including .qa-screens/overflow-check.html,
which overflows because it is the fixture for that check.
A cache-token drift, caught before it shipped
Adding the section picker changed js/rz-design-studio.js and its stylesheet, and the two existing
adopters — datahallAI.html and dc-conventional.html — still referenced them at ?v=1.130.0 and
?v=2.0.0. Both would have kept serving the previous file from cache while the six new pages served
the new one: the same asset live under three URLs. All nine references now carry one token.
v3.3.2PATCH
A retraction, and the flow-direction invariant that does hold
v2.18.0 shipped a MONITOR on this line, printed on every ship-gate run since:
> 2,880 of 9,630 tagged lines carry a flow class opposing their declared direction.
That number was wrong, and the error was mine. It assumed the fR / fD classes mean forward
and fL / fU mean reverse, so every leftward-drawn pipe whose logical flow is forward counted as
a defect. Re-measured against the geometry the lines actually carry, the pairing is exact and has no
exceptions at all:
class
direction the line is drawn
count
fD
downward
3,744
fR
rightward
1,512
fL
leftward
1,494
exceptions
0
The two attributes are orthogonal and both were right the whole time. The class names which way
the dashes travel across the screen; data-direction names the logical from-to of the line in
the process. A pipe drawn right-to-left whose flow is logically forward correctly carries fL.
What replaces it
The monitor is gone and the invariant that does hold is now a gate, checked at runtime across every
diagram: a line's flow class must match the direction its own endpoints run. That catches the
defect the monitor was reaching for — dashes travelling backwards up their own pipe, which on a
mimic reads as flow going the wrong way — without inventing a conflict between two attributes that
never disagreed.
6,750 tagged lines are checked. Proven red by injecting one reversed class into a detached copy: the
gate names the line, the view, the class, and the direction the line is actually drawn.
Why this is in the changelog rather than quietly deleted
A wrong number printed on every gate run is worse than no number, and a number that is retracted
without saying so is worse still. The retraction is recorded in the four places the original was
published: the gate's own output, this changelog, the plan file and the project memory.
v3.3.1PATCH
A gate for the hall plan — and it caught a duplicate CDU on its first run
Three owner rejections in a row on datahallAI.html were the same class of defect, and **no gate
saw any of them**: the mimic renders, the engine tests pass, the numbers are traceable, and the
drawing still puts the CDUs on the wrong wall, animates nothing out of them, prints one temperature
for a twenty-row hall and leaves the CRAH banks unlabelled. A drawing can be entirely correct by
every existing check and still say the wrong thing.
tools/test-datahall-ai-hall-plan.mjs asserts what the owner actually asked for, measured on the
rendered DOM and counted against the engine at run time (never a literal, so a basis change
moves the test with it rather than breaking it):
one CDU glyph per installed unit — no aggregation divisor;
the CDUs stand between the two banks, i.e. in the cross aisle, at the end of the rows they serve;
every row carries its own flow animation, not just the two header trunks;
every aisle reports temperature at several points, in both banks, all of them reading;
the points along one aisle do not all print the same number;
every CRAH bank states how many units it stands for.
Run against 01488121 — the build the owner was looking at — it reports 7 broken contracts.
Against the fix, none. Wired into ship-gate.sh.
What it caught
The hall had 56 CDU targets for 55 installed units. The extra one was a legacy "mechanical bank
chip" at x=752 that wrapped a second data-rz-equipment="cdu:1" around itself — a clickable CDU-01
sitting 190 px from where CDU-01 actually stands, opening its HMI from a place no CDU occupies.
Both chips existed for "the counts a per-glyph drawing cannot show", and after v3.3.0 the drawing
shows them: the CDU count stands on its own gallery, and every CRAH bank carries its unit count on
its face. The CDU chip is removed; the CRAH one stays as a single hall-level summary.
standarization/DATAHALL_AI_STANDARD.md gains the hall-plan contract: where each band stands and
why, the flow-animation binding, the aisle-instrumentation rule (and the trap that three readouts
sharing one RZ_SIM seed key print one number three times), and how to measure legibility here —
effective px rather than the font-size attribute, getBoundingClientRect() rather than
getBBox(), which ignores transform and reported 34 collisions where the reader sees 2.
v3.3.0MINOR
The CDUs move to the end of the rows they serve, and the hall says what it is doing
Owner, on the live drawing: *"Kok masih sama? Nggak proper positioning rack, cdu dan jumlah dan
aliran simulasi air cdu ke row tidak ada. Tidak ada crah dll", then "No hac temp indication in
several point across row, and cdu weird positioning"*.
The first answer was not in any diff. Production serves origin/main, and two releases sat
unpushed — v2.19.0 and the whole v3.0.0 hall rebasing. Nothing the owner was looking at had ever
been deployed. Deploy lag looks exactly like a broken fix; check
git rev-parse origin/main HEAD and curl the live URL for a marker before diagnosing a rendering
complaint. What follows is what was still open once it was deployed.
CDU positioning
A CoolIT CHx1000 is an end-of-row CDU. The drawing put four galleries on the short walls —
which are the ends of the cross aisle, not the ends of the rows. Rows run top to bottom on this
plan, so a row's end is the cross aisle: a row's coolant used to leave the drawing sideways,
travel the length of the hall and come back. All 55 installed CDUs now stand in the cross aisle,
spread across the same span the rows occupy, and the row directly above or below each one drops
straight into it. One glyph per installed unit — no aggregation divisor, so the count on the
drawing is the count in the registry.
The water, animated, at the engine's own rate
Three animations existed on this plan: two header trunks and one blink. The twenty row branches
were static and nothing moved out of a CDU at all. The supply path now flows end to end — CDU band,
row drop, row riser, and the return — 61 animations in place of 3. The dash period is not
decorative: it comes from the row flow the engine publishes (tcsFlowRackLpm × racksPerRow, clamped
to 1.2–6 s), so a row moving more litres a minute visibly moves faster. Motion that encodes a
number is a reading, not decoration.
CRAH
They were drawn. They just said nothing: twenty banks whose only number was ≈ 4.5 units, which
reads as noise. A bank is an integer of real machines, so the remainder is spread one per bank and
each bank now carries its own count on its face (T01 ×5). The owner's other question — *"memang
nggak perlu crah utk hot aislenya?"* — had no answer anywhere on the drawing: each contained aisle
now shows its return path overhead to the CRAH wall it faces, drawn as a path because a ceiling
plenum takes no floor.
HAC temperature, and the legibility floor
There was one readout pair per containment pair, parked in the cross aisle, bank A only: five
numbers for a hall of twenty rows, none of them standing in the aisle it measured. Each aisle now
carries three points at a quarter, a half and three quarters of the row, in both banks — 30 hot
and 30 cold, all reading. The id scheme (H{pair}{point}t) already allowed it; the loop bound in
upd() was capped at si<1 and the drawing emitted only b===0. The seed key now carries the
aisle and the point too, because all three used to read 'page1' and printed the same number
three times.
Then the sizes. audit-legibility reports this page rather than gating it, and it was reporting
584 sub-floor labels: the aisle temperatures rendered at 3.8 px against an 8.5 px floor, which
is why they read as absent. Raising them was previously blocked by overlap, so the space came
first — the bank captions turned 90° into the left margin (they had run across the top inside the
north CRAH band, under the row labels: three strings on one line), the CDU caption moved to the
clear strip east of the band, and the CRAH id and its count merged onto one line. With that room,
the readouts go to 8.5 px — the floor itself — row and bank ids to 6.8, and the **measured
text/text overlap count is unchanged at 2**, both of them pre-existing corridor-services labels.
Verified on production after deploy: 440 racks, 20 CRAH banks, 56 CDU glyphs, 61 animations,
30 + 30 aisle readouts with none left showing an em dash, no page errors, and the render-audit rule
set clean at 390/768/1440 in both themes.
v3.2.0MINOR
The missing voltage level, and the array that blanked the page
Owner, from the live single-line: *"ini gimana sih tulisannya 20kv dan 150kv secara bersamaan. Ini
harusnya memang 150kv dan ini juga aneh masak 150kv di step-down ke 400v pakai trafo."* Both halves
were right, and both are arithmetic.
check
number
facility at the worst weather bin
702 MVA
current if that were fed at 20 kV
18,887 A
the ring main the drawing showed carrying it
630 A — thirty times over
current at 150 kV
2,702 A
And a 2.5 MVA cast-resin unit substation is a 20 kV machine. At 150 kV the impulse withstand is
650–750 kV BIL, so the device the drawing implied cannot be built.
Why no gate caught it
electrical.voltageLL (400) was the only voltage in the whole engine — js/dcai-engine.js
labelled its own block "LV distribution" — so every kV string on the page was page-authored prose
inside a declared-basis escape hatch, and a search of tools/ found zero tests asserting any
voltage anywhere. A contradiction between two prose strings is not something a numeric gate can see.
The chain, now four levels
PLN 150 kV 4 × 250 MVA circuits (3 duty + 1) from two independent substations · 2,702 A
↓
150/20 kV 16 × 100 MVA ONAF — one per hall per feed
↓ secondary 2,887 A on a 3,150 A busbar · 41 % normal, 82 % carrying the hall alone
20 kV board fault 20.6 kA into 31.5 kA gear · BUS TIE NORMALLY OPEN, because two mains
↓ paralleled would deliver 41 kA and exceed it
20 kV feeders 5 per hall per feed at 630 A, each carrying 8 unit substations
↓
20/0.4 kV 2.5 MVA drawing 72 A at 20 kV
Every figure is an engine parameter with its own basis hook, printed on a new MV & HV chain
card. The source boxes now read one voltage, taken from the engine, on all six drawings that carried
the contradiction.
Gated — the complaint made executable
tools/test-dcai-voltage-chain.mjs is new: three descending levels, **every drawn transformer ratio
must step exactly one of them** (a 150 kV → 400 V label fails), every rendered kV must be a
published level, a reading within 10 % of one, or a declared gear class, and no source box may state
two different voltages. Proven RED against the pre-fix tree, where it stops at the first assertion
because the engine published no level above 400 V at all.
Found while shipping: an array in the snapshot blanks the whole page
Publishing the three levels as a convenience array alongside the scalars took the entire cockpit to
AUTHORITY UNAVAILABLE with nothing logged anywhere. The registry generator digests a nested array
to a string ("[3 items] sha1:…"), the page's authority check compares typeof registry value
against typeof live value, and string !== object made datahallSnapshotValid() reject the whole
snapshot. This is the same class as the engine-version pin that once blanked eight cockpits: a
silent, total failure from a change that looked purely additive. The array is gone — the three
scalars already said everything it said — and tools/test-dcai-engine.mjs now walks every snapshot
branch and fails on a nested array, so the next one fails a gate instead of a page.
Also corrected
Four transformer mimic sites hardcoded 10 MVA against the engine's 2.5, and a VCB spec claimed
12.6 MVA where 20 kV × 630 A is 21.8. Both now read the engine.
Honest boundary
250 MVA per 150 kV circuit is an ADOPTED thermal rating for an ACSR double-circuit line, not a PLN
grid study: the page states what the intake requires, not that it has been granted. Sixteen
150 kV transformer bays plus four line bays is a very large switchyard, and that is what 628 MW at
150 kV costs. The single-line still draws the 20 kV level and below; the 150 kV switchyard is
published in figures and named on the source boxes, not yet drawn as its own band.
v3.1.0MINOR
One PDF shell, and a document can now be issued partial or whole
Owner backlog item (22): "export to PDF with a mature template, partial or whole", starting from
cdu-checklist.html but required across all the engines.
171 gensets modelled, 8 drawn. Tests derive every expectation from the snapshot.
Diagrams (WP3). SLD per hall on the 40 groups with the trunk loading DERIVED; overview genset
bank with the facility pool counted; hall mimic as 10 row strips of 88 (≈ today's node count, not
6,000 rack glyphs); rack architecture redrawn for GB300 (18 compute trays, 9 NVSwitch trays, 8 power
shelves — 5 duty, N+3, 4+4 unachievable); cooling P&ID as banks with the cliff panel printing
both sides of the economiser inequality (HTW required 37.0 °C vs achievable 37.0 °C, margin 0.0 K,
cliff 34 °C) in place of a dry cooler that used to leave fluid colder than the air; floor plans and
the building isometric relabelled from the engine.
Documents (WP5). BoD drawer, BoD PDF, Tech Spec and FAQ describe the plant the engine computes
— the COP is printed with its Carnot-fraction derivation, the busway with its loading, the gensets
as a facility pool of 4 MW machines (the 2.75 MW class was rejected by arithmetic: >200 sets). The
PDF's per-rack table became per-group. The GB200 basis survives only as a labelled retired
reference and in the platform selector as Retired GB200 split-domain basis.
Registry. R8 "every parameter rendered or declared internal" flipped STRICT in this
commit, as its v1.136.0 flip condition required: 171 of 200 read by the page, 29 declared internal
with a written reason each (ambiguous-leaf twins the adapter reads under an alias, consistency
flags the gate asserts, per-bin values, null-while-free-cooling planes).
Honest boundaries
The building grid is still the retired-basis 72 × 48 m plot and says so on the floor plan: two
62 × 31 m halls and 270 UPS frames per hall do not fit the drawn rooms, and re-gridding the site is
its own change. The Network tab draws 27 representative racks and labels them as such (Track A §A7
redesigns it). The AI page is not yet in the coverage gate as a monitor; the coverage tool is bound
to the Conventional registry. 142 kW per rack is ADOPTED (NVIDIA publishes no rack power); annual
figures are ASSUMED (no TMY). Nothing was tuned: the PUE is what the declared terms produce.
Execution note
Six Opus agents were dispatched in worktrees and all six were terminated by a session rate limit
mid-task, on branches cut from main rather than the switched page. Their partial work was merged
one package at a time and completed by hand; the record is in the plan file.
Standards: ACCURACY_VALIDATION.md (Rule 5 rewritten for GB300 vocabulary; acceptance list
re-pinned), DATAHALL_AI_STANDARD.md (capacity table, LV grouping), manuals and PRD rewritten
against the snapshot. Ship gate: 57 gates; probe-accuracy-validation re-pinned to the snapshot and measured at 82/82 (AI-Test-3 inverted: at GB300 the page must NOT say rack-pos or 66 kW and MUST say 142 kW; TS-AI and BoD-AI assert the derived COP is printed with its Carnot derivation).
v1.136.0MINOR
Added — a GB300 / 500 MW basis engine for the AI campus, built alongside the retired GB200 one
The owner asked (2026-09-05) for the AI/HPC facility at **300–500 MW IT, four halls, NVL72-class
racks at 100–140 kW or newer, and ruled that 500 MW means rack IT**, that the cooling basis
should be "the best", and that nothing may be invented to reach a PUE. That is a different basis,
not a parameter change, so it ships as a new engine pair — js/dcai-model.js (authored leaves,
every one with a // source: line and an evidence class) and js/dcai-engine.js (pure functions,
one frozen snapshot) — while js/datahall-model.js + js/datahall-calculations.js stay
byte-frozen under the ship gate and keep proving 57/57 against their own worked examples. They are
retired, the way conv-engine.js retired the 1.85 MW hall. datahallAI.html still loads them;
switching the page is the next ship (Track A §A2b) because it touches ~150 literal sites and ten
gate fixtures.
Basis: 880 racks/hall × 4 × 142 kW = 499.84 MW rack IT; fabric (7,040 switches at an
ASSUMED 4 kW), OOB and a 2 % storage/management allowance take the total IT to 539.05 MW.
The ship workflow now invokes the full browser Design Studio regression so all three document
selections, distinct bodies/TOCs, current-versus-study scope and export flow remain release-blocking.
Final strict site gates and release evidence are recorded by the ship workflow for this revision.
v1.134.12PATCH
Accuracy gate follows the canonical basis drawer and reconciles equal scopes
Fixed
Migrated the Conventional and Data Hall accuracy probes from retired page-local drawer selectors to the canonical data-basis-param / #rz-basis-drawer contract; authored parameters are checked for value, scope, source and evidence without inventing a formula they do not have.
Corrected cross-page IT reconciliation to compare the 30,000 kW campus roll-up with four selected halls at 7.50 MW each; the gate reads hall count and both scope values from CONV_CALC.snapshot, so a future scenario change cannot silently reintroduce a hard-coded 30 MW assertion.
Preserved the two intentional page-local operational drawers (state and margin) while proving the three engine-backed Data Hall KPIs use the shared registry drawer.
Removed the retired 1,850 kW scenario and dependent 2,682.5 kW, 58.1 L/s and 45,900 L values from active Conventional overview fallbacks, generated Tech Spec narrative, PRD and Manual. Current operation now remains 30,000 kW IT / 43,500 kW facility / 943.0 L/s CHW / 744,144 L usable fuel everywhere.
Replaced global PDF string searches with labeled output-row assertions and added PRD/Manual parity checks, closing the false-green path in which stale narrative text could satisfy the accuracy gate.
Removed the remaining 99.98% uptime placeholder from the overview sidebar and basis drawer; YTD uptime now fails closed as UNAVAILABLE until an authenticated downtime event log exists.
Corrected the last CHWS first-paint fallback from 7.2°C to 19.4°C and gated all 23 duplicated current-value KPI, callout and sidebar surfaces against the governed snapshot.
Corrected the generated WUE worked-calculation source from the retired 37 L/min note to the labeled 600.0 L/min campus equivalent, bound its provenance to CONV_CALC.snapshot.water.flow_lpm_for_wue, and added dedicated output and source assertions.
Standardized the Design Studio scope contract on the implemented current-plus-study identifier; the 40 MW IT / 58 MW facility case is a governed planning-study comparison, while the adopted current snapshot remains 30 MW IT / 43.5 MW facility.
Changed the UNAVAILABLE uptime drawer output from healthy green to an explicit amber unavailable state and strengthened Manual/PRD retirement checks for 1,850 kW, both 58.1/58.2 L/s rounding variants, 37 L/min, 45,900 L and 99.98% through adversarial fixtures.
Split first-paint and runtime validation: the accuracy gate now parses raw cockpit markup before any script runs, then independently verifies all 23 engine-updated duplicate surfaces after initialization.
Regenerated the Conventional parameter registry after the consumer inventory changed and recorded every protected Manual rebaseline as an explicit old-to-new telemetry-document replacement contract.
Documentation
Updated the Conventional operations standard and Obsidian mirror to the adopted 4 × 7,500 kW operating scenario and recorded that browser probes must follow canonical component IDs, compare like-for-like engineering scopes, use the exact current-plus-study contract and independently gate runtime, fallback, generated and public-document consumers.
Verification
Accuracy validation: 83 passed, 0 failed.
Release ship gate: 41/41 passed.
Canonical browser audits: responsive layout, light/dark coverage, WCAG critical/serious, interactions, chart provenance, page access gates and hero-fit all clean.
v1.134.11PATCH
Both reported backlogs closed and gated: 77 unasserted → 0, 27 unrendered → 0
Two figures had been sitting in the registry gate's REPORTED line for five releases. Chasing
them found that **one was a bad measurement and the other was two different problems counted
together.**
"77 parameters with no gate asserting them" was mostly the measurement. The tests field
only grepped gate source for the path token, so it could not see the formula gate evaluating 51
of them, or the provenance rule covering every authored constant, or a drawer hook, or a
declared distribution. An assertion is an assertion whether it names the path in source or
reaches it through the registry. Corrected, the figure fell from 77 to 13 — and all thirteen
were TEXT parameters the numeric provenance rule could never reach.
So they got real assertions: evidence-class values must be terms from the taxonomy (a first
regex /_evidence_class$/ missed meta.evidence_class, whose leaf is dot-separated — the one
label the check did not cover); meta.version must be semver and agree with the registry
header; the basis and study document pointers must name files that exist; the active
scenario must be one the engine declares, and its label must be that scenario's own label; the
adopted chiller type must follow the study's heat-rejection type — the conflict that took two
releases to settle is an assertion now; hall.chillers_allocated must stay null and its reason
must be a real sentence; and the design duty must not be below the load being carried.
Now 0, and gated (R7). A new parameter with nothing asserting it does not ship.
"27 parameters no cockpit renders" was two problems in one number. Some were genuinely
missing from the screens, and they were the ones that matter most: the coil approach and the
supply-path mixing that decide whether a 25.4 °C supply is credible, the **chiller specific
power the entire COP figure rests on, the cooling-tower range, the plant design duty**,
the UPS module rating and modules per system the loading percentages are measured against,
the hall's design capacity, utilisation and cabinet ratings, the fire tank level and stored
volume, the fuel specific consumption. All sixteen are on the cockpits now, each a basis hook.
The other thirteen were never meant to be rendered — unrounded *_exact twins kept so identity
tests compare without rounding noise, a legacy site.it_design_kw alias that predates the
campus model, document pointers, and branch-level evidence labels that each parameter already
carries more specifically. They carry display: internal **with a written reason of at least 40
characters**.
Now 0, and gated (R8). A parameter that is neither rendered nor declared internal does not
ship — and "internal" without a reason is refused.
The drawer gate rejected four of these hooks before they shipped, each the same defect in
miniature — a parameter attached to an element that renders something else. The hall facility
load was hooked to a line reading the hall's IT load; the selected cabinet peak to the wrapper
of the design-average cell; the cabinet design average to a cell reading "500 × 20 kW", two
numbers where a check can only guess which one is meant; the fuel specific consumption to a line
reading "744,144 L ÷ 15,503 L/hr", neither of them the rate. Each got its own element rather
than a looser check. **A hook must point at the element that renders exactly the value it
Auth security: remove offline-root from public bundle + narrow retry — Codex review H-1/H-2
Security / Fixed
Removed offline-root recovery from the public bundle (Codex review H-1). Shipping any recovery
passphrase/hash in client JS is a published credential — it was readable (the passphrase was even in a
code comment), guessable/crackable, and unlocked client-side root with no outage (it also fired on ordinary
invalid_credentials). offlineRoot() is now a no-op stub; no passphrase/hash/salt remain in auth.js.
The need it addressed (a paused/blipping Supabase locking out login) is covered by the keep-alive workflow
(project won't pause) + the network retry in rz-supabase.js. Client-side role is presentation-gating only,
never a server authorization boundary.
Narrowed the Supabase fetch retry (Codex review H-2). rzFetch no longer allowlists the whole
/auth/v1/* namespace (which includes stateful writes: signup, OTP/recover, user updates). It now retries
only GET/HEAD and the password-grant login token POST (/auth/v1/token?...grant_type=password); everything
else is single-attempt, preventing double-submit of stateful Auth/REST writes. Verified unit 8/8.
v1.129.1PATCH
Governed agent harness standard
Added
Public long-running agent harness standard:standarization/AGENT_HARNESS_STANDARD.md now defines the frozen work-item contract, lifecycle, exact-revision evidence, independent acceptance, bounded autonomy, private-state containment, compact CLI checkpoint, Session Manager manual, and required change-and-lesson ledger.
Automatic cross-CLI handoff contract: Claude and Codex exchanges use watched, idempotent inbox envelopes with acknowledgements and exact component revision references, eliminating manual task-file drag-and-drop from the normal review loop.
Changed
Machine-checkable instructions now name the actual Session Manager test/manual gates and the canonical RZ ship gate. The standard no longer references proposed privacy scripts as if they already existed.
Provider unavailability is recorded independently from technical rejection: it cannot consume an implementation iteration or silently shrink the required reviewer quorum.
Candidate and accepted revisions are now explicitly separate in the durable checkpoint and owner-visible CLI indicator; PASSED remains unavailable until independent evaluation pins the exact accepted revision.
Fixed
Local agent-gateway authorization now gives the middleware-signed direct-peer verdict precedence over framework-synthesized forwarding headers, while forged client locality remains rejected. This preserves passwordless loopback operation without weakening reverse-proxy authentication.
Node 24 CJS/ESM export drift could make the CLI emit no machine token while the server derived a predictable credential from an empty identifier. Empty identifiers now fail closed, Linux uses a dependency-free machine ID source, other platforms accept the package default-export shape, and a validated random explicit token disables legacy machine-derived acceptance.
Turbopack could eliminate a direct runtime-secret property read from the packaged server, causing the current explicit token to fail despite being present in the child environment. The server now uses a fixed-name runtime-dynamic lookup, with a source regression guard and installed-package authorization probes.
A package could pass structural artifact policy yet fail to boot when a temporary Next.js output directory was baked into the server and pruned by the canonical allowlist. Release verification now requires a clean .build/next artifact and an installed-package /healthz probe in addition to pack validation.
A fresh Oh My Pi credential database existed as an empty file, so the OmniRoute adapter failed on its first write. Runtime setup now boots the official local auth-broker lifecycle before adapter configuration and verifies the provider record without exposing credential values.
The OmniRoute postinstall check used the legacy unsuffixed wreq-js native filename and warned even while the shipped GNU-libc binary loaded correctly. Candidate resolution now recognizes GNU, musl, MSVC, and legacy package names, with cross-platform regression coverage.
Documentation parity is restored across the public standard, RZ changelog/version/cache artifacts, Session Manager manual, private cross-CLI handoff, and owner knowledge vault.
Verification
Session Manager: 375 unit/integration tests pass, including lifecycle, authorization, bridge, redaction, manual parity, and recovery contracts; the 375 px manual view and desktop shell remain intact.
OmniRoute: 41 focused authorization/CLI tests plus 10 postinstall regression tests pass; the production service binds to 127.0.0.1:20128, and health, dashboard, landing, and root-follow probes return HTTP 200.
OmniRoute credential boundary: the installed package accepts the current explicit token; four former machine/empty-ID variants return 401 and a forged-locality request returns 403. Oh My Pi and Letta setup/status return 200 with OmniRoute present; model discovery returns 123 entries and an inference probe returns 200.
OmniRoute artifact: the canonical package contains 21,017 policy-approved entries (166,276,290 bytes packed; 767 MB unpacked), installs successfully, boots from .build/next, and exposes the required runtime manifests.
RZ website: the complete task verify ship gate passes all 13 blocking groups for this release.
v1.129.0MINOR
Telemetry cockpit PRDs and operator manuals
Added
Twelve public engineering documents for six telemetry cockpits: separate, implementation-traceable
PRDs and technical Manuals for EPMS Telemetry, AI/HPC Data Hall, Conventional Data Center, Data Hall
SCADA, CDU Mini-BMS, and the RZ cockpit visual prototype. Every PRD follows the mandatory 11-section
contract; every Manual records engineering basis, inputs, formulas or explicit fixture provenance,
worked examples, standards, glossary, assumptions, and limitations.
Two additive documentation routes on every cockpit: visible PRD and Manual links use each
cockpit's existing control style without changing its authorization gate, scripts, or operating flow.
Public documentation hubs and discovery coverage: a new prd/ hub, six paired entries in the
Manual hub, twelve search records, and generator support for PRDs in sitemap.xml, llms.txt, and
Adversarial re-audit of every prior owner review comment against the live code
(owner: "semua comment saya tidak di eksekusi dengan proper"). 17 of 19 items
verified genuinely engine-wired; 2 real defects of the "shipped but degraded"
class were found and fixed. Full ledger: dcmoc/OWNER_COMMENT_AUDIT.md.
Fixed
Strategic Planning — single-source lock was conditional (StrategicPlanningDashboard.tsx).
Total Land Area / Grid Capacity / Climate / Target PUE were only disabledwhen a
site fed them, so with no site selected they degraded to editable phantom inputs showing
seed values (10000 m² / 20 MW) — contradicting the page's own "LOCKED (single source)"
banner. Now always read-only mirrors; with no site the page shows a clear empty state
("Select a site to run feasibility → Site Intelligence") instead of a fake editable seed.
Rack form factor was a fake CAPEX cost driver (CapexEngine.ts). rackFormFactor
(std42u 1.0 / tall48u 0.90 / OCP 1.15, sourced EIA-310 / OCP Open Rack v3) only fed the
reported floorSpace metric — switching rack form changed reported m² but $0 of CAPEX,
though v1.115.92 claimed it a "real cost driver". Now multiplied into the space-driven
building (shell + civil) category, so 42U/48U/OCP move the CAPEX total as intended.
Added
DC Incidents — Case Library (dc-incidents.html), a root-only knowledge base of major data-center / cloud incidents worldwide. Ranked hub (transparent magnitude composite: blast radius 35% · users 25% · financial 20% · duration 20%) → each incident opens a dedicated dossier page (incident-<slug>.html) with a full sequence of events (SOE), root cause, correction of errors (COE), contributing factors, lessons learnt, engineering improvements, technical deep-dive, and a provenance-mandatory reference list (every fact traces to a public post-incident report).
Data-driven build:data/incidents/*.json corpus → tools/build-incidents.py renders the hub + all incident pages from one audit-passing template (gate markup, dark palette, mobile patch, version stamp, cookie consent). New ship gate tools/test-incidents-corpus.mjs rejects any incident missing sourcing/SOE/COE/≥2 references.
Gating + landing padlock: root-only page-access feature flag dc-incidents; enforceTierFeatureAccess('dc-incidents') on every page; /dc-incidents.html added to ROOT_ONLY_PATHS; a 🔒 DC Incidents entry on the landing-page menu. Anonymous visitors see the locked overlay; root accounts see the content (verified headless).
Phase A reference incident: AWS S3 US-EAST-1 (28 Feb 2017) — the archetype of COE/SOE post-incident culture — hand-built from the official AWS postmortem. Gated pages are intentionally excluded from the public sitemap / llms / search-index. Gates green: page-gates, dark-coverage, responsive-layout, mobile-responsive, version-stamp, script-tags, incidents-corpus.
Operations OPEX breakdown now recomputes when the deferred rz-engine.min.js finishes loading — the models.opex.fullBreakdown memo gates on the useEngineReady() signal (same class of fix as the WS2 Financial-statements engine tick). Previously, if the Operations page mounted from persisted state before the engine script loaded, the breakdown could stay empty until an input changed. End-to-end verified all four OPEX-program surfaces render with 0 console errors (Operations breakdown · Finance P&L/Balance-Sheet/Sankey · Maintenance 5-strategy deep-dive · CAPEX read-only mirrors).
Finishes the CAPEX Assumptions input-dedup (WS4). Seven fields that were still editable in BOTH the CAPEX Assumptions drawer and Requirements → Infrastructure — Deep-Sea Water Cooling, Green Certification, Renewable Energy, Design Fee, PM Fee, Contingency, Project Year — are now read-only mirrors on the CAPEX page (Edit ↗ jumps to Requirements), matching the established Building/Fire/Substation/Power-Backup pattern. Requirements → Infrastructure is the sole editor (it already carries the richer editors — sustainability + solar/BESS sizing, the delivery-basis group, and Design Margin ⇄ Contingency binding at Growth & Availability 1.5). No CAPEX cost input is editable in two places anymore; the CAPEX drawer reflects one shared source. Removed the now-dead setDeepSea setter. walk 31/0, trace-parity 116/116, enum-coverage 25. The DCMOC OPEX program (WS0–WS4) is complete.
Maintenance-strategy deep-dive on Maintenance → Strategy. The planned-maintenance regime expanded from 2 (OEM-full / standard-annual) to the full 5-strategy worldwide taxonomy — Full OEM-compliant · Predictive/CBM · Hybrid (criticality-tiered) · Standard annual + compliance-only · Reactive/run-to-failure — researched from SFG20 / OEM RCM / NFPA 25/70B / Uptime O&M. A dropdown + per-strategy cards surface each strategy's PM frequency, task scope, contract-vs-in-house split, indicative in-house FTE/10 MW, availability delta and annual cost index, alongside the engine-computed PM labor-hours, site FTE and availability % for the current project.
Changed
DATA.maintenance.ops.pmRegime carries all 5 strategies (aligned key-for-key with DATA.opexModel.maintenanceStrategies): pmHoursMult scales PM labor demand in models.maintenance.opsHeadcount; failureMult scales the deferred-servicing failure exposure in models.maintenance.availabilityImpact — so the strategy choice flows through to staffing (Staffing page) and computed availability (Risk page) app-wide (predictive/CBM gives the best availability 99.982%, reactive the worst 99.968% at the reference 20 MW / Tier III). Store pmRegime enum + DowntimeCalculator widened to 5. Sourced. Engine 762/0, reference-parity 155/0, walk 31/0, trace-parity 116/116, calibration 19/0.
Detailed project financial statements on the Financial → Pro Forma (Full) tab, driven by the same engine cashflows the KPIs use (no re-computed or fabricated financials):
Income Statement (P&L) — per-year Revenue → OPEX → EBITDA → Depreciation (tax shield) → Taxable Income → Tax → Net Cash Flow (after-tax), with EBITDA/net margins. Labelled to match the engine's convention exactly (taxableIncome = max(0, EBITDA − depreciation), netIncome = EBITDA − tax = unlevered FCF; depreciation is a non-cash tax shield added back), so the statement reconciles line-to-line.
Balance Sheet — assets (net PP&E + cash) = liabilities (debt schedule from an adjustable debt-ratio + debt-rate) + equity (paid-in + retained earnings). Cash is the accounting identity plug (cumulative net income + D&A − principal), so the sheet balances by construction — a live check asserts a < $1 residual every year. Operating P&L is unlevered (project basis, matches NPV/IRR); interest is a financing memo.
Sankey — the actual money distribution for a stabilized year: Revenue → each OPEX group (from the WS1 per-country breakdown, scaled to the P&L OPEX) + EBITDA → Tax / Net Cash Flow. Depreciation (non-cash) is correctly excluded; renders only when the cash chain is positive, else an honest note.
Changed
Guarded the OPEX-breakdown call against the deferred-engine race (persisted CAPEX can mount the statements before window.RZEngine loads) with an engine-readiness tick, so the Sankey populates deterministically. DCMOC build green; walk 31/0, trace-parity 116/116.
v1.116.0MINOR
DCMOC — complete per-country OPEX model + maintenance-strategy taxonomy
Added
Complete per-country OPEX model. The Operations page previously showed only "Energy Cost 24h". It now renders a full annual operating-cost breakdown — grouped Energy / People / Contracts & Maintenance / Security / Compliance & Licenses / Connectivity / Real Estate & Tax / Overhead — with $/yr, % share and a per-line basis. New engine model models.opex.fullBreakdown (additive; totalAnnual untouched, all existing gates green) covers electricity, water, in-house manpower, outsourced M&E O&M, M&E preventive maintenance, ICT/network maintenance, outsourced cleaning, physical security (guards), security-system maintenance, software/BMS/DCIM licenses, certification renewals, annual permits, corporate internet/bandwidth, land lease, property & local tax, insurance, carbon levy and G&A overhead.
Per-country sourced cost factors (40 countries). A 3-agent research loop (deep-research → adversarial fact-check → merge) produced sourced values for the six genuinely location-variant factors — water tariff $/m³, industrial land lease $/m²·yr, property-tax %, enterprise DIA $/Mbps·mo, cleaning $/m²·yr and fully-loaded guard $/guard·yr — added to CountryProfile.opex and regenerated into DATA.countries. Each cell carries a source_url+quote+confidence provenance row in tools/dc-corpus/opex-research/facts.jsonl; screening-confidence cells (benchmark-extrapolated) are logged. The fact-checker corrected NZ/CA water (USD conversion), ZA internet and NG/KE/ZA/CO guard clamping.
Worldwide maintenance-strategy taxonomy + selector.DATA.opexModel.maintenanceStrategies (full OEM-compliant · standard annual + compliance-only · reactive/run-to-failure · predictive/CBM · hybrid), each with contract-vs-in-house split, in-house FTE/10MW, availability delta and cost index (SFG20 / OEM RCM / NFPA / Uptime). A dropdown on the Operations OPEX card drives the contract / PM / manpower / availability mix coherently (highest-cost OEM = best availability; reactive = lowest cost, worst availability).
Changed
Every new DATA value carries a DATA.sources entry (opexModel, opexModel.maintenanceStrategies, countries.opex). Engine 762/0, value-bindings 85/0, reference-parity 155/0, walk 31/0, trace-parity 116/116, enum-coverage 25, model-calibration 19/0.
v1.115.94PATCH
DCMOC — CAPEX infra dedup complete: Power Backup single-owner + fuel AutoField at its home
Changed
Finishes the CAPEX Assumptions dedup. The Power Backup block (UPS / generator / fuel autonomy) was the last section editable in both the CAPEX Assumptions tab and Requirements → Infrastructure. It's now read-only mirrors on the CAPEX page, and the fuel-autonomy AUTO/override field moved to its canonical home in Requirements → Infrastructure → Systems (AUTO from the tier's Uptime backup band; tick to override) — so the auto-with-override lives where the field is owned, with exactly one editor. No CAPEX infrastructure parameter is editable in two places anymore. walk 31/0, trace-parity 116/116.
v1.115.93PATCH
DCMOC — CAPEX infra dedup, cont.: Fire + Substation now read-only mirrors
Changed
Continuing the CAPEX Assumptions dedup — the Fire Protection (suppression + alarm) and Substation & Grid (front-of-meter scope + utility rate) sections were editable in both the CAPEX Assumptions tab and Requirements → Infrastructure, writing the same store. They're now read-only mirrors on the CAPEX page (value + "Edit in Requirements ↗"), so each has one owner. Editing them in Requirements still flows to the CAPEX total. walk 31/0, trace-parity 116/116, enum 25.
> Still remaining: the Power Backup block (UPS / generator / fuel) is dual-editable — its dedup also moves the fuel AUTO/override field to Requirements, so it's a careful follow-up.
v1.115.92PATCH
DCMOC — rack-form is now a real cost driver + AutoField extracted & rolled to CAPEX
Added
Rack Type (42U / 48U tall / OCP) now drives floor space — previously an inert control. Backed by engine DATA.requirements.rackFormFactor (48U tall packs ~14% more usable U in the same footprint → 0.90× floor; OCP Open Rack is 21″ wide + open busbar aisle → 1.15×), wired into the CAPEX floor-space calc. Picking a rack form now moves floor space (and its building cost) + shows a hint of the effect. Sourced (OCP OpenRack v3 21″; EIA-310 48U vs 42U).
CAPEX "Fuel Storage" now auto-derives with override — the AUTO/OVERRIDE field primitive (previously embedded in the Financial page) is extracted to a shared components/ui/AutoField.tsx and applied to fuel-storage hours: AUTO follows the Uptime backup-autonomy band for the tier (II 48h · III 72h · IV 96h); tick to override for remote/hurricane sites. "Auto tapi bisa override."
Changed
Engine chain rebuilt (762/0, catalog 234 fns / 159 sources); gates green (walk 31/0, trace-parity 116/116, enum 25, reference-parity 155/0). Completes Workstream C of the IA/auto-derive program.
v1.115.91PATCH
DCMOC — Time-to-COD tooltip + Target Rack Density auto-derives from architecture
Added
"Time to COD" now explains itself — a hint on the Requirements field: months from today to the Commercial Operation Date (the date the facility starts revenue service), derived from Target COD in Project Overview; drives schedule pressure, phasing + interest-during-construction.
Target Rack Density (Max) auto-derives from the selected AI architecture — picking an architecture (GB200 ≈ 132 kW/rack peak, etc.) auto-fills the target ceiling; the field shows an AUTO chip when it matches the architecture, or a one-click "↺ match" button + "overridden" note when you've changed it. Owner: "target rack density bukannya auto dari architecture rack." Manual override still allowed. Pure frontend; walk 31/0.
> Note: rack-form (42U/48U/OCP) wired as a real cost driver + the AutoField roll-out to more CAPEX inputs are a follow-up (Workstream C, part 2).
v1.115.90PATCH
DCMOC — Site Intelligence: guided Auto-optimize + blink-highlight what to edit
Added
The Edit-Criteria drawer now shows you what to fix and can fix it. On open, the parameters that are dragging the site's score down (the drivers of every weak axis) pulse — so you're not guessing which of 36 fields to touch. A new "Auto-optimize criteria → best score" button solves each weak-axis driver to its best realistic value on the real score model (autoOptimizeSite, reusing the axis-lever bisection), applies them, highlights exactly what it changed, and reports the score move (e.g. 42 → 57 · 2 weak axes lifted · 6 criteria set to best). It never regresses, and is honest that some axes can't reach "Good" by criteria alone. Pure frontend (reuses the site scoring + lever solver); walk 31/0, site page 0 errors.
v1.115.89PATCH
DCMOC — Site Intelligence edit-criteria: per-parameter help + country auto-derive
Added
Every site criterion now explains itself. The Edit-Criteria drawer's 36 parameters each carry a plain-language "what it is + a worked example" tooltip (e.g. Cable Landings — number of submarine-cable landing stations serving the region; more = diverse international connectivity + lower latency. e.g. Singapore ~4, an inland site 0) — so the numbers and scales aren't a wall of unexplained values.
The categorical criteria now auto-derive from the per-country research tables (countryBaselineEnums): earthquake risk from the seismic zone, flood/cyclone/coastal risk from the natural-disaster table, construction-labor from the talent pool, fuel availability from the diesel table, and government support from the tax-incentive programs — shown as a cyan "baseline" chip (and named in the select placeholder) so far fewer fields sit blank. Editing still overrides to a "custom" value; the store stays unset (= engine baseline) until you do. Pure frontend; walk 31/0.
v1.115.88PATCH
DCMOC — CAPEX "Assumptions & Config" dedup: one owner per field
Changed
Removed the duplicated content from the CAPEX Assumptions & Config sub-tab so each thing has a single edit home and the wiring can't diverge:
The project-timeline Gantt (which the Construction page already owns, with the 4-level WBS + planned-vs-actual EVM) is gone from CAPEX — replaced by a compact "Construction timeline: N mo (L2 cost basis)" line + an "Open full schedule & EVM in Construction ↗" jump.
Building Type, Site Condition and Market Condition — which are (or now are) owned by Requirements → Infrastructure — are read-only mirrors on the CAPEX page (value + "Edit in Requirements ↗"), matching the existing shared-canonicals pattern. Market Condition was added to Requirements → Infrastructure so all three sit together.
Net: no field is editable in two places. Editing them in Requirements still flows through to the CAPEX total (same store). Pure frontend; walk 31/0, trace-parity 116/116.
> Note: several other infrastructure parameters (substation / UPS / generator / fuel / fire / floor / seismic) are still editable in both the CAPEX Assumptions tab and Requirements → Infrastructure — a broader dedup pass for a follow-up.
v1.115.87PATCH
DCMOC — alternative power tech wired into a real CAPEX/OPEX/CO₂ delta
Changed
The Alternative Power section is now a techno-economic comparison, not just a reference table. Each on-site power option (reciprocating engine, open/combined-cycle turbine, SOFC, SMR, geothermal, solar+BESS, diesel) now shows its CAPEX, annual OPEX and annual CO₂ sized to this project's facility load (IT × PUE, continuous prime) plus the Δ vs the diesel baseline — green when lower, red when higher. Backed by new engine DATA.fuelGen.altPowerTech.capexUsdPerKw + lcoeUsdPerKwh (Lazard LCOE+ 2024/25: gas-CC ~$76/MWh, nuclear ~$182/MWh; + NREL ATB + vendor est) with a DATA.sources entry. Example at 2.5 MW IT: SMR carries ~+$25M CAPEX but ~−$3M/yr OPEX and ~−19 kt/yr CO₂ vs diesel — the real high-capex/low-carbon-and-fuel tradeoff. Screening, not a procurement basis. Engine chain rebuilt (762/0, 158 sources); walk 31/0.
v1.115.86PATCH
DCMOC — polish round 3: card-consistency capstone
Changed
Aligned the Planned-Maintenance Compliance Regime card to its page's card idiom (added the shadow-sm dark:shadow-none its sibling cards use) — the last real in-page consistency gap on the session's new components. (Cross-page card differences are a pre-existing DCMOC pattern, not introduced here, and consolidating onto the shadcn <Card> would move against the dominant slate-card idiom, so left as-is.) walk 31/0; no engine change.
Completed the rz-* token sweep on the remaining session components: the Phased Financial Improvement Recommendations cards + icon (cyan→rz-info), and the CDU water-balance range sliders (accent-cyan-500→ the --rz-accent-info var) — the new UI now sits entirely on the token palette.
Added
The Alternative On-Site Power Technologies table highlights the row matching your current fuel-type selection ("your selection" chip + subtle tint), connecting the modelled choice to the broader technology landscape. Verified: selecting Fuel cell tags the SOFC row. No engine change; walk 31/0.
v1.115.84PATCH
DCMOC — IST fail→issue loop + UIUX token polish of the session's new components
Added
A failed IST scenario now auto-raises a linked open Cx issue (High severity), so a failed integrated test surfaces in the Issues & Punch list, the readiness guidance and the PDF — passing or clearing it closes the linked issue. Closes the loop from "test failed" to "tracked remediation."
Changed
Design-token polish across the session's new components (from a UIUX review). Introduced a shared StatusChip (rz-* semantic tokens) and routed every status/label badge through it so "pass" is rz-data, "fail" is rz-alert, and measured/info accents are rz-info site-wide — the IST tab, alt-power-tech maturity chips, and the pmRegime "Selected" pill were using raw emerald/rose/cyan/violet families that fought the palette. Also: PhasedFinancial negative states red-500→rz-alert; PageDescription + CDU + commissioning info accents →rz-info; the two new accordions standardized on ChevronDown; IST Pass/Fail tap targets raised to min-h-[32px]; and the one real a11y gap — the Phased Financial phase row (<tr onClick>) — is now keyboard-operable (role="button", tabIndex, Enter/Space handler, focus ring). No engine change; gates green (walk 31/0), no new console errors.
v1.115.83PATCH
DCMOC — IST scenarios in the Commissioning PDF export
Added
The Commissioning PDF now includes an "Integrated Systems Tests (L5)" section — each applicable IST scenario (redundancy-filtered) with its category, duration and Pass/Fail/pending status. Closes the gap where the IST scenarios drove the on-screen readiness but were absent from the exported report. Export probe 44/44, walk 31/0; no engine change.
v1.115.82PATCH
DCMOC — IST scenario pass/fail now drives the commissioning readiness index
Changed
The rich IST scenarios are now the single source for the ist readiness key. Marking scenarios Pass/Fail in the IST Scenarios tab feeds the engine readinessIndex (via checklistStats['ist'] = passed ÷ applicable, redundancy-filtered) and the overview per-level readiness bar + the Tests-passed tally — so the integrated-systems tests actually move operational readiness, not just a local count. The old shallow CX_CHECKLIST['ist'] accordion is removed from the Cx Checklist tab (replaced by a pointer to the IST Scenarios tab) so there is exactly one IST source — no double-count. Verified: a Pass on the IST tab shows "1/10 passed" on the checklist pointer and lifts the ist readiness. Pure frontend; gates green (walk 31/0, trace-parity 116/116).
v1.115.81PATCH
DCMOC — editable pass/fail tracking on the IST scenarios
Added
The IST Scenarios now carry editable Pass/Fail tracking, like the Cx checklist. Each integrated-systems-test scenario has a Pass / Fail / clear control (in its expanded body) and a status badge in the header; the tab shows a live tally — passed / failed / pending + a % pass progress bar. State persists in the shared cxTracking store under an ist-scn: namespace (no collision with the readiness-driving checklist ticks), survives reload, and is plan-mode-clean until touched. Pure frontend reuse of the existing tracking store; no engine change. Verified: marking Pass updates the tally and persists across reload. Walk 31/0.
v1.115.80PATCH
DCMOC — alternative on-site power technology research reference
Added
An "Alternative On-Site Power Technologies" reference on the Fuel & Generator page — closing out the A-slice ask ("deep research cari alternative tech lain"). A sourced screening comparison of the credible behind-the-meter options for an off-grid / prime-power data center beyond the reciprocating genset: reciprocating gas engine, open- and combined-cycle gas turbine, solid-oxide fuel cell (SOFC), small modular reactor (SMR), geothermal/EGS, solar+BESS, and diesel backup — each with electrical efficiency, operating CO₂ (kg/kWh), TRL, power density (MW/acre), deploy window, H₂-readiness, maturity, and a note. Backed by engine DATA.fuelGen.altPowerTech with a DATA.sources entry citing 2025-26 industry reporting (Goldman Sachs, DataCenterDynamics, Bloom Energy, GE; NuScale's 77 MWe NRC Standard Design Approval, May 2025). Reference/education only — collapsible, and explicitly not wired into the cost model (the fuel-type selector remains the modelled path). Engine chain rebuilt (762/0, 158 sources); walk 31/0.
v1.115.79PATCH
DCMOC — review-hardening of the power-source / fuel / water batch
Fixed
Carbon Scope-1 now uses the selected fuel's consumption rate (fuelTypeModel.effLPerKwh), not a fixed diesel 0.3 L/kWh. Previously fuel-cell (0.20) and biogas (0.32) selections computed their fuel volume — and therefore Scope-1 CO₂ — on diesel efficiency, giving a wrong carbon number.
The HVO cost/CO₂ comparison panel is now gated to fuelType === 'diesel'. HVO is a diesel substitute, and the panel's baseline is the diesel 2.68 kgCO₂/L factor — showing it for a gas/biogas selection produced a nonsensical saving computed against a non-diesel fuel volume.
Prime (off-grid, continuous) power no longer double-counts test fuel. With gensets running 8760 × utilisation, the periodic test schedule is a subset of continuous operation; the separate test-fuel line is now zeroed for prime (standby/hybrid keep it). Prime CO₂ corrects slightly (e.g. 20,236 → 20,207 t/yr at 2.5 MW).
Guarded the generator OPEX-breakdown percentage divisor against a zero total; removed a dead/misleadingly-named annualGlycolMakeupPlusWaterM3 field from models.water.coolingLoop; the IST scenario tab now renders unordered lists as <ul> (was <ol>) and guards the risk-badge style lookup.
All from an adversarial review of v1.115.72–.78. Engine chain rebuilt (762/0, bindings 85/0); gates green (walk 31/0, trace-parity 116/116, enum 25, fuel-gen probe 5/5).
v1.115.78PATCH
DCMOC — Commissioning Integrated Systems Test (IST
scenario deep-dive)
Added
A new "IST Scenarios" tab on the Commissioning page with the scripted Level-5 integrated-systems-test procedures the checklist was missing. Backed by engine DATA.commissioning.istScenarios (sourced: Uptime Tier Certification IST practice + NFPA 110/72/75, ISO 8528, IEC 62040-3, ASHRAE TC9.9). Ten scenarios, each with purpose, pre-conditions, scripted method steps, acceptance criteria, what to observe during the transient, the systems involved, duration, risk and the governing standard — specifically covering the tests requested:
Dual-source (A/B feed) transfer — either feed carries 100% with no IT drop.
Bus-tie breaker & interlock — tie closes on a source loss under a valid sync-check permissive; interlock blocks an unsafe close.
Mechanical redundancy degradation (N+2 → N) hotspot test — deliberately shed cooling units step-by-step to N, map where rack-inlet hotspots form, and observe the surviving plant's ramp-up response (fan/valve/flow) to hold setpoint. This is the headline test the owner asked for.
Cooling unit failover & ramp-up, black-building / pull-the-plug, generator step-load + fuel endurance, UPS transfer & autonomy, fire/EPO cause-and-effect, and BMS/network failover.
Scenarios are gated to the project's redundancy (appliesFrom): a N+1 design sees the single-path tests; the dual-source and N+2→N degradation tests unlock at 2N. The tab shows the in-scope count and estimated test-hours, and lists what unlocks at higher redundancy. Engine chain rebuilt (762/0, catalog 234 fns / 157 sources); gates green (walk 31/0, trace-parity 116/116, enum 25).
An always-visible "Improvement Recommendations" panel on the Phased Financial page. Previously the quantified levers were only reachable by clicking a red IRR/NPV/PI number; now they sit on the page as a ranked, comprehensive list. It combines:
Solved levers — the revenue-uplift and CAPEX-cut magnitudes that reach the 12% hurdle, computed by bisection on the same cashflow model (with the target $/kW·mo and resulting IRR).
Structural levers with the real on-page figures — capture the incentives already modelled (tax + grid $), reduce interest-during-construction (the live IDC $ over the build months), re-phase the build (defer later-phase CAPEX), and trim OPEX drivers (PUE/staffing/SLA) — each with a jump-to-tab.
A prominent "Auto-optimize revenue → 12% IRR" button that runs the existing deterministic bisection solver and previews the allowlisted revenue move before applying.
The panel adapts its framing to whether the program is above or below the hurdle, and is honest that structural levers are directional while revenue/CAPEX levers are solved. Pure frontend (reuses the optimizer + decision-explain); no engine change. Gates green (optimizer 8/8, financial-auto 9/9, walk 31/0).
v1.115.76PATCH
DCMOC — two planned-maintenance compliance regimes: OEM-full vs standard-annual
Added
Planned maintenance now has two compliance regimes on the Maintenance → Strategy tab, wired end-to-end:
OEM-compliant (full) — every OEM PM task at the manufacturer's interval: maximum reliability and warranty compliance, maximum manpower.
Standard (annual) — most tasks consolidated to annual, with only 1–2 critical systems (UPS batteries, gensets) kept at OEM frequency: far fewer PM labor-hours, at a ~12% higher failure exposure from deferred servicing.
Backed by engine DATA.maintenance.ops.pmRegime (pmHoursMult 1.0 vs 0.55, failureMult 1.0 vs 1.12, sourced). models.maintenance.opsHeadcount scales PM labor demand by the regime, and models.maintenance.availabilityImpact scales the failure rate — so choosing standard-annual honestly lowers labor-hours in Staffing while nudging the computed availability in Reliability/Risk. The selector shows the live delta and, importantly, is honest that at a small single-DC site the total headcount stays floor-bound by 24/7 emergency-response coverage (the saving lands on vendor/overtime labor cost; the FTE drop appears at larger campuses).
Held in the sim store inputs.pmRegime (default oem-full). Enum-coverage gate extended to pmRegime (2/2). Engine chain rebuilt (762/0, bindings 85/0, catalog 234 fns / 156 sources); gates green (walk 31/0, trace-parity 116/116, enum 25).
v1.115.75PATCH
DCMOC — per-page "what this page is for" descriptions, site-wide
Added
Every DCMOC page now explains itself. A single registry-driven PageDescription component is injected once in the Shell, above every tab's content, and renders a one-line "what this page shows + when to use it" note keyed by the active tab. Covers all ~48 tabs (Requirements, Capacity, Fuel & Generator, CDU, Architecture, CAPEX, Financial, Reliability, Operations/Staff/Maintenance, Carbon, Commissioning, Scenarios, Diagnostics, and the rest). No more guessing what a page is for. One component + one registry — not a per-page copy; adding a tab is a one-line registry entry. Gates green (walk 31/0); no engine change.
v1.115.74PATCH
DCMOC — Simulation KPIs + Cause-Effect levers get trace/diagnostic modals
Added
Click-to-trace diagnostics on the Simulation (Staff Model Config) KPIs. The four KPI cards — Internal Staff Cost, Vendor Labor Cost, Parts & Consumables, Hidden Turnover Loss — are now clickable and open a diagnostic modal that explains, from the SAME live model the card renders, how the number is built (headcount × labor multiplier, the vendor-premium strategy multiplier, the AQI consumables multiplier + filter-life, the turnover amortisation) plus quantified levers with jump-to-tab.
Cause-Effect Lever Map is now interactive and honest. Each lever row is clickable → the same diagnostic modal with its direction (raises/reduces/mixed) and effect. Fixed a real bug: the "Air → Direct Liquid Cooling" lever displayed a literal unfilled ~$X/yr placeholder — it now computes the actual annual energy saving from the live IT load at the PUE delta ($0.10/kWh screening) and shows the real PUE figures from the engine.
Changed
The shared DiagnosticModal gained an explain variant — a neutral cyan "Trace — how this is calculated" framing (vs the red "threshold breached" framing) so a healthy KPI breakdown isn't mislabelled as a fault. Reused, not re-implemented per card. Gates green (walk 31/0, trace-parity 116/116); no engine change.
v1.115.73PATCH
DCMOC — CDU liquid-cooling water & glycol consumption model
Added
Water & Glycol Balance section on the CDU / Liquid-Cooling page — liquid cooling was modelled for hydraulics and refrigerant but not for water use, which for an evaporative-tower site is the dominant sustainability cost. New engine model models.water.coolingLoop (backed by DATA.coolingWater, ASHRAE TC9.9 + cooling-tower practice, sourced) computes the full balance:
Technical coolant loop — closed water + glycol loop (~8 L/kW IT charge), user-set glycol fraction (20–35%), and annual makeup (~15%/yr of charge from leaks/drain-refill/service), split into water vs glycol.
Heat-rejection tower losses — evaporation (≈ latent heat of the rejected duty, IT×PUE × 1.4 L/kWh_th × wet-fraction), blowdown (evaporation ÷ (cycles-of-concentration − 1)), and drift (~0.5% of evaporation with modern eliminators).
Heat-rejection selector — Evaporative tower (max water) / Hybrid-adiabatic (~half) / Dry cooler (no tower water, but higher fan energy/PUE), plus glycol-% and cycles-of-concentration sliders.
Outputs annual m³ by stream, glycol makeup (flagged as chemical, not water), and WUE (L/kWh). Verified live: a 2.5 MW evaporative site ≈ 2.64 L/kWh WUE; switching to a dry cooler drops tower water to zero. The balance is also written into the CDU PDF export.
Engine chain rebuilt (terser + catalog 234 fns / 155 sources + value-bindings 85/0 + test-rz-engine 762/0); gates green (walk 31/0, trace-parity 116/116, enum-coverage 24). DATA.sources.coolingWater entry added (no economically-material literal left in the model body).
Power source + fuel type on the Fuel & Generator page — the model now covers three power topologies and six fuel/generation types, and they drive the whole app (not just a label):
Power source — Utility grid + standby gensets (gensets run ~grid-outage hours only), Prime power (off-grid) for sites like Ireland where the DNO has no grid capacity so on-site generation is the primary continuous supply (gensets run 8760h × 0.85 utilisation), and Hybrid (grid + on-site sharing ~55% of the year).
Fuel / generation type — Diesel (EN590/ULSD), HVO-100 (renewable diesel, −90% lifecycle CO₂ per EN 15940), Natural gas (reciprocating gas engine, −24% CO₂), Solar + BESS hybrid (shaves genset run-hours ~30%), Fuel cell (SOFC on gas), and Biogas/RNG (near carbon-neutral). Each carries efficiency, CO₂/unit, cost premium and run-hours factors from engine DATA.fuelGen.powerSourceModel / fuelTypeModel (ISO 8528 duty ratings; EN 15940; DATA.sources).
The selection is held in the simulation store so every related surface reacts:
Fuel & Generator — annual fuel, consumption, and CO₂ scale with the run-hours implied by the topology (verified live: standby ≈ 14 h/yr → prime ≈ 7,446 h/yr; prime CO₂ ≈ 20,236 t/yr, prime + HVO ≈ 2,061 t/yr = −90%).
CAPEX — prime removes the utility substation + grid-connection front-of-meter cost entirely (off-grid) and oversizes the gensets (prime-rated); hybrid at ~half grid cost.
Carbon & ESG — Scope-1 generator emissions now use the topology run-hours and the fuel-type CO₂ factor instead of a fixed 200h diesel approximation.
Architecture SLD — prime power relabels the utility intake to an off-grid On-Site Plant node and the generation lane from STANDBY to PRIME POWER PLANT (continuous), with a paralleling-switchgear stage in place of the ATS.
audit-dcmoc-enum-coverage extended to gate powerSource (3/3) and fuelType (6/6) against their engine DATA maps (their keys live in JS, so tsc can't catch drift). Engine chain rebuilt (terser + catalog + value-bindings 85/0 + test-rz-engine 762/0); gates green (walk 31/0, trace-parity 116/116, arch-diagram 7/7, reference-parity 155/0).
v1.115.71PATCH
DCMOC — CAPEX-total trace now decomposes the real cost stack
Fixed
The CAPEX-total trace was inaccurate — it showed the total as computed from just "IT Load + Contingency %", as if CAPEX were a contingency assumption rather than a calculation. It now decomposes into the REAL components that sum to the total: Hard Cost (Σ all disciplines — electrical/cooling/UPS/building/generator/network/fire/seismic/security/commissioning/testing/permits × their multipliers) + Soft Costs (design + PM) + Contingency + Front-of-Meter (substation/grid/switchgear). Four new drill-down trace nodes (capex.hardTotal / capex.softTotal / capex.contingencyAmt / capex.fomAmt) read the engine's own costs/softCosts/contingency/fomTotal, so clicking through the trace reaches the actual cost calculation. Value-bindings + trace-parity 116/116 green.
v1.115.70PATCH
DCMOC — Trace popover Collapse/Expand-all fixed + made reusable
Fixed
The Trace-Number popover "Collapse all / Expand all" toggle now actually works per node. It previously flipped a single boolean that showed/hid one flat list (owner: "no effect"). Replaced with a reusable per-node collapsible tree: new hooks/useCollapsibleTree.ts (tracks expanded ids, toggle/expandAll/collapseAll/allExpanded) + components/ui/CollapsibleTree.tsx (nested renderer with per-node chevrons + a renderNode callback). The popover's dependency tree is now genuinely expandable node-by-node, Expand-all/Collapse-all act on every node, and filtering still falls back to a flat matched list. The primitive is reusable by any tree UI (owner: not hardcoded per-use). Trace-parity 116/116 unaffected.
CAPEX Engine no longer duplicates Requirements inputs. The Project-Parameters panel had editable Region/Country, IT Capacity, Redundancy, Cooling, and Rack Density selects — the exact shared canonicals a banner already said Requirements owns, so the page could fork them (two owners of one value). They're now READ-ONLY derived cards (value + "requirements" chip + "Edit ↗" jump to Requirements); only CAPEX-only assumptions (building type, UPS/generator, fuel storage, utility rate, green cert, contingency, fees, FOM quality selects, substation) stay editable here. One source of truth (owner: "jangan ada duplicate").
Setup Wizard removed — the Scenario-Controls "Wizard" modal (Region / Reliability Tier / Technical Systems / Maint. Strategy) duplicated Requirements configuration; deleted the trigger + mount so there's a single configuration path.
Gates: walk 31/0 · trace-parity 116/116. No engine change.
Generators showed 115% utilization — impossible for a self-sized component. Root cause: the capacity component table counted gensets from IT load (ceil(IT/2000), engine equipScale) but divided FACILITY load (IT × PUE) by that capacity → utilization = PUE ≈ 115–127%. Generators back the WHOLE facility (IT + cooling + losses), so they must be sized on facility load — this was a genuinely under-sized genset plant, not just a display glitch. Now capacity-adapter.ts equipmentTable() derives the genset count from facility kW, so utilization can never exceed 100% (verified: Generators 115% → 94% at 110 MW IT).
At-Risk / Watch rows are now clickable → diagnosis modal on ALL four capacity system tables (Power + Cooling + Rack & Space + Network), not just the power utilization bars. The status chip opens the shared DiagnosticModal with the already-computed remediation (add N units / shed X MW / raise rating) + quantified levers + jump-to-Requirements/Phase-Plan. Closes the "every red value clickable" mandate for the capacity page.
Gates: walk 31/0 · trace-parity 116/116. No engine change.
v1.115.67PATCH
DCMOC — futureExpansionMw wired: Architecture reserve now shown as capacity headroom
Fixed
futureExpansionMw (Architecture rail) was a dead control — set by the user, drove nothing. It now surfaces on the Capacity Planning page as an "Expansion Reserve" KPI (reserved IT MW + the implied ≈MW facility-power headroom at the current PUE), so the declared future growth visibly reserves capacity the utility feed and footprint should be sized for. Display-only wire — deliberately does NOT alter the utilization/forecast math (kept the sensitive capacity model untouched). Last of the store-wide dead-control sweep.
Gates: walk 31/0 · trace 116/116. No engine change.
Found by a store-wide dead-control audit (agent), directions confirmed with the owner:
Removed
AI Chip Type (Primary) select (Requirements → Workload) — a dead control redundant with the Arch Profile Picker right above it, which is the real AI-hardware selector (applies density / cooling / tier / archKey). The chip dropdown set a store field that drove nothing. Removed the select, the aiChipType/AiChip store field + type, and its Summary row.
contractorAvail — a CapexInput interface field with no UI control and no reader; removed.
Fixed
Long-Shift Permit is now a real control with a cost effect. The PP 35/2021 warning (12h shift in Indonesia) was shown but the includeLongShiftPermit flag was orphaned — no toggle, no effect. Added a permit checkbox under the warning; without the permit a 12h ID shift now carries a screening +10% labor OT penalty (≈14h OT/week per PP 35/2021), with it the penalty is removed. The flag flows into the staffing cost memo (and its deps), so the toggle visibly moves monthly labor cost.
rackForm (42U/48U/OCP) left as-is — it's a documented design spec shown in the Summary/dossier, not a misleading cost control.
DCMOC — 3 more dead controls wired: transformer lead/type + delivery method
Fixed
Three more dead controls (same class as v1.115.64, found by continuing the sweep): the Requirements → Infrastructure Transformer Lead, Transformer Type, and Delivery Method selects were declared in CapexInput and written to the CAPEX store, but calculateCapex never read them — picking Expedited transformer, Oil-filled, or EPC/Turnkey moved CAPEX by $0. Now wired to real screening multipliers (DATA.capexDetail.fomTxLeadMult/fomTxTypeMult/fomDeliveryMult + capex-data.ts twin), default option = 1.0 (baseline unchanged). Deliberately NOT reused the engine advanced-model's transformerLeadMult/deliveryMethodMult — their enum domains differ (standard/extended/emergency vs the UI's standard/expedited/long_lead; dbb/db vs design_build/design_bid_build), which would have silently fallen back. contractorAvail left alone — it's an unused interface field with no UI control (nothing to wire).
tools/audit-dcmoc-enum-coverage.mjs extended to the 3 → 22/22 clean.
DCMOC — 8 dead CAPEX controls now actually affect cost
Fixed
Eight CapexDashboard FOM selects were dead controls — declared in CapexInput but never read by calculateCapex, so the user could pick Power Distribution, PDU type, Cabling, Floor, Security level, Fiber entry, Site condition, and Market condition and CAPEX would not move at all (a "control that does nothing"). Each is now wired to a real screening multiplier (DATA.capexDetail.fom*Mult + capex-data.ts twin), combined into the cost stack. Every one defaults to 1.0 at its UI default option so baseline projects are unchanged — only non-default picks move CAPEX (verified: default $60M → retrofit + military security + overheated market + intelligent PDU ≈ 1.40× → $80M).
The new fom*Mult maps are deliberately DCMOC-specific names: the audit gate (below) caught that the engine's own advanced capex model already had a marketConditionMult/siteConditionMult with DIFFERENT enum domains (buyer/balanced/seller), which a naive add would have silently collided with — so the DCMOC maps are namespaced to avoid it.
Changed
tools/audit-dcmoc-enum-coverage.mjs extended to cover the 8 newly-wired selects — 19/19 mappings clean; it's what surfaced the naming collision during development.
DCMOC — permanent gate against the silent enum→DATA-key fallback bug class
Added
tools/audit-dcmoc-enum-coverage.mjs (STRICT ship gate). The same bug class bit us twice (v1.115.61 market northern_virginia, v1.115.62 four CAPEX selects) — a UI <select> value used as a key into an engine DATA map that has no such key, silently falling back to a neutral 1.0 multiplier / $1M cost and mis-costing a real design choice with no warning. Rather than keep finding these by hand, this gate mechanically verifies every cost/value UI option resolves to a real key: it loads the engine DATA (vm sandbox), extracts each select's option domain from CapexDashboard.tsx (+ curated array-selects for redundancy/seismic), and asserts membership against DATA.capexDetail.* / DATA.markets (cityMarket alias-aware via resolveMarketKey). 11/11 mappings clean (rack/cooling/building/fire/alarm/ups/gen/substation/redundancy/seismic/market); exit 1 on any gap. Self-tested: removing a key makes it fail with the exact field "value" → map has no key. Wired into the DCMOC ship-gate list (CLAUDE.md) so a new option or renamed key can never silently regress this class again.
v1.115.62PATCH
DCMOC — CAPEX select options that silently fell back to default cost, now resolve to real values
Fixed
Four CAPEX selects had UI option values with no matching engine DATA key → silent fallback (same bug class as the Northern Virginia market fix; found by an audit sweep). Every one under-/over-costed a real design choice with no warning:
Rack densityultra (75 kW/rack) had no rackMult key → used 1.0× instead of 1.9× (electrical/building CAPEX understated ~47% on the highest-density option).
Fire suppressioninert (IG-541) and sprinkler (wet-pipe) had no keys → both 1.0× instead of inergen 1.2× / water 0.6×.
Fire alarmbeam (beam/aspirating detection) had no key → 1.0× (added a real 1.2× premium).
Substationpad_mounted_11kv and dedicated_66kv had no keys → both fell to the $1M shared cost instead of ~$2M / ~$5.5M (front-of-meter CAPEX understated 50–80%).
Fix: completed DATA.capexDetail (rackMult/rackKw/fireSuppressionMult/fireAlarmMult/substationCosts) in the engine + the capex-data.ts fallback twin with every UI option value as a real, sourced key. Additive — no rename, no persisted-value migration.
Changed
Two stale-memo wiring fixes (audit): Construction Engine's WBS/Gantt tree now re-derives when IT load changes (was missing itLoad from its useMemo deps — read via getState()); Executive capacity sparkline now subscribes to occupancyRamp instead of reading getState() outside the memo, so it updates when the ramp changes in isolation.
DCMOC — fix Northern Virginia market → real $215 colo rate, not the fallback band
Fixed
City-market → colo-rate resolution was broken for Northern Virginia. The <select> emits northern_virginia, but the alias map only had virginia → n-virginia, so northern_virginia resolved to the non-existent key northern-virginia and the flagship market silently fell back to the JLL band ($185) instead of its real $215/kW·mo rate — directly undercutting the "se akurat" AUTO revenue. Extracted the duplicated (and broken) alias into one resolveMarketKey() (lib/market-key.ts), fixed the alias, and adopted it in all three surfaces that map city → DATA.markets (Financial AUTO, FinancialPage pro-forma, Strategic Planning). Verified: selecting Northern Virginia now sets revenue AUTO = $215 and writes it through app-wide. Probe _dcmoc_financial_auto_probe strengthened with the market-resolution assertion (regression guard).
v1.115.60PATCH
DCMOC Financial — AUTO-by-default with per-field manual override
Changed
Financial Pro-Forma is now AUTO-first. Owner: "masih banyak data manual — saya mau auto, tapi bisa manual override klw di tick, se akurat dan optimum." All 15 parameters (revenue, discount/WACC, project life, escalations, tax, depreciation, NRC/MRC, contract, take-or-pay) default to an accurate value derived from REAL data — market colo rate (DATA.markets[market].coloPrice, e.g. N.Virginia $215) else the JLL/CBRE 2026 band × tier; Damodaran regional WACC (DATA.discountDefaults); country economy (tax / inflation / wage growth); design CAPEX/kW for NRC; Contract Duration (Requirements). Each field shows a green AUTO chip whose tooltip names the exact source.
Per-field override. A tick on any field flips just that one to OVERRIDE (editable, amber chip); the rest stay AUTO. Un-tick returns it to live AUTO. Replaces the old single all-or-nothing userEdited boolean that froze the entire block on any edit, and removes the un-sourced 120 + electricityRate×500 / 0.06 + riskPremium magic formulas (last holdout of the number-provenance audit).
Revenue write-through preserved. The effective revenue (auto market colo OR manual override) writes through to the sim-store SSOT, so Executive / Monte Carlo / Report / Phased Finance all compute on the same basis. Store clamps 50–500/kW·mo. Opening Financial applies the market-accurate basis app-wide (honest consequence: the default with no market selected is the JLL mid-band $185, more defensible than the prior conservative $150 seed).
DCMOC — honest payback propagated to comparison/phased surfaces + PDFs
Fixed
Payback honesty reached the remaining surfaces: ReportDashboard's negative-NPV narrative used a paybackLive <= projectLifeYears heuristic that printed "payback 15.0 yr" for a never-reached payback (clamp == life) → now reads the paybackReached flag. Scenario Comparison (KpiCell + matrix row) and Phased Financial (headline + per-phase table + scenario table) now show "> N yr" for unreached payback and suppress the distorting delta; the Phased Financial PDF (weighted, per-phase, scenario rows) matches. Extends the v1.115.57 fix so no surface — screen or export — prints a fabricated payback year.
v1.115.58PATCH
DCMOC — honest break-even occupancy: "not reachable" instead of a clamped "100%"
Fixed
Break-even occupancy no longer lies at the clamp: calculateFinancials clamped min(1, opex/revenue) — when OPEX exceeds revenue even at FULL occupancy the card printed "100%" as if break-even existed there. FinancialResult gains breakEvenReachable; the Financial card now prints fault-red "not reachable — OPEX > revenue @100%", and the Financial PDF's occupancy table no longer places its "← Break-Even" marker on the fabricated 100% row. (FinancialPage's overview stress card already handled this honestly with its own bisection — the engine + card + PDF now match.) Completes the honest-metrics family: revenue SSOT → EBITDA/IRR diagnosis → IRR n/a → payback not-reached → break-even not-reachable.
v1.115.57PATCH
DCMOC — honest payback: "> 15 yr (not reached
" instead of a fabricated "15.0 yr")
Fixed
Payback no longer fabricates a year: calculateFinancials fell back to paybackPeriodYears = projectLifeYears when cumulative cash never turns positive — every surface then printed "15.0 yr" as if capital were recovered in year 15. FinancialResult gains paybackReached / discountedPaybackReached flags + one shared fmtPayback() formatter ("> 15 yr (not reached)"); adopted on the Executive IRR-card sub, Financial engine payback card (simple + discounted), Portfolio comparison row, and the Cash-Flow chart no longer draws a payback marker on a curve that never crosses zero. Same honesty class as the IRR "n/a" fix (v1.115.56).
v1.115.56PATCH
DCMOC — honest IRR "n/a" + last stale-basis strings unified
Fixed
IRR no longer fabricates "0.0%": when the DCF never turns cash-positive (NPV < 0 and the solver pins ≤ 0) there is NO positive root — the Executive card now shows n/a ("no positive root (cash-negative)"), the diagnosis modal explains it, and the ƒx trace node returns null to match (trace-parity baseline 117→116 while the default project is cash-negative — the node honestly has no number to compare).
Stale revenue-basis strings unified: lib/screening.ts REVENUE_PER_KW_MONTH was a duplicate 150 literal — now re-exports DEFAULT_REVENUE_PER_KW_MONTH so the two names can never drift; Financial overview stress-test fallback used $280 while its own Pro Forma ran $150 (same page, two bases) — now falls back to the live store basis with an honest label; fin.ebitdaY5 trace template and the site.ctx binding formula no longer cite the dead $280.
v1.115.55PATCH
DCMOC — below-hurdle IRR diagnosis on the Executive KPI, Auto-optimize wired to the real solver
Added
Executive IRR card diagnoses itself below the 12% hurdle: fault-red + "BELOW 12% HURDLE — click for diagnosis"; the DiagnosticModal explains the equity case vs a DC fund's cost of capital and offers levers (raise the write-through revenue basis, re-phase the build via Phased Finance, cut OPEX drivers) — and its Auto-optimize button invokes the page's real 3-objective solver (bisects revenue to the hurdle, preview-then-Apply). Verified headless end-to-end: red card → modal → Auto-optimize → "3 objectives evaluated" results panel, 0 errors. Completes the mandate pair with the negative-EBITDA card (v1.115.53).
v1.115.54PATCH
DCMOC — revenue basis edit WRITES THROUGH: Financial field → sim-store SSOT → every DCF surface
Fixed
Financial revenue field was local-only: the "Revenue per kW/month" input (Financial → Pro Forma) edited only this page's state — Executive/Monte-Carlo/Report/Phased-Finance kept the old basis, so the EBITDA-diagnosis lever "raise the basis in Financial" silently did nothing app-wide. The edit now writes through to the sim-store tunable (the same one the optimizer moves), with reverse sync so an optimizer Apply refreshes the field while the page is open. Store clamps the tunable to $50-500/kW·mo (a typo can no longer poison every DCF).
Verified headless end-to-end: set $200 in Financial → store reads 200 → Executive negative-EBITDA chip clears. Probes: walk 31/0 · trace-parity 117/117 · optimizer 8/8 · fuelgen-scale 5/5.
v1.115.53PATCH
DCMOC — negative-EBITDA diagnosis on the Executive KPI
Added
Executive EBITDA card now diagnoses itself when negative: at the honest $150/kW·mo screening basis a small project can run operating-cash-negative (the old $280 default masked it) — the card turns fault-red ("NEGATIVE — click for diagnosis") and opens the shared DiagnosticModal: why (year-5 revenue vs OPEX, live figures), levers with jump-to-tab (raise the revenue basis in Financial — JLL 2026 $140-230 band; Run Optimization to bisect revenue to the 12% hurdle; cut OPEX drivers in Operations), and the honest note that this is real screening economics, not a bug. Closes the "every red value clickable" mandate for the Executive KPI row. Verified headless: card renders red, modal opens with levers (screenshot).
v1.115.52PATCH
SecondBrain — OmniRoute portal pill
Added
OmniRoute pill in the SecondBrain nav, third portal beside Vector Index and Sessions (rose→violet CTA): one click opens the locally-installed OmniRoute dashboard (http://localhost:20128/) — the MIT free AI gateway (250 providers, 90+ free tiers, one OpenAI-compatible endpoint at /v1). OmniRoute v3.8.48 installed globally via npm and persisted as a systemd --user service (omniroute.service, auto-restart, boots with login). Local-app pattern identical to the Sessions pill.
tools/_dcmoc_fuelgen_scale_probe.mjs — headless verification of the Y-slice's behavior change: seeds a 2.5 MW project (genset class N× 2.5 MW) then a 150 MW project and asserts the unit class switches to N× 3.0 MW (DATA.fuelGen scale rule, AssetGenerator-consistent); asserts Phased Finance break-even revenue derives from the live $150/kW·mo basis (a $280 regression would print ~$27x); screenshots FuelGen/Executive/Financial/Phased Finance for visual review. 5/5 green.
Verified
Screenshot review of the finance surfaces on the unified $150 basis: Phased Finance blended IRR 13.7% ≥ 12% hurdle, NPV +$2.2M, MoIC 1.15×, payback 7.6 yr — coherent. Known honest consequence: the default 2.5 MW Indonesia project's Executive year-5 EBITDA reads slightly negative at conservative wholesale $150 (raise revenue in Financial or run the optimizer to clear it) — this is the real economics, not a bug; the previous $280 default was masking it.
v1.115.50PATCH
post-audit certification sweep — full site gate suite green, 1 mobile fix
Fixed
setup-supabase.html mobile overflow (+20px @390px): injected step-list <li>/<span> tokens (URLs, keys) pushed past the viewport — mobile guard added (overflow-x: hidden + overflow-wrap: anywhere on li/span/code). Responsive-layout render gate back to CLEAN 116/116.
Verified (certification, no changes needed)
Full remaining gate suite after the X/Y/Z audit ships: dark-coverage CLEAN (117 pages × both themes) · responsive-layout CLEAN · mobile-responsive strict pass · interactions CLEAN (palette/diagrams/scrolly real-input) · a11y CLEAN (0 critical/serious × 2 themes) · page-gates CLEAN · hero-images CLEAN · article-charts provenance CLEAN (29) · explain-db ALL GREEN · dc-corpus ALL GREEN.
v1.115.49PATCH
LTC — trace coverage completed: 48 traceable values across every surface
Added
Trace everywhere: the DETAILED tab's full output grid (20 values — liquid/air heat, flow, pump, CDU, PUE, COP, WUE, energy, OPEX gross/credit/net, carbon ±, densities, control index, future factor, risk, confidence) and all 6 MODEL VALIDATION values are now click-to-trace targets, joining the KPI strip, RESULTS cards and P&ID equipment — 48 traceable values total. New dependency entries (futureFactor, confidence, rack/design density) mirror the compute chain. DETAILED-grid ƒx badges render via CSS ::after so the engine's per-render setText() cannot wipe them (verified surviving recompute).
v1.115.48PATCH
LTC — trace indicators now VISIBLE + RESULTS values traceable
Fixed
The v1.115.47 trace indicators never rendered (owner: "belum ada tracenya"): the ltc-traceable class was applied to the KPI value element while the CSS targeted the card — so no ƒx badge or dotted underline appeared anywhere. The whole KPI card is now the trace target (badge + underline render, keyboard-focusable), 12 RESULTS-card values gained inline ƒx chips (Supply/Return temp, ΔT, Flow, Capture, CDU count, Pump Power, System COP, PUE, WUE, Total Facility, Annual Energy), and the SYSTEM OVERVIEW label carries the hint "klik equipment / angka ƒx untuk trace perhitungan". Verified: 6 ƒx cards + 12 traceable rows render, card click opens the trace, 0 console errors.
v1.115.47PATCH
LTC — click-to-trace engine (DCMOC pattern
, information-rich P&ID, clean steppers)
Added
RZ Trace — the DCMOC trace-engine indicator, adopted into the LTC lab. Headline values (KPI strip, P&ID equipment) carry a dotted-amber underline + ƒx chip; click opens a trace popover: big live value + label + provenance chip (ENGINE · models.ltc.compute / INPUT), description + formula from the page's own 95-key parameter metadata, and clickable dependency pills (each dep's live value — click to drill deeper, with a breadcrumb back). Footer: "Open in DETAILED tab" (jumps to the live equation inspector) + "Copy trace" audit text. Content 100% sourced from the existing PARAM_TOOLTIPS metadata + live model — only the ~22-KPI dependency map is new, and it mirrors the compute chain.
Information-rich SYSTEM OVERVIEW (owner: values were being overlooked): every equipment now shows its live headline value inside the diagram — Dry cooler "2,244 kW rejected", Pump "124.1 kW · 28 m head · η 78%", CDU "10 CDU · ΔT 8.1 K", IT racks "12.00 MW IT · 97.1% captured" — plus the loop labels. Equipment groups are clickable (hover highlight) → their trace.
Clean − / + steppers replacing the ugly native number-input spinners (hidden page-wide): on all 6 primary-slider value boxes and every number input in OTHER PARAMETERS (84 steppers), stepping by each input's real step/min/max and firing the live recompute; 32px touch targets on mobile.
Changed
Modelling block diagrams — essentials first, nothing removed: dense sub-value lines are de-emphasized (dimmed) and light up on hover; every block keeps its existing click-to-detail popup (the detail layer is one click away, per "yang utama aja, detailnya klw di klik — jangan dihilangkan").
v1.115.46PATCH
DCMOC Z-SLICE — hardcode audit FINISHED: 227 → 0 WARN, every number carries a documented basis
Fixed
Hardcode gate v2 (string/comment-blind): the v1 scanner flagged ~140 false positives — numbers inside STRING literals (regulation cites like "PP 35/2021" parse as division) and comments. Scanner now blanks string/comment spans before token matching, adds calendar/physical conversions to the ignore set (365 d/yr, 4.33 wk/mo, 4046.86 m²/acre, 255 RGB, 2^32 LCG divisor, 168 h/wk) and recognizes more provenance keywords (EPA/CBRE/BNEF/IEA/Uptime/JLL/USGS/ASCE/DEFRA/World Bank/audit-ok).
All 45 real findings triaged to zero: RevenueEngine guard fallbacks ($185 MRC / $280 NRC) now single-source from the JLL/CBRE-cited defaultRevenueInputs (were duplicate literals that could drift); every remaining screening constant honestly annotated at the point of use — vendor-labor ×1.30 premium (Simulation ↔ Sensitivity parity pair), labor 4% / maintenance-CPI 3% escalators (Report/TaxIncentive/MaintenanceStrategy), diesel CO₂ 2.68 kg/L (EPA/DEFRA) + HVO +3% volumetric / −90% lifecycle, Uptime tier availability band 99.700–99.995% (dimension-explain ↔ value-trace parity pair), rack density classes 6/12.5/25/75 kW, NFPA 2001 clean-agent options, forecast-vs-budget skews, Numerical-Recipes LCG constants, PDF/SVG layout offsets (audit-ok).
Gate result: 227 WARN → 0 WARN across 227 scanned files — "jangan ada angka yang muncul tiba-tiba" now holds mechanically, and the gate stays in the ship suite to keep it that way.
v1.115.45PATCH
LTC — clickable engine-verified fix suggestions on out-of-band metrics
Added
Every out-of-band MODEL VALIDATION metric now tells you how to fix it. Under an amber row, up to two suggestion chips appear (e.g. PUE → "Economizer ↑", Flow intensity → "Supply Temp ↓"), computed by perturbing each impact parameter ±1 step through the REAL models.ltc.compute and ranking which moves pull the metric back toward its band — no heuristics, no hardcode. Chips are clickable: one click applies that ±1 step to the actual input (sliders sync, auto-run recomputes, the band marker moves live). Verified: six clicks walked PUE 1.28 → 1.26 toward the design band; 0 console errors; parity 15,750/0; calibration gate 38/0.
v1.115.44PATCH
DCMOC Y-SLICE — X-audit deferred violations: fuel/gen sizing to DATA, grid economics sourced, one depreciation convention, portfolio on live revenue
Fixed
Fuel & Generator sizing single-sourced + scale bug: DATA.fuelGen gains genset unit classes (2.5 MW ≤100 MW-IT / 3 MW above — CAT 3512E-3516E / MTU 20V4000 class), tank module 20 kL (UL-142 class), PM economics ($18k/unit + $5/kW, CBRE FM 2025) and per-genset environmental-compliance admin; FuelGenEngine reads them with parity fallbacks. Also fixes a real inconsistency: the engine displayed flat 2,500 kW units at every scale while AssetGenerator switches to 3,000 kW above 100 MW — spec and quantity now agree.
Grid Reliability economics sourced: diesel $1.25/L (IEA 2025 C&I band), UPS $27/kW·yr (blended VRLA/Li-ion, Uptime 2025), dual-feed $50/kW·yr, BESS $300/kWh (BNEF 2026 C&I) hoisted to one sourced constant block; fuel burn rate now reads DATA.fuelGen.genEfficiencyLPerKwh — the same EPA Tier 4 source FuelGenEngine uses.
ONE depreciation convention: new DEFAULT_DEPRECIATION_YEARS = 15 — Benchmark/Monte-Carlo/Scenario/Portfolio used 7 and Report used 20 while Financial/Investment/trace used 15, so the same project tax-shielded differently per page.
Portfolio on the live revenue basis: calculatePortfolio no longer hardcodes $120/kW·mo — the dashboard passes the sim-store revenue tunable, making portfolio IRRs comparable with every other surface.
Hazard-pay screening constants: AQI-driven O&M labor premium (0.15%/point above AQI 100, +10% hazardous step) hoisted with the EPA-band rationale; dead first-assignment code removed.
v1.115.43PATCH
LTC — MODEL VALIDATION: the model now shows its receipts vs ASHRAE/OCP/Uptime bands
Added
models.ltc.validation(model) in the shared engine + DATA.ltcCalibration.validationBands — six benchmark bands, each MIRRORING its DATA anchor (PUE = pueMatrix.liquid tier4–tier2 [Uptime 2026, corpus hyperscale p10/p50 as fleet ref]; WUE = water.wueByType immersion…rear-door envelope; loop ΔT, supply temperature and flow intensity = DATA.cdu.bands [ASHRAE TC9.9 W-classes + OCP cold-plate]; pump fraction ≤3% IT [OCP hydraulic guidance]) with a DATA.sources provenance entry.
MODEL VALIDATION card in the LTC right rail: live band bars with a value marker per metric, green in-band / amber out, an "N/6 in band" pill, and the source citation on hover. Honest by design — the default config correctly reports PUE 1.28 above the 1.10–1.22 design band and flow intensity 2.0 above the OCP 1.0–1.5 LPM/kW band (redundancy + hydraulic margin inflate design flow), pointing the user at exactly what to tune. Updates on every recompute.
Permanent gate tools/test-ltc-calibration.mjs (38 checks): bands finite + sourced + mirrored to their DATA anchors (a divergent copy fails), validation structure, and the default-config findings LOCKED — a future shift in what the model reports is a visible gate failure, never a silent change (MODEL_CALIBRATION_STANDARD).
DCMOC X-AUDIT — total wiring audit: 35 fixes, fabricated-PDF financials killed, ONE revenue basis everywhere
Fixed
Report/PDF fabricated financials (worst find): ReportDashboard's PDF export hardcoded NPV $15M / IRR 18.5% / payback 3.2 yr while the screen showed the real engine numbers — the exported document now uses the SAME live financialResult as the page (audit X, 3 opus agents over all 33 modules).
ONE revenue basis app-wide: DEFAULT_REVENUE_PER_KW_MONTH aligned to the sim-store optimizer tunable ($150 — the constant said $280 while PhasedFin read the store's 150, so the same project showed two different IRRs). Every surface (Executive, Financial, Investment, Report, Benchmark, Monte-Carlo, Scenario Comparison, dashboard DCF, ƒx trace) now reads the LIVE store value with the constant as fallback — an optimizer Apply that raises revenue now moves every IRR on every page, not just Phased Finance.
Executive design margin wired: hardcoded ×1.2 provisioning margin now reads designMarginPct (the Requirements slider actually moves the Executive capacity KPI).
Carbon tooltip prices matched to engine ($25 vs $35, $50 vs $68 contradictions), Benchmark depreciation aligned, ~30 more magic literals moved to constants/DATA or labeled screening across CarbonEngine, CBMEngine, BenchmarkEngine, AssetLifecycleEngine, CapacityPlanningEngine, Reliability, DesignTools.
WorkOrderView Rules-of-Hooks bug: overloadAlerts useMemo sat AFTER the null-country early return (hook count could change between renders) — moved above the guard, null-safe.
Added
tools/audit-dcmoc-hardcodes.mjs — permanent WARN-level heuristic gate scanning module render paths for unexplained numeric literals (225 advisory findings baseline; violations get fixed or annotated, the gate keeps new ones visible).
Comparison charts wired to the site (Q): Talent/Grid/Disaster/Tax country charts always include the Requirements country (Oman-class bug — top-15 slice dropped it); selected bar highlighted.
Capacity margin coherence (R): utilization ≡ 1/(1+margin) structural floor exposed — margin-aware bands (OK starts above the floor; "margin 30% = Watch forever" absurdity gone), built-capacity denominator from the committed phase plan, white space AUTO-derived (racks × 0.72 m² ÷ 35% gross-up) instead of guesswork, space lever no longer jumps to Staff Model Config.
Added
ScoreValue 100% (M): shared gradient primitive (green→amber→red, direction-aware: risk/utilization lower-better) + ƒx trace + explain tooltip — adopted across Site-Intelligence (5 children + comparison matrix), Results, Sustainability pillars (4 new trace ids), Reliability, Financial, Construction (SPI/CPI 1.2-scale), Commissioning, Assets, Risk, Capacity. 20 new per-score explain keys (838-entry DB).
Run Optimization is now a real solver (U): 3 objectives (blended IRR ≥ hurdle via a page-parity closure + 40-iteration bisection with residual line; capacity margin-aware band check; budget vs CAPEX P80 AACE band) → per-objective status + preview-then-Apply through the tunables allowlist guard. Project Status derived from lifecycle (was hardcoded "Design").
FAQ → glossary with analogies (W): 41 analogies (Lightbulb callouts) + 39 new entries (incl. Delivery & PM category: RFI/Submittal/ITP/NCR/punch/turnover/DoA/decision velocity) + auto Engine Glossary rendering all 838 explain-DB entries with combined search; 25 core DB entries now carry analogies site-wide.
LTC — micro-polish: live run stamp, preset dots, hover/edit feedback
Changed
Live run stamp: "Last run HH:MM:SS ✓ Completed" now updates on every recompute (auto-run included), not only on the manual button.
Preset dots on the three architecture-filled primary sliders (Liquid Capture, Supply/Return Temperature) — green dot with "Auto-set by the architecture preset — adjust freely" tooltip, mirroring the AUTO taxonomy.
Feedback: KPI cards get a subtle hover lift; the slider value box pulses on drag (both prefers-reduced-motion aware).
Every input is now classified, ending the "which is which" confusion: OTHER PARAMETERS regroups into three numbered sections — 1 · Design inputs (20 — your decisions), 2 · Auto-filled (11 — filled from the country & architecture presets with sourced engine data; chip shows AUTO · country/AUTO · architecture, flips to OVERRIDDEN the moment you edit, with a per-field ↺ that restores the preset value and recomputes), 3 · Optimization & calibration (6 — targets and tuning knobs). A 3-step guide strip (1 Set design → 2 Review auto-filled → 3 Tune & optimize) sits under the INPUTS header; single-column readable layout in the rail. All element ids/bindings untouched; the page's advanced-mode toggle still works (tier grids keep the input-grid class).
Firm-precision pass: thousand separators on all ≥4-digit values in the result cards, quick-KPI snapshot and model diagrams ("23,312 LPM"), dT→ΔT, tabular-nums on numeric columns so figures align. Display only — engine math untouched (parity 15,750/0).
The dense legacy blocks no longer render as an all-expanded wall. Every relocated analytical block (24 across the four tabs) is now a titled collapsible card — collapsed by default, the most useful one per tab open (FLOW leads with the visual Energy + Loss Breakdown instead of the raw variable dump). Each tab opens with a one-line guide of what lives inside; tab buttons gain icons. Accordion heads are keyboard-accessible (aria-expanded, focus-visible) and the titles carry the existing hover-explanations. All element ids/bindings untouched; charts and controls fully functional when expanded (verified: Pareto expands + renders, scenario replay, 0 console errors).
Requirements · Average Rack Density dropdown vanished on scroll (owner, Android): options committed on pointerdown, so on touch a scroll gesture's first contact instantly selected an option and closed the list. Selection moved to click (the standard pattern — a real swipe never synthesizes a click), plus touch-pan-y overscroll-contain on the list, max-h-56 → max-h-72 (the old height also hid the last rows — why it looked like options stopped at 75 kW), and a label-forwarding guard on the combobox root (Fields wrap controls in a <label>; clicks on non-labelable descendants were re-dispatched to the button). Fix lives in the shared CreatableCombobox — every dropdown in the app benefits.
Added
AI-era density presets (engine-true nominals from DATA.requirements.archProfiles): 120 kW GB200 NVL72 / OCP ORV3 · 140 kW GB300 NVL72 · 200 kW Rubin VR200 class (analyst est.) — average density takes the NOMINAL rack kW; peak (EDPp, e.g. GB300 192 kW) is the CAPEX power-provisioning side. Target Rack Density max raised 250 → 300.
Red-value diagnostic adoption widened (owner: "tidak hanya ini tapi semua"): Construction EVM SPI/CPI < 1 (live EV/PV/AC, behind-plan phases, forecast slip + quantified levers), Financial health grade D/E + CPI/SPI passthrough (composite formula with live terms), Reliability composed availability below tier target (reuses the page's own availExplain — modal can never disagree with the panel), Sustainability overall grade C/D + every pillar score < 50 (thresholds SOLVED from the pillar formulas: score ≥ 50 ⇔ PUE ≤ 1.35 / grid share ≤ ⌊45/gi⌋% / WUE ≤ 1.1). Investment & Maintenance SLA skipped honestly — rich inline diagnostics already exist there (no double-diagnosis).
Changed
Commissioning WBS card now full-width (a 7-level WBS with sub-activities was cramped in a half column).
Dead code removed: openWorstPhase (superseded by the diagnostic modal) + unused leverNav.
v1.115.35PATCH
LTC — country presets now pull the crawled, sourced tariffs from the shared engine
Changed
Country profile selection re-baselined to the sourced engine data (deliberate + disclosed): picking a country in the LTC lab now fills electricity price, grid-carbon intensity and water tariff from the shared DATA.electricityTariffs / DATA.waterTariffs (EIA/Eurostat/PLN/EMA/Ember/PUB…, each with DATA.sources provenance) instead of the page's older inline numbers — e.g. Indonesia: 0.08→0.082 $/kWh (PLN B3 2024), carbon 0.73→0.726 kgCO2e/kWh (Ember). Inline profile values remain the offline fallback; computeModel formulas untouched (parity 15,750/0).
Reliability: Redundancy What-If via engine reliability.kOutOfN — UPS + genset at N/N+1/2N with per-step cost and the technology-economics verdict incl. the explicit "negligible — the simpler N+1 option wins" branch + $/downtime-minute break-even; chain-derived SPOF list (single-path series elements + remediation); Monte-Carlo availability band (200 seeded-LCG re-runs, MTBF/MTTR ±20% → P10/P50/P90 vs tier target).
Sustainability: facility-level water table via engine water.facilityFootprint (climate-adjusted WUE, upstream-power water — with the honest note that on fossil grids the upstream share dominates), CUE row + ERF rendered honestly as 0 (no heat reuse modeled), 24/7 CFE note card (annual % vs not-modeled-hourly, no fake numbers).
Financial: WACC basis + scenario NPV table (base / rate+200bp / rev−15% / capex+10% via engine roi, NOT-ROBUST banner), break-even occupancy (bisection on the page's own cashflow), TCO $/kW·mo KPI benchmarked against the market colo price — replacing a mislabeled DATA.decision.revenuePerKwMonth fallback (key never existed; $280 always fired while labeled "engine") with the real DATA.markets colo price or an honestly-labeled screening default.
Strategy as a % mix (RTF/PPM/PdM editor summing 100%, discrete buttons = presets) + selectable SLA (2hr/4hr/NBD radio in the SLA tab) in store/simulation — and the Risk/Simulation "Expected Availability"/"Financial Exposure" cards now COMPUTE from these choices (design vs adjusted delta shown) instead of a hardcoded tier lookup.
Staffing page: "Campus Operations Model (engine)" accuracy card (nDc, tech basis strings, labor-hours strip, phenotype) + 18↔20 reconciliation — modeled base FTE vs employed FTE (+relief/shrinkage) explicitly reconciled; org chart re-captioned. Maintenance now precedes Staffing in the menu (regime determines headcount). 7 new engine-gate assertions (availability ordering, campus split, skeleton floor).
I — Construction "Delivery Governance" dossier beside BOQ: engine DATA.dossier.deliverables (9 playbooks: IFC/RFI/Submittal/ITP/NCR/look-ahead/variation/punch-list/turnover — purpose/trigger/owner/approver/cycle-time/hold-impact/standard) + DATA.dossier.pmFramework (control accounts, decision-velocity delegation bands, reporting rhythm, stakeholder-pressure responses, ISO 19650-5 security exposure, senior/intermediate PM charters) + models.dossier.deliveryGovernance (document volumes scaled by MW+months, variation bands $-scaled to CAPEX, program windows). New print-window dossier (13 sections, LIVE-WIRED vs STANDARD-PRACTICE chips, live tracked risks/issues).
J — Energy Audit & Management + Renewable & Storage Economics (2 new Sustainability tabs): engine models.energy.enpi (ISO 50001/IPMVP claim test — numerator/denominator/boundary/climate-normalization/window/exclusions PINNED and rendered verbatim; SUPPORTED/UNSUPPORTED verdicts), models.energy.storageEconomics (cycle+calendar degradation curve, augmentation plan, dispatch constraint, grid-services band, per-region regulator notes), models.energy.renewableProject (hybridScreen coverage → NPV @ regional WACC / IRR / payback → bankability gate with the honest "GRID-ONLY WINS HERE" verdict — technology-economics balance, not enthusiasm). Renewable tab gated on the CAPEX renewable selection; tariff ±30% / sizing ±50% scenario sensitivity. 5 new engine-gate assertions.
L — Capacity Planning tabs equalized: Cooling / Rack & Space / Network now carry the same component·capacity·utilization·status table + escalation levers as Power (cooling duty = IT heat not IT×PUE; density ceilings per cooling tech; binding-constraint row; network clearly labeled screening) — and the generic 5-card strip + Key Insights are tab-specific instead of duplicated.
Changed
Walk probe extended with program assertions (Cx WBS sub-activities, sane Cx duration, Construction WBS, Delivery-Governance button, capacity cooling table, campus ops card, strategy mix) — 31/0; full suite green (trace 117/117, optimizer 8/8, synergy 6/6, export 44/44, 0 console errors).
v1.115.32PATCH
DCMOC Architecture — BESS/PV in the diagram + Mechanical & Cooling view
Added
On-site renewables drawn (Workstream K1): when the CAPEX questionnaire selects Solar/Solar+BESS, the Architecture diagram now renders a Solar PV Array + BESS + PCS (sizing badges from renewSolarMwp/renewBessMwh) in an ON-SITE RENEWABLES group, tied to the MV bus — the owner-reported missing BESS. New vendored IEC-style symbols: solarArray, inverter, pump, heatExchanger (assets/engineering → 24 symbols).
Mechanical & Cooling view (Workstream K2): third Architecture diagram view — a dedicated cooling schematic via computeCoolingLayout, fully dynamic on the set parameters: air → cooling towers · chillers · CHW pumps · CRAH/in-row; liquid/RDHx → dry coolers · plate-HX (FWS/TCS) · primary/secondary pumps · CDU loops; deep-sea → intake (depth) · titanium plate-HX (loop ΔT) · seawater pumps (pipeline km), chiller-less. Unit counts from equipScale, duty = IT heat (not IT×PUE), supply/return loop edges, per-node hover detail.
v1.115.31PATCH
DCMOC — Commissioning WBS: L0–L6 × 3–5 sub-activities on the recursive Gantt
Added
Commissioning WBS (Workstream E2): new DATA.commissioning.cx.rich.subActivities (25 sub-activities across L0–L6, ASHRAE Gl.0/BCxA practice — FAT witness, pre-functional checklists, startups, point-to-point, power-path/cooling-chain integration, load-bank stages, IST scenario matrix, burn-in, punch/turnover) + models.commissioning.programWbs — levels get CALENDAR windows from the log-damped program schedule with crew-days effort attached; sub-activities distribute serial-with-overlap; drivers substituted live (witnessed-unit counts, IST scenario count per redundancy class, BMS point-count note). L5 flagged critical (endurance burn-ins are a hard calendar floor).
The Commissioning page timeline is now the shared recursive Gantt (expand a level → its sub-activities, per-bar diagnosis tooltip), replacing the flat 7-bar strip; legacy strip kept as engine-absent fallback.
Auto/Manual Optimizer (Workstream D): lib/optimizer/tunables.ts — the reviewable ALLOWLIST of fine-tune parameters Auto may move (revenue $/kW·mo band 100–300, CAPEX contingency 5–15%, design margin 5–30%, maintenance in-house share), with LOCKED_FIELDS (IT MW, tier, cooling, redundancy, country, workload, phases) asserted before every Apply. lib/optimizer/optimize.ts — deterministic bisection over the page's OWN verdict model (capex-weighted phase IRR), honest infeasibility (band edge can't clear → "nothing applied" verdict). Phased-Finance red IRR/NPV/PI modals now carry Auto-optimize → proposal preview (from→to, IRR/NPV deltas, guard note) → explicit Apply writes only revenuePerKwMonth (new persisted sim tunable; all revenue math now derives from it).
Capacity Planning At-Risk/Watch chips → shared DiagnosticModal: the inline expansion is replaced by the Workstream-C modal (reason + bisection-solved levers + jump-to-tab), consistent with Phased-Finance.
Permanent gate tools/_dcmoc_optimizer_probe.mjs (8 assertions): red IRR → modal → Auto-optimize → feasible proposal → Apply clears the NO-GO → requirement base data byte-identical. 8/8 green.
v1.115.29PATCH
article-28 — hero image + fixed card thumbnail
Added
article-28 cover image (assets/article-28-hero.webp, 1200×509): generated dark-editorial hero — a datacenter rack wall compressing through golden data streams into a single glowing laptop (the article's Compression Horizon motif, matching its gold/amber palette). Inserted at the top of the article body per the article-27 pattern.
Fixed the broken-looking article card: the articles-grid thumbnail now uses the hero image instead of the dark OG card (which rendered as an empty block with overlapping title); dimensions corrected (1200×509).
Commissioning "Program Duration 7832 d" (owner: "angkanya ngacau"): models.commissioning.programRich summed per-level LABOR-days serially and the page displayed it as calendar wall-time (100 MW → 7832 d ≈ 356 mo). The engine now also returns calendarDays/calendarMonths (the log-damped, capped programSchedule wall-time — crews work levels in parallel; never above the single-crew serial total) plus laborDays + crewEquivalent. Page headline shows CALENDAR (100 MW ≈ 20 mo, ~17.8 crews); labor effort is labeled secondary. Golden cx-calculator parity untouched (durationDays semantics preserved); 4 new engine-gate assertions.
Added
Universal detailed tooltips (owner: "tooltip untuk semuanya… harus detail"): 74 new bindings across 12 modules (Financial, Maintenance, Staffing, Capacity, Construction, Commissioning, Reliability, Sustainability, CAPEX, Risk, Asset Intelligence) — every KPI strip now carries a 2-4 sentence tooltip (definition + engine basis + drivers + decision caveat). Explain DB enriched with 10 detailed finance/energy keys (irr, pi, blended-irr, payback, hurdle-rate, discount-rate, idc, landed-cost, lcoe, bankability) → 818 entries, gate 10/10.
4-level Construction WBS (owner: "jangan hanya 2 level… in such detail"): new DATA.construction.wbs (72 L3 tasks + L4 work-packages across all 22 sub-phases, engineering-real bases) + models.construction.detailedSchedule — expands the CAPEX timeline into phase → sub-phase → task → work-package, durations distributed serial-with-fast-track-overlap, procurement-bound tasks carry longLeadWeeks basis + CRITICAL-path flag (e.g. transformer 60–120 wk ⇒ PO before NTP) + screening unit counts. GanttChart rebuilt as a recursive N-level renderer (collapse/expand per node, per-bar diagnosis tooltip: basis · depends-on · critical/risk); legacy 2-level API preserved for CapexDashboard.
Red-value diagnostic modal (Workstream C start): shared RedValue + DiagnosticModal (components/ui/RedValue.tsx) — any breached value renders red + clickable → modal with WHY (computed reason), threshold vs actual + gap, quantified levers with jump-to-tab, and an Auto-optimize slot (Workstream D). Adopted on Phased-Financial Blended IRR / Total NPV / PI.
Changed
Operations menu order: Maintenance now precedes Staffing (the maintenance regime determines headcount).
Version-stamp audit: corpus crawl artifacts (tools/dc-corpus/raw/*.html) excluded — not site pages.
v1.115.27PATCH
LTC SYSTEM OVERVIEW — clean SVG P&ID from the RZ Engineering Symbol Atlas
Changed
The P&ID is now a crisp, theme-aware inline SVG built from the RZ Engineering Symbol Atlas — equinor ISO P&ID symbols (PP001A centrifugal pump, PV005A gate valves; recolored to currentColor) + lucide fan/server (dry-cooler fans, rack servers), with a plate-HEX/CDU line-art vessel: dry cooler → valve → pump → valve → CDU/plate HEX ⇄ IT racks with liquid cold plates. Blue supply / red return dashed lines with arrowheads and animated flow (reduced-motion aware). Every label is live-wired — Supply/Return °C, total flow LPM, pump kW, CDU count update with the model; no baked text, so nothing bleeds. Replaces the muddy cropped-PNG strip (pid-day/night.png deleted; atlas source symbols vendored to assets/ltc/src/ for provenance). Sharp at any width, both themes.
v1.115.26PATCH
LTC Modelling Lab is now ONE app — old page fully consolidated into the dashboard
Fixed
"Seperti ada 2 apps." The dashboard revamp had left the ENTIRE original page visible below it (hero "Root Only Module", Legal Notice, "Integrated Engineering Architecture Focus", "High-Fidelity Engineering Calculator" + Workspace Controls) because relocateExistingPanels() targeted class names that don't exist (.pipeline-section/.sankey-section/.calc-output-section/.report-section — zero matches), so all ~27 analytical blocks stayed put and the tabs stayed empty. Rewritten as a route-everything consolidation: every child of the old .calc-output-panel is routed into a dashboard tab by its REAL class/contained-id — FLOW gets the pipeline/variable-flow/energy-breakdown blocks; CHARTS gets Target Guidance, Constraint/Degraded-Mode, Pareto, Monte-Carlo and Sankey; REPORT gets the Scenario Manager, Provenance Ledger, CSV + Executive/Engineering export packs and Methodology; DETAILED (catch-all — nothing can be orphaned) gets the quick-KPI snapshot, the full 23-card output grid, block/control diagrams, Explorer, Rule DSL, Equation Inspector, Calibration Quality, Self-Test and the rest. Fixed-overlay modals (#blockDetailModal, #sankeyDrillModal, #modeHelpModal) move to <body> so a hidden tab can't trap them; the full 61-input legacy panel folds into OTHER PARAMETERS (with a settings count); the emptied old shells + chrome are hidden under body.ltc-consolidated (fail-safe: if consolidation throws, the flag is never set and the original page still renders). The legal notice text now lives in the dashboard's SCENARIO INFO card (Terms/Privacy links kept).
Acceptance verified headless: single shell (page height ~1,577px, previously multi-screen dual-render), DETAILED 14 blocks / FLOW 3 / CHARTS 5 / REPORT 4, all functional; every element id preserved (engine bindings intact); both themes; 0 console errors. Gates: ltc-parity 15,750/0, dark-coverage, page-gates, a11y, interactions all CLEAN.
v1.115.25PATCH
Fix: login no longer drops when navigating — Supabase session rehydration
Fixed
Login state "lepas" on navigation/back — real root cause.auth.js getSession() reads only the rz_premium_session localStorage mirror; when that mirror was missing/expired while the Supabase session token (sb-*-auth-token) was still valid, every page showed "Login" although the user was signed in (the Supabase→sitewide bridge was never implemented — rz-supabase.js "plan B4"). New rehydrateFromSupabase() in init(): when the mirror is absent but a Supabase token exists, it lazy-loads the shared client, reads supabase.auth.getSession(), rebuilds the mirror (tier from the profile when available, role via the allowlist), re-renders the auth UI and dispatches rz-auth-change (so page gates like the LTC lab unlock immediately). Async + fail-silent; skips entirely (no module load) when no Supabase token exists, so offline/demo flows are untouched.
Second real cause on the homepage:index.html loads auth.min.js, a minified twin that was a week stale (2026-07-16) — the homepage always ran old auth logic regardless of auth.js fixes. Rebuilt auth.min.js from the fixed auth.js (terser) and bumped its ?v=.
Cache-bust: auth.js?v=2026-07-24a unified across ALL 156 pages (three divergent versions found: 2026-07-23 ×154, 2026-07-16b, 2026-07-14a).
Verified headless: Supabase-token-only → mirror rebuilt in <10ms (cached), user pill shows, LTC gate unlocks; navigation keeps state; logged-out still shows Login. 0 console errors.
Double tooltip (owner screenshot): Requirements Field labels set a native browser title even when the field also renders a rich RZExplain panel — the black native box covered the panel header. Native title now renders ONLY when there is no explainKey (requirements/ui.tsx).
Added
Phased-Financial KPI tooltips: Total NPV, Payback, and Investment cards (previously bare) now carry detailed tooltips — definition + formula basis + what moves them + decision caveats (e.g. payback ignores time-value; NPV can be positive while IRR misses hurdle).
KpiCard.explainKey: KPI cards can now bind the shared RZ_EXPLAIN_DB (rich Explain tooltip) instead of a hardcoded tip — the enabler for the universal-tooltip sweep.
DCMOC Architecture · Single-Line Diagram: the cooling/BMS lanes are hidden in SLD view, but their labeled group frames (COOLING PLANT / NETWORK & CONTROL) were still drawn — leaving empty boxes. The group loop now applies the same lane guard, so an empty frame is never rendered in SLD. Logical view unchanged (DiagramSvg.tsx).
Changed
DCMOC Operations · Staff Model Config: the controls+results block was locked to full viewport height (h-[calc(100vh-140px)]), so the shorter right panel left a large dead gap and the below-fold sections (Cause-Effect Lever Map, Environmental Impact) were undiscoverable. The right panel is now an in-page tab menu (Overview · Environment & Risk · Power & Cooling · Cause-Effect Map); forced height dropped so height follows content. No dead gap; every section reachable (SimulationDashboard.tsx).
v1.115.22PATCH
TCO — replace stretched market cards with a rate-outlook chart
Changed
TCO market-data card fixed: the earlier empty-area fix had stretched the 8 market KPI cells to fill the left column (huge cards, tiny content). Reverted to compact cells and added a real Market Rate Outlook line chart (build $/MW, colo $/kW, power $/kWh — indexed to Year 1 = 100 over the analysis period, from MARKET_DATA/country projections; day/night-aware, re-renders on region/country/theme change). Column now shows analysis, not stretched whitespace.
v1.115.21PATCH
TCO — flatten the saturated results panel
Changed
TCO Comparison results panel de-slopped: the right-hand results panel was a saturated indigo gradient (#312e81→#4338ca, CLAUDE.md rejected-pattern #7). Flattened to a clean flat instrument surface (#0e1626 + hairline border) so it matches the site's flat aesthetic + the spares dashboard; inner KPI cards / chart / white text unchanged.
v1.115.20PATCH
LTC — coolant thermophysical properties surfaced from the shared engine
Added
The RESULTS Flow & Temperature card now shows the selected coolant's Cp / density / viscosity / thermal conductivity, read live from the shared DATA.coolants (including the crawled/sourced viscosity + thermal-conductivity fields) — connecting the v1.115.10 data crawl to visible UI use. Switching coolant updates the props from the engine. No hardcode.
v1.115.19PATCH
Spares — tab-group chips for the module nav
Changed
Spares module tab bar tidied: the 25-tab wall now shows its group labels (ANALYTICAL / OPERATING ENGINE / REFERENCE / SUPPLY CHAIN & TRANSPORT) as clear blue section chips instead of tiny inline text, so the tabs read as grouped sections; active tab is a solid blue pill. Dropped the faint | dividers.
v1.115.18PATCH
Spares — left-rail input unit labels
Fixed
Spares ANALYSIS SETUP rail units: added the missing unit hints to input labels — Annual Failure Rate λ (/yr), Installed Base (units), Lead Time Sigma (wk), Qualified Alternates (suppliers) — so every input reads self-explanatory (the rest already carried wk/$/% units).
v1.115.17PATCH
Spares — KPI strip tidy: no truncation, one-line values
Fixed
Spares 8-module KPI strip readability: the single-row 8-across layout squeezed each card so word values truncated ("ESSEN…", "CRITI…", "Strat…") and labels clipped. Reworked to a 4×2 grid (8-across only ≥1650px) so every card is wide enough for one-line values; text values (tier/quadrant/status words) use a slightly smaller size via a JS-set .sp-kpi-txt class, numbers stay large, white-space:nowrap prevents mid-word breaks. Full labels + status colors + count-up preserved.
v1.115.16PATCH
LTC — REPORT A/B scenario compare
Added
Scenario compare in the REPORT tab. "Pin current as A" snapshots the current design's engine KPIs; as you tune inputs, a table shows A (pinned) vs B (current) with color-coded deltas (green = better) for PUE / System COP / Total Flow / Pump Power / Risk / Net OPEX / Carbon. Snapshots and deltas are pure engine values — no hardcode.
v1.115.15PATCH
Spares KB — deep-research crawl-enrich into the shared engine
Added
Spares knowledge base enriched with real, sourced facts via the provenance-mandatory dc-corpus pipeline (every fact carries a source URL + verbatim quote or it's rejected — nothing fabricated). New sources appended to tools/dc-corpus/sources.yaml (segment spares): OEM/FM reliability (Riello UPS ~250,000 h MTBF, Oxmaint, IEEE-493-lineage), procurement lead-times (DC Atlas / Wood Mackenzie Q2-2025 — transformers ~128 wk, gensets/switchgear/chillers/UPS), and obsolescence/DMSMS (JEDEC J-STD-048 6-mo LTB / 12-mo LTS). New negation-guarded extract patterns (mtbf_hours, mttr_hours, lead_time_weeks, obsolescence_months) in extract.mjs. 25 spares facts aggregated into 3 monotonic multi-source distributions in the engine @@CORPUS block: lead-time p10/p50/p90 = 16/30/52 wk, MTBF = 2.8k/4.8k/250k h, obsolescence = 4/8/12 mo. Wired as DATA.sparesPricing.corpusBands — additional sourced sanity bands, NOT replacing the golden reliability/sparesCatalog/classes values — with DATA.sources provenance (asOf 2026-07-23). Shared engine → benefits the spares calculator + DCMOC. Gates: test-dc-corpus 2847/0, test-rz-engine 747/0, value-bindings 85/0, calibration 19/0, min reproducible; rz-engine.min.js?v → 2026-07-23-sp. Unfetchable sources (403/404/429) were dropped, mttr kept below the distribution floor rather than fabricated.
DCMOC Architecture diagram — interactive topology tracing: hovering (or tapping) any node now shows a rich equipment tooltip (title · badge · counts · description) and highlights the edges it connects to while dimming everything else, so the power/backup/cooling/control topology reads itself. The hovered node's cell lights up (lane-accent glow); pan/zoom/Logical-SLD unchanged; prefers-reduced-motion respected via short CSS transitions. next build green, arch-probe 7/7, 24 interactive nodes.
v1.115.13PATCH
LTC — power-breakdown donut: center total + per-slice %
Added
POWER BREAKDOWN donut now shows the facility total in the ring center (e.g. "15.42 MW Total") and each legend row carries both kW and its % of total — all from model.breakdown (engine single-source), matching the reference.
v1.115.12PATCH
DCMOC Architecture diagram — label/badge polish
Fixed
DCMOC Architecture diagram label overlaps: the per-node redundancy/use-case badge (e.g. "2N", "AI") was drawn at the cell bottom where the sub-label already sits, colliding (MV Switchgear "1× boards", IT-Load "+10% margin"). Symbol nodes now render the badge as a top-left corner pill (the ×N unit count stays top-right), clear of the title/sub labels; phase boxes keep their bottom-centered NOW/PLAN badge. next build green, arch-probe 7/7.
v1.115.11PATCH
LTC — sensitivity tornado in CHARTS, engine-computed
Added
Two-directional sensitivity TORNADO in the CHARTS tab (renderTornado). For each of the engine's DATA.ltcCalibration.impactParams it perturbs the input −1 and +1 step through window.RZEngine.models.ltc.compute and draws a horizontal bar spanning ΔPUE(low)…ΔPUE(high) on a zero-centered axis (top-8 by range, blue→red). Fully engine-computed — no hardcoded deltas; wired to the same single source as the right-rail sensitivity. Both themes; responsive.
v1.115.10PATCH
Shared engine data crawl: coolants/CDU vendors, refrigerants, electricity+water tariffs, lead times
Added
Real sourced DC data crawled into the shared rz-engine.js (consumed by LTC + TCO + DCMOC via the engine — no page-local copies). Six new curated CSVs under /data/ + engine DATA keys, every value carrying a DATA.sources entry (source+asOf+unit+method):
DATA.coolants extended — added pg40 + dielectric_1p/2p and viscosityMpas/thermalCondWmk/ashraeClass on all fluids (ASHRAE HoF 2021 ch.31 / IAPWS-IF97 / OCP Immersion Fluid Base Spec).
DATA.refrigerants extended — added R744 + Novec7000 and priceUsdPerKg/pfas on all 11 (Chemours/Honeywell/3M TDS · EPA SNAP · ASHRAE 34-2022).
DCMOC Architecture diagram — blocks → proper engineering standard symbols
Changed
DCMOC Architecture-Engine single-line diagram rebuilt from filled blocks to real engineering standard symbols. Vendored a shared engineering-symbol SVG library assets/engineering/ (20 IEC/single-line symbols, viewBox 0 0 48 48, fill=none stroke=currentColor — utility pylon, transformer two-circle, MV switchgear, breaker, disconnect, ATS, genset+G, fuel tank, UPS, battery, PDU, rack, spine/leaf fabric, chiller, cooling-tower, dry-cooler, CDU, CRAH, pump, heat-exchanger, BMS, IT-load) + manifest.json — the canonical seam the parallel session's symbol DB merges into. New tools/build-engineering-symbols.mjs inlines them into dcmoc/.../diagram/symbols.gen.ts. Rewrote palette.tsx to render each node AS its symbol (44px), DiagramSvg.tsx so the node is the symbol + a label below (no filled box), inside labeled group boxes (POWER TRAIN / GENERATION / COOLING PLANT / NETWORK & CONTROL / IT HALL / GROWTH PHASES) with A/B bus voltage labels, day/night CSS-var surfaces, pan/zoom + Logical/SLD switch intact. layout.ts: enlarged cells to fit the symbols + labels and fixed the IT-HALL 3-cell / NETWORK-lane overlap. Topology/edges/buses unchanged; tsc --noEmit clean, next build green, arch-diagram probe 7/7. (Excludes the concurrent LTC session's uncommitted rz-engine.js edits.)
v1.115.8PATCH
LTC — sensitivity + design sub-scores single-sourced to the engine; DCMOC PUE fallback reconciled
Changed
Sensitivity is now real, engine-computed — no fabricated numbers. The dashboard's SENSITIVITY ANALYSIS previously invented deltas from hardcoded coefficients (*0.15, *0.003, *0.05, 0.002, *0.001, refs 24/40) whenever it couldn't read the page's impact table. Added models.ltc.sensitivity(input) to rz-engine.js — it perturbs each of DATA.ltcCalibration.impactParams (the single-source copy of the page's IMPACT_PARAMS) by its step through models.ltc.compute and returns ranked ΔPUE/ΔCOP/ΔOPEX/ΔCarbon/ΔRisk, mirroring the page's pdfImpactRows exactly. renderSensitivity now calls it (offline fallback reads the page's own #impactRows table, else "—"); the fabricated block is deleted.
Design sub-scores single-sourced. Efficiency/Resilience/Sustainability/Operability were UI-invented ((2.0-pue)/(2.0-1.02), annualGwh*0.4, :30). Added models.ltc.designSubScores(model) with its thresholds in DATA.ltcCalibration.designSubScores (+ DATA.sources entries); renderDesignStatus reads it — no magic numbers in the UI.
DCMOC pue.ts fallback reconciled to the engine (was a divergent second source): air 1.42→1.50, liquid 1.08→1.15, DEFAULT_PUE 1.42→1.50 — now mirrors DATA.pueMatrix[type].tier3 exactly (engine stays primary via getPUE).
Owner mandate "wired satu sama lain jangan hardcode sendiri2": LTC sensitivity, design sub-scores, PUE (LTC/TCO/DCMOC) and all LTC data now resolve through the one shared engine.
Notes
Additive only — test-ltc-parity.mjs still 15,750/15,750 with 0 drift (computeModel output unchanged). Engine chain green: test-rz-engine 747/0, test-value-bindings 85/0, test-reference-parity 155/0; rz-engine.min.js reproducibly rebuilt.
v1.115.7PATCH
LTC Modelling Lab — data basis migrated to the shared RZ engine + micro-interaction polish
Changed
LTC data basis is now the shared RZ engine (was 100% inline heuristics). Lifted the LTC lab's inline constant tables + every "magic" coefficient out of computeModel into provenance-tagged rz-engine.js DATA — DATA.coolants, DATA.rackProfiles, DATA.coolingArchProfiles, DATA.ltcClimate, DATA.ltcCountryProfiles, DATA.ltcCalibration — each with a DATA.sources entry (ASHRAE HoF / OCP / Uptime where empirical; method:"model-calibration" labeled honestly where a coefficient is a modeling choice). Added models.ltc.compute() that runs the LTC formulas from those DATA. js/ltc-system-modelling-lab.jscomputeModel() now delegates to window.RZEngine.models.ltc.compute (inline tables retained as offline fallback), and the LTC page now loads rz-engine.min.js. LTC + TCO + DCMOC now share one sourced knowledge base.
Formulas unchanged — golden-parity locked. New gate tools/test-ltc-parity.mjs drives the old inline model and the new engine model over 250 seeded input combinations (15,750 field checks) and asserts bit-identical output (max relative drift 0.000e+0). Engine chain re-run green: test-rz-engine 747/0, test-value-bindings 85/0 (catalog regenerated), test-reference-parity 155/0; rz-engine.min.js reproducibly rebuilt.
Micro-interaction polish (js/ltc-lab-ui.js + css/ltc-lab.css). KPI values now count up (eased, prefers-reduced-motion aware) with thousands-separated flow (e.g. 23,312); tab switches get a fade+slide entrance; a one-time card-entrance stagger on load; WCAG 2.4.7 :focus-visible rings on all sliders/tabs/buttons/preset-cards; and hover-explanations (RZExplain) wired across the dashboard (22 glossary terms: PUE/COP/WUE/CDU/ΔT/economizer…).
v1.115.6PATCH
Spares: interactive heatmaps + count-up + 445-part catalog wired into the engine, calc & DCMOC
Changed
Spares interaction layer (spares-readiness-calculator.html): the Supply-Risk × RPN and Kraljic heat-maps now show real values on hover / focus / tap (band × band + the current part's RPN + alternates + the recommended action) via a reusable cell tooltip — fixing the "colored boxes with no value" complaint. Replaced the cheap sp-num color-flash with a smooth rAF count-up + underline-settle on KPI/result numbers (prefers-reduced-motion safe). Added semantic status colors on readiness/hub % (green ≥95 / amber ≥80 / red), P(stockout), and tier (VITAL red / ESSENTIAL amber / DESIRABLE green), and a Run-Analysis loading→done state.
445-part catalog wired as a shared engine data source (was browse-only). New tools/build-spares-catalog-aggregate.mjs distils the curated 445-part catalog (js/spares-parts-catalog.js) into a compact per-commodity DATA.sparesCatalog aggregate (p50 MTBF/MTTR/lead-weeks/cost/alts/crit/EOL + part count) in rz-engine.js, exposed via models.spares.catalog.{commodities,byCommodity,classAggregate} with a DATA.sources.sparesCatalog provenance entry. The spares page gains a "Pick a real catalog part" searchable picker (filtered by commodity) that drives the real inputs (cost←cTyp, lead←ltTyp, alternates←alts, severity←crit, demand←MTBF, LTB←life/eol) with catalog vs screening provenance chips; screening defaults remain the fallback. DCMOC spares-adapter.ts layers the same catalog aggregate over its 6-class screening table into the shared models.spares.newsvendor. Golden-compare: the default screening path is unchanged; picking a real part deliberately shifts to catalog values. Engine chain green (test-rz-engine 747/0, dc-corpus 2700/0, value-bindings 85/0, calibration 19/0); rz-engine.min.js?v → 2026-07-23-cat. (Deep-research crawl-enrich of the KB follows in a later ship.)
v1.115.5PATCH
LTC Modelling Lab — TEMPERATURE PROFILE chart to match reference
Changed
TEMPERATURE PROFILE (LIQUID LOOP) rebuilt to the reference: two dot-marked line series (Supply blue / Return red) on a fixed 20–40–60 °C axis with 0–100% loop-length X labels, a top-right legend, and colored endpoint value pills — replacing the thin two-line sketch (odd 18/28/39 ticks, leftover monospace label font, no markers/legend/axis). Still live-wired to model.input.supplyTemp/returnTemp.
tco-calculator.html — Region → Country cascade (owner: "region dulu baru negara", like DCMOC). Replaced the flat 12-metro dropdown with two selects: Region (Americas / Asia Pacific / Europe / MENA / Africa / LATAM) → Country / Location filtered by region, built on load from RZEngine.data.countries (40 countries) grouped by .region, with every existing rich US/global metro kept under its matching region (no metro lost). A metro selection uses the unchanged MARKET_DATA record (golden-compare: N. Virginia / Dallas / Singapore TCO identical to before); a country selection derives a market record from the engine — power = economy.electricityRate, staff = baseSalary_Engineer×12 (fallback 85k×constructionIndex), buildCost = 12.0×constructionIndex, tax = economy.taxRate (wired to the WACC tax-shield), with per-region screening estimates for colo/land/CAGR (seeded from regional metro averages) and projection curves escalated via the country's inflationRate/laborEscalation. Snapshot card gains Corp-Tax + Currency cells + a provenance note ("engine: power/tax/labor/build-index; screening: colo/land/cagr"). Default stays Americas / N. Virginia (unchanged view). Verified: Indonesia → power $0.090, buildCost $7.8M/MW (12×0.65); UAE → $10.2M/MW (12×0.85), tax 9%.
tco-calculator.html — empty bottom-left area fixed: the left .tco-simulator-inputs column stretched to the grid cell but the market-data card's content ended ~1332px short, leaving a dead band. Gave the inputs column height:100% and made the snapshot grid flex:1 with grid-auto-rows:minmax(44px,1fr) so the KPI cells consume the height — dead space 1332px → 21px (padding only).
spares-readiness-calculator.html — hero banner de-boxed: the data-rz-hero + rz-hero-fit blur-letterbox rendered a black box when the cover image was absent. The banner is now default-hidden and reveals ONLY on a successful image load (assets/spares-readiness-calculator-hero.webp); onerror removes it — no dark void until the owner supplies the image.
v1.115.3PATCH
LTC Modelling Lab — topbar brand + design-status radial to match reference
Changed
Topbar brand now matches the reference: "Bagus Dwi Permana / LTC Modelling Lab" with a light hex-tile mark (fa-diagram-project on a blue-tint tile), replacing the dark gradient chip + "LTC System Modelling Lab / Comprehensive…" line.
DESIGN STATUS restructured to the reference layout — the blue ring sits on the left with the score % inside, and "Design Score" + the word grade (Excellent/Good/Fair) sit to the right of the ring (was a centered ring with the grade below). Sub-bars unchanged below.
Confirmed (screenshot, seeded root session): the light-theme topbar is white and the legacy DC-Solutions/Glossary back-links are hidden on desktop; both themes verified 0 console errors.
v1.115.1PATCH
Critical Spares Engine — rebuilt to the reference 3-column dashboard
Changed
spares-readiness-calculator.html — actual reference LAYOUT (the prior ship only recolored the old top-to-bottom sprawl; owner rejected it). Rebuilt the page body into the reference's dense 3-column analyst dashboard (.sp-shell grid, 264px / fluid / 312px, contained ~1460px so no empty voids; rails stack < 1180px):
LEFT rail — ANALYSIS SETUP: Select Commodity + 12 compact Fleet & Operational inputs (rail_* mirrored two-way to the real c1_*/s_*/r_* calc inputs via sp_bind + native event dispatch, so the existing calc reruns), Edit All Inputs, SCENARIO Active pill, ▶ Run Analysis (→ recalcAll()), Last-run stamp.
CENTER: the 8-module KPI strip + all 25 module tabs + panes, restyled to the flat reference card language.
RIGHT rail: View Methodology, Monte-Carlo card, Portfolio Summary, Supply-Risk × RPN 3×3 heat map, Quick Actions (Export Full Report / Save / Reset — wired to existing handlers). Filled by renderSpRails() appended to recalcAll (original body untouched).
Workflow-Overview flowchart hidden (not in the reference). Hero-banner band added at top (assets/spares-readiness-calculator-hero.webp via js/rz-hero-fit.js blur-letterbox; collapses cleanly to nothing until the owner adds the image — no dark void). Day + night both screenshot-matched to uiux_day.png / uiux_night.png. All calc functions + element ids preserved.
v1.115.2PATCH
LTC Modelling Lab — pixel-match rebuild to reference: real P&ID + de-terminal skin + gate fix
Changed
SYSTEM OVERVIEW is now the reference photorealistic P&ID. The generic outlined-box SVG was replaced by the actual reference equipment render (dry-cooler bank → pump → plate heat-exchanger → cold-plate racks, blue supply / red return pipes with valves), cropped from the reference mockups into assets/ltc/pid-day.png + pid-night.png and theme-swapped live. Only the Supply/Return temperature chips are overlaid, live-wired to the model (they cover the baked labels). This closes the owner's "P&ID tidak sama" report.
De-terminalized the whole dashboard skin. Every JetBrains Mono declaration in css/ltc-lab.css (~20 rules, incl. all KPI numbers) is now Inter sans to match the reference's clean SaaS look — no monospace, no thin-outline instrument aesthetic. Font Awesome icon fonts untouched.
Right rail matched to reference: COMPLIANCE & STANDARDS rows are now green-check circles + "Compliant" (ASHRAE TC 9.9 / ASHRAE 90.4 / ISO 14644 / ANSI/BICSI 002 / NFPA 75 / Uptime Institute), not score-bars; DESIGN STATUS radial is the reference blue ring + word grade (Excellent/Good/Fair). All still wired to model.scores.*.
Fixed
LTC page false access-gate lock ("keep asking Sign In"). The tier gate flashed "Pro or Educator access required" for a logged-in owner because _rzAuth/_rzFeatures load after the gate's first checks and the heavy dashboard init widened that window. Added an email-allowlist fast-path in the page's inline gate script: an allowlisted root email (bagus@/admin@resistancezero.com) unlocks on the very first synchronous check, before _rzFeatures exists — no gate flash. Logged-out still gates (login modal renders on top, z=99999).
Initial-paint gap carried from v1.114.x: the new dashboard now renders the cached model at the end of init() (defer ordering had the engine's first applyModelToUI fire before the scaffold existed).
Notes
Backend math untouched — computeModel and all solvers unchanged (the only engine-file edit remains the additive return fields + hook from v1.114.0). Every KPI is engine-wired (values differ from the mockup's illustrative figures because they are computed from the live default inputs).
v1.114.2PATCH
Fix: login state now persists on every page — real root cause
Fixed
Login button wrongly shown on pages after logging in (recurring; the earlier ?v= cache-bust fixed a real drift but not this). TRUE root cause in auth.js: pages that carry their OWN hardcoded nav auth markup (#navAuthWrap with navLoginBtn/navUserBtn) — calculators, spares-readiness-calculator.html, dc-market-tracker, etc. — make injectAuthButton() skip injecting the shared rzLoginBtn/rzUserBtn (line 543), and then updateAuthUI() early-returned because those shared buttons don't exist (line 634) → the page's own Login button was never switched to the account widget even with a valid localStorage.rz_premium_session. Calculators masked it with their own local sync; spares had the markup but no sync, so it always showed "Login". Fix: updateAuthUI() now falls back to the page-local nav* button/avatar/email/badge/dropdown IDs when the shared rz* ones are absent — one additive change that syncs auth state on every page carrying #navAuthWrap, permanently (idempotent + safe on pages that also self-sync). Verified headless on spares + a manual page + tco. Cache-bust: auth.js?v=2026-07-16b → 2026-07-23 unified across all 155 pages that load it.
v1.114.1PATCH
LTC Modelling Lab — mockup topbar
Added
LTC Modelling Lab topbar (mockup match) — js/ltc-lab-ui.js now injects (non-destructively, via buildTopbar()) the reference top navigation to finish the pixel-match: a center nav (Home / DC Solutions / Technical Manuals / LTC Modelling Lab [active underline] / Contact — all real destinations) and a right action cluster — What's-new bell → changelog.html, report-doc icon → REPORT tab, Save Scenario → the existing #saveScenarioBtn (or focuses the scenario-name field on the REPORT tab when unnamed), Export Report → the existing #exportExecutive. Every control wires to a REAL existing handler — no fake buttons. The existing theme toggle, auth widget, and mobile hamburger are preserved: the center nav + action cluster are desktop-only (≥ 981px), the legacy back-links are hidden on desktop and remain in the mobile drawer. Styled for both themes under body.ltc-revamp in css/ltc-lab.css. Fixed an initial-paint gap from the same session: because both the engine and the UI module are defer, the engine's first applyModelToUI fired the __ltcModelHook before the dashboard scaffold existed, so the new KPI strip/schematic/right-rail showed placeholders until the first input — init() now re-renders the cached model once the scaffold is built.
css/ltc-lab.css + js/ltc-lab-ui.js — a JS-driven visual reflow of ltc-system-modelling-lab.html into a pixel-matched 3-column dashboard (INPUTS rail · center breadcrumb/title/badge/SYSTEM OVERVIEW schematic/KPI strip/tabs · right rail COMPLIANCE·DESIGN STATUS·SENSITIVITY·SCENARIO INFO), for both light and dark themes. All new CSS is scoped under body.ltc-revamp (light :root:not([data-theme="dark"]), dark [data-theme="dark"]). The backend math is untouched — the UI module is additive, builds its scaffold on DOMContentLoaded, and RELOCATES the existing input/output panels into the new tab panels (RESULTS · DETAILED CALCULATION · FLOW DIAGRAM · CHARTS · REPORT) via appendChild, so every existing element id / chart SVG id+viewBox / the <script src> list / version stamp / #rootGate+#rootLoginBtn are preserved.
6 primary sliders (IT Load / Liquid Capture / Supply Temp / Return Temp / Pump Head / Pump Efficiency) two-way bound to the existing number inputs (inItLoad/inLiquidCapture/inSupplyTemp/inReturnTemp/inPumpHead/inPumpEff) — moving a slider mirrors the numeric box and dispatches input/change on the underlying input so auto-run recomputes. 3 architecture preset cards set inCoolingArchitecture + fire change. An OTHER PARAMETERS collapsible holds the remaining ~45 inputs.
Wired renderers invoked via a 1-line additive hook at the end of applyModelToUI (window.__ltcModelHook): KPI strip (IT Load / Liquid Cooling / Total Flow / Pump Power / PUE / System COP), theme-aware SYSTEM OVERVIEW SVG schematic (dry-cooler → distribution loop w/ center pump + blue supply / red return animated flow lines → CDU/HEX → IT-load rack, prefers-reduced-motion aware), RESULTS tab (FLOW & TEMPERATURE · PUMP & HYDRAULIC · PERFORMANCE · POWER BREAKDOWN donut · TEMPERATURE PROFILE line chart · ENERGY BALANCE table), COMPLIANCE list (6 standards, compliant when the mapped sub-score ≥ 70), DESIGN STATUS radial gauge (scores.total + grade + 4 sub-bars mapped from the 5 sub-scores by averaging), SENSITIVITY top-5 impact-on-PUE bars, and SCENARIO INFO. Every displayed number derives from the engine model object — no hardcoded KPI literals in markup.
assets/ltc/ — vendored 4 clean engineering-symbol glyphs from the RZ Engineering Symbol Atlas (tabler outline: pump.svg, heat-exchanger.svg, dry-cooler.svg, rack.svg, currentColor-recolorable).
Changed
js/ltc-system-modelling-lab.js (additive only — NO formula changed): exposed already-computed liquidCop, airCopEff, massFlowKgS on the top-level computeModel return object (previously only in internals), and added the window.__ltcModelHook(model) call at the tail of applyModelToUI.
ltc-system-modelling-lab.html (2-line edit): load css/ltc-lab.css?v=2026-07-22-ltc after the inline styles and js/ltc-lab-ui.js?v=2026-07-22-ltc after the engine script. Everything else preserved.
spares-readiness-calculator.html — "Critical Spares Engine" reference reskin (day + night): re-pointed the page's amber accent tokens to a reference instrument-blue design system and restyled the hero (breadcrumb + title + Analysis-Summary badge + Analysis-ID/Created/Scenario meta trio), the 8-module KPI strip, tabs, module/result cards, tables, badges, and buttons to the reference's flat-card visual language — sampled palette (day #f6f7f9/#1c2534, night #061119/#f1f5f9), all 25 tabs kept. Added a footer RZEngine v2.5.1 stamp (reads RZEngine.DATA.version), analysis-header meta (auto Analysis-ID/timestamp), active-KPI-card sync to the current tab, and count-up/entrance micro-interactions (no sound; prefers-reduced-motion safe). Engine wiring: the statistical kernels are now single-sourced to RZEngine.models.spares.{normInv,normCdf,poissonCdf} (local implementations kept as fallback) — the "update the engine → updates everywhere" contract; all module algorithms preserved unchanged.
tco-calculator.html — engine-wired + dropdown parity: market rates (power $/kWh, colo $/kW-mo, CAGR) now sourced at load from RZEngine.data.markets (syncMarketRatesFromEngine, byte-identical to the inline seeds); annual power cost and cost/kW-yr delegate to RZEngine.models.opex.powerCostAnnual / models.tco.costPerMwYear via guarded helpers with inline fallback — golden-compare identical, no result drift. Dark-mode <option> styling added to .tco-input-field selects (matched the canonical .input-field look; blanket rename avoided per the v1.4.1 regression note). .tco-page-background → inset:0 hardening.
Fixed
Auth login state not persisting on manual/* pages (recurring): all 40 manual/*.html loaded a stale cache-busted auth.js?v=2026-07-14a while the rest of the site loaded the current ?v=2026-07-16b (Supabase-primary login + updated session handling). Returning users ran the old cached auth.js on manual pages → the nav showed "Login" even when logged in. Re-stamped all 40 to ?v=2026-07-16b; headless-verified the account widget shows and Login hides with a seeded session. Root cause was cache-bust drift, not the injection logic.
v1.113.3PATCH
article-28 editorial overhaul: skin compliance + real source links + research-backed depth + OG photo card
Fixed
article-28.html reading layout — the page was missing the <div class="container"> wrapper inside <article class="article-content"> that every compliant editorial article uses (see article-27.html). Without it the prose ran full-bleed (no centered column, no L/R margins) and the shared editorial runtime could not build the sticky related rail (it keys off .article-content > .container). Added the wrapper → prose now sits in the centered reading column with margins, and the "Related" rail hydrates from the new Continue-Reading block. This is the true root of the recurring "no margins / doesn't follow the article standard" report (the earlier .article-body wrapper alone was insufficient).
Changed
article-28 skin → editorial compliance (rejected-pattern #7/#9): removed the gradient drop-cap (linear-gradient first-letter → solid --rz-art-accent), flattened the gradient h2::before accent to a 2px solid rail, and deleted the bespoke hand-rolled scroll-progress bar + its script (the shared js/rz-article-editorial.js runtime owns read-progress). Body callouts were already editorial-flat via css/rz-article-dark.css. Wide 7-column data tables now scroll inside the reading measure (.ws-lever-table clamped to min(--rz-measure, 880px)) so they no longer sprawl past the prose (fixes the responsive wide-article-table gate; note this site's editorial measure is 46rem @ 20px root = 920px).
article-28 content depth (owner: "harus sangat2 lebih detail", concrete max limits + conditions + per-process cost/duration/needs/risk): added a hardware-ceiling table (what fits on 16GB/24–32GB/70B/CPU-only, with the Q4 ≈ 0.5GB/1B rule), expanded the local-AI pipeline table to needs·cost·time·ROI·risk with real figures (QLoRA 65–70B on one 48GB GPU ~24h/~$15–30, RAG < ~100K interactions/mo cheaper than fine-tune, MCPVerse best-model 57.77%), a provider-notes table (Claude MCP / ChatGPT function-calling / Ollama local GGUF / Gemini long-context grounding), concrete BitNet 70B numbers (4.1× faster · 8.9× throughput · 7.16× less memory · 71.4× matmul energy, from-scratch-only caveat), DistilBERT (97% at −40% params) + dataset-distillation (50 IPC → −5.9% on ImageNet-1K), and rewrote the 3 generic FAQ into concrete max-and-conditions answers. Honesty guardrails: the "~4,000×" density figure is labeled illustrative (the solid, sourced number is corpus→FP16-70B ≈ 100–150×; cascade chart + CSV corrected from ~60 TB to ~15–20 TB), "30–60%" prompt gain labeled reported, Extropic "~10,000×" labeled aspirational.
article-28 sources — every reference is now a real clickable link (arXiv abs/, DOI, publisher pages) with [proven]/[reported]/[frontier]/[claim] confidence tags (owner: "sources reference ini tidak ada linknya"). Fixed the wrong glossary anchor on Landauer and corrected the series-nav (was mislabeled "Article 23"/prev article-22 → Article 28/prev article-27).
Open Graph index card (tools/build-og-images.py → assets/og/index.webp) — the generic gradient card now composites a circular profile photo (assets/profile-photo.jpg, anti-aliased mask + brand hairline ring) over an index-aligned dark hero (near-#0a0e1a base + low-opacity gold/mint aurora, RZ mark, resistancezero.com footer). Graceful fallback if the photo is absent.
v1.113.2PATCH
Fix: mobile hamburger appearing + misfiring on desktop
Fixed
The injected mobile hamburger (js/rz-mobile-nav.js) was hard-coded with inline display:inline-flex !important, which overrode the stylesheet's desktop .rz-nav-burger{display:none} rule — so on pages without a pre-existing .hamburger (e.g. the LTC labs) the burger appeared at desktop width and, when clicked, set body{overflow:hidden} while the drawer CSS only exists ≤768px → no menu opened and the top HUD collapsed. Now the injected burger's visibility is gated by matchMedia('(max-width:768px)') (+ change listener), so it shows only on mobile and never misfires on desktop. Cache-busted js/rz-mobile-nav.js?v= across 119 pages.
v1.113.1PATCH
DCMOC sidebar color restored — semantic lifecycle-phase coding
Fixed
Sidebar lost its color (owner report) — the v1.110.0 de-slop killed the 13-color engine rainbow but overshot to all-grey (only the active engine had amber), so the "13-Layer Lifecycle" nav read colorless. Restored color by lifecycle phase (not the rejected rainbow): engine badges + icons now colored by phase — Plan & Design (1-5) instrument-cyan, Deliver (6-7) signal-amber, Operate (8-10) data-green, Optimize (11-13) mint. Color now MEANS the lifecycle stage (design.md §5); the active engine keeps its solid signal-amber "you are here" chip. Shell.tsx only, no engine change. tsc 0 + build, screenshot-verified.
v1.113.0MINOR
dossier Permitting Matrix → EPC-grade register
Added
DATA.dossier.permittingMatrix deepened — 10 → 21 permits across the full DC lifecycle (zoning/EIA/building/fire/grid-interconnection · excavation/crane/confined-space/hot-work/working-at-height/traffic/temp-power/temp-water/fuel/hazmat · water-discharge/MV-HV-energization/telecom · occupancy/operational-license/periodic-inspection). Each permit carries the full EPC fields: phase, purpose, requiredDocs[], submissionFlow, durationWk, dependency, approvalRisk, inspection, renewal, standard. STANDARD-PRACTICE reference (jurisdiction-agnostic; validate against the AHJ).
Dossier Section 2 render — flat table → detailed per-permit blocks grouped by lifecycle phase (Pre-Construction / Construction / Energization / Occupancy-Operation), each showing purpose · required docs · submission flow · dependency · inspection · renewal · standard + risk chip + duration. Flat-table fallback for older engines. Dossier 380 KB → 423 KB.
Platform post-drafts for Article 28 in Article/Post Draft/Article 28/ (LinkedIn, X thread, Mastodon, Medium long-form, Facebook) — completes the CONTENT_LINKAGE_PLAYBOOK §1 distribution mandate.
Rebuilt llms-full.txt (130 pages) so the AI-content mirror includes Article 28.
v1.112.1PATCH
Article 28 deepened — more detail + comprehensiveness
Changed
Expanded article-28.html substantially: +2 sourced charts (the ~60 TB corpus → ~14 GB capability cascade on a log scale; efficiency headroom of analog/neuromorphic/reversible/thermodynamic/brain substrates), +2 tables (a substrate comparison matrix and the full 9-stage local-AI pipeline), a Capability Density formula callout, a concrete no-GPU recipe (ollama + qwen + persistent chat + RAG + tools, mapping to the site's own local stack), an in-body FAQ, and a Sources & further reading list. Now 4 charts + 3 tables + SVG diagram, ~2,900 words.
New datasets data/article-28/compression-cascade.csv + substrate-efficiency.csv (both provenance-tagged).
article-28.html — "The Compression Horizon: how close can a laptop get to a datacenter's mind?" (Future Forward). A rigorous, bold-but-fenced take on two impossible-sounding targets — 300 TB of data into 1 KB, one CPU-only PC equal to a 1 GW datacenter — mapping the hard walls (Shannon, Kolmogorov complexity, Landauer's principle) against the real frontier that climbs them (dataset distillation, BitNet b1.58, implicit neural representations, hyperdimensional computing, Matryoshka embeddings; reversible/analog memristor/photonic/neuromorphic/thermodynamic compute; algorithmic-efficiency halving). Introduces the Capability Density / Compression Horizon framing, a default→genius local-AI pipeline with an inline SVG block diagram, 2 sourced charts (bits-per-weight; compute-per-capability over time), and a myths-vs-reality confidence table.
Cross-linkage: prepended to insights.html (Latest Publications) + articles.html grid; 6 new glossary terms (Shannon entropy, Kolmogorov complexity, knowledge/dataset distillation, hyperdimensional computing, memristor, RAG); search-index.json (id 46) + search-sections.json; sitemap.xml (145 URLs) + llms.txt; OG card assets/og/article-28.webp; two provenance-tagged datasets in data/article-28/.
Verified
Headless: 2 charts render, SVG pipeline diagram, 8 sections, 0 console errors; article-charts provenance gate CLEAN (source+basisTag on both); dark-coverage CLEAN; script-tags CLEAN; article-28 passes version-stamp + mobile + responsive. (Pre-existing gate noise on tools/dc-corpus/raw/* version stamps and setup-supabase.html mobile overflow is unrelated and untouched.)
v1.111.2PATCH
SecondBrain → Claude Session Manager entry pill
Added
"Sessions" entry pill in the SecondBrain graph nav, next to "Vector Index" (amber→green filled-glow CTA, so the two portals read as a pair) → opens the local Claude Session Manager at http://localhost:8770/.
The manager itself is a local-only tool (lives in ~/.claude/rzsessions/, auto-started by a systemd user service; not in this repo): a dark, reference-grade 3-column web GUI that lists every AI-CLI session on the PC — Claude (~/.claude/projects) + Codex (~/.codex/sessions) + persistent ollama chats — each with the correct provider logo, title, model, folder, event count, and a per-session Resume button that opens a terminal running claude --resume / codex resume / rzchat --resume in the session's cwd. Search, provider/model/folder filters, sort, and a detail panel (Summary + Key Topics + Metadata). Backend is bounded-parse (never full-reads the 80 MB+ Codex logs) + mtime-cached; localhost-bound, UUID-validated resume.
Verified
Headless: pill renders next to Vector Index → localhost:8770, 0 console errors. Manager: 71 sessions (8 Claude + 63 Codex) render, provider filters + detail panel + dry-run resume all work; favicon 204 (no stray 404). systemd user service active + enabled. (Session data never leaves the PC; nothing session-related is committed to the repo.)
v1.111.1PATCH
worked-calc accuracy fixes — adversarial review
Fixed
Fire-water density mislabeled (HIGH) — the sprinkler-demand calc tagged "NFPA 13 OH-2" but used 12.2 L/min·m² (that's Extra-Hazard-1). Corrected to the true OH-2 density 8.15 L/min·m² (0.20 gpm/ft²; 1 mm/min = 1 L/min·m²). (Only surfaces for water-suppression projects; novec default = zeros.)
Availability rendered "100 %" (MED) — a redundant chain rounds to 7+ nines and the renderer caps at 4 decimals → an unqualified, physically-impossible "100 %". Now clamped to 99.9999 % (4-nines display); the Annual-downtime step carries the real per-topology figure. _availRaw (the reliability-model match) unchanged.
Transformer step-3 unit mix (MED) — the count step used kW while the prior steps were MVA/MW, breaking the audit trail; now ⌈facMVA / unitMVA⌉ (all MVA).
WUE step not self-contained (MED) — the annual-water step referenced IT_kWh with no kWh value shown; now the IT-energy step outputs kWh and the water step shows the full WUE × kWh ÷ 1000 arithmetic.
Disclosure notes (LOW) — UPS autonomy flagged "(screening default — validate vs project UPS spec)"; availability inputs note the 3-group (ups/crac/chiller) screening scope vs a full IEEE 493 RAM study.
Verified
engine 747/0 (_availRaw still == models.reliability) · parity 155/0 · bindings 85/0 (catalog 221fn/123src) · export 44/0 · tsc 0 + build. Hand-verified UPS/cooling/fuel calcs reconcile. Availability now renders 99.9999% (screenshot-confirmed). Engine ?v -m→-n. 0 CRITICAL in review; 1 HIGH + 3 MED + 2 LOW all resolved.
Owner: dossier "masih jauh" from a real EPC reference. Deepened engine-sourced across 4 sections (dossier 272 KB → 380 KB, all numbers track the live project).
Added — Engineering Calculations = WORKED CALCS (Section 4, the biggest gap)
BOQ dossier input-sync fix — dossier tracks the live project
Fixed
Dossier showed the wrong project (usa · 1 MW vs the app's Indonesia · 2.5 MW). The CAPEX store persists inputs; on reload after any CAPEX edit, onRehydrateStorage respected the STALE persisted itLoad/location instead of reconciling to the Simulation store (the SSOT the top bar reads). So buildBoqModel built the header — AND the CAPEX results the whole BOQ decomposes — on stale inputs. Now onRehydrateStorage ALWAYS reconciles the shared canonicals (itLoad, country/location) from the live Simulation store (capex-specific fields like fireType/redundancy stay persisted), so the dossier total recomputes on the live project. Belt-and-suspenders: withProjectMeta + all 3 dossier callers (CapexEnginePage, ConstructionEngine, TraceValue) now pass location/itMw from the Simulation SSOT (same source as the top bar), so the header can't diverge.
Verified
Seeded headless test (persisted capex = usa/1000, sim = ID/2500) → dossier header now reads Indonesia / 2.5 MW (was usa / 1 MW). tsc 0 + build · trace-parity 117/117 · walk 24 · export 44. No engine change (DCMOC-only).
v1.110.1PATCH
deep-BOQ correctness fixes — adversarial review
Fixed
Seismic zone now consistent + functional (HIGH) — the Seismic Zone dropdown never affected CAPEX cost (CapexEngine hardcoded zone2), while the BOQ take-off read input.seismicZone (default zone1) — so civil rebar/bracing quantities were calibrated to a different zone than the cost basis. Now CapexEngine reads input.seismicZone (the dropdown works: zone0 0.2× → zone4 8× on the seismic category), BOQ driver + store defaultInputs both default zone2 — all three agree. (Deliberate CAPEX re-baseline for non-zone2 selections — the control now does something.)
Lump-fallback line coherence (MED) — the zero-load reconciliation fallback hijacked the first real component row (qty 0 / rate 0 / full total); now pushes a dedicated Lump sum line (qty 1) so the discipline tree reads coherently.
Removed 3 dead unitRates keys (MED) — securityPerRack/cxPerMw/permitPerMw were pre-deepening anchors referenced by no leaf (a future author using securityPerRack would get a wrong $1,500/rack vs the real acs+cctv+intrusion path).
Four-phase DCMOC overhaul (owner: dossier too shallow "3 levels deeper" + "design warna terlalu AI slop, harus intuitive" + KPI cards uneven/overflowing).
Vector Index dashboard (Apps/second brain/vector-index.html)
Smooth cluster density replaces the ugly overlapping convex-hull polygons — d3.contourDensity() soft isobands per client-cluster, cached in world space so zoom/pan stay smooth.
Constant-size category labels (screen space, fade out on zoom-in) instead of labels that ballooned when zoomed.
Live left-panel controls: Namespace (collection filter), Time Range (mtime filter), and Cluster Algorithm (client re-cluster on the 2D projection — KMeans / DBSCAN / agglomerative) now actually drive the canvas. Similarity-threshold fade + smooth animated zoom buttons.
3D enhanced + interactive: raycaster hover-tooltip + click-to-select fills the right panel (was inert); round sprite points, additive glow, depth fog, auto-orbit (pauses on interaction), fit-to-bounds.
Detail + scalable + robust tabs: shared sortable / filterable / paginated table (Documents · Chunks · Embeddings) with row-click → locate the point in the visualization; richer Overview (cluster mini-table · recency · token stats) with distribution bars whose width equals the % shown; Analytics gains a similarity-to-centroid histogram (oversized SVG charts replaced with bounded bars); full Settings health (index · projection · per-collection · cluster-size tables + copy rzmem viz). Right panel adds a vector sparkline + nearest-neighbors. Tab reveal animation, Esc clears selection, tabular-nums, solid gate scrim (no glassmorphism).
Fixed the toolbar "aneh behaviour" root cause: the empty-state #emp lived inside the vis-network container, which wipes its children on first render — so every later rebuild (Force / filter / search / focus / orphan) threw Cannot read properties of null (reading 'style'). Moved #emp out of the container + defensive guards → all controls work without crashing. Hierarchical layout now auto-fits (no more off-screen "explosion").
Prominent "Vector Index" entry CTA: filled cyan→violet gradient pill with glow + shine (replaces the tiny buried icon), clearly signalling a major feature.
China duty band (HIGH) — CN.importDutyBand was high (17%) but DATA.sources documented punitive (~30% China↔US). For AI DCs the frontier equipment is predominantly US-origin (Section-301 / retaliatory tariffs dominate), so CN is now punitive — matches the documented intent + the realistic worst-case, and retires the previously-unused punitive band.
GB300 cooling-kit duty (HIGH) — the $50k/rack cooling kit is added after the CapexEngine cost loop, so it bypassed landedFactor — import duty was silently zero on imported cooling hardware. Now applies landedFactor(country,'coolingKit') (new equipmentShareByCategory.coolingKit 0.90); a 72-rack GB300 in a high-duty country was ~$0.4M short.
Dossier landed table completeness (MED) — the per-category supply-chain table disclosed only 5 of the duty-receiving categories; now shows all material equipment disciplines (adds bms/security 0.60, fireSuppression/fireAlarm 0.50) so the disclosed uplift matches the cost math.
Type widening (MED) — CountryProfileLike in BoqDossier now includes gpuExportTier/customsLeadBand (were read via a cast, hidden from tsc).
Verified
engine 710/0 · parity 155/0 (after country regen: CN punitive) · calibration 19/0 · bindings 85/0 (catalog 216fn/118src) · trace-parity 117/117 · export 44/0 · tsc 0 + build. Split-rule / null-safety / country-coverage / honesty PASSED the review (0 CRITICAL). Engine ?v -i→-j. Closes the review-discipline gap on Ship-C (which shipped without a dedicated review).
v1.107.0MINOR
Vector Index dashboard revision + SecondBrain gate/lock/modal fixes
Added
Vector Index dashboard — 8 functional tabs (Apps/second brain/vector-index.html): Overview (stat cards + category/collection distribution + top documents), Documents, Chunks, Embeddings (filterable + paginated tables), Vector Visualization (the canvas), Search (semantic/text), Analytics (SVG bar charts — category, cluster sizes, collection split, token histogram), Settings. All driven from the local vector-index.json (no backend).
Prominent, labeled entry point — the SecondBrain toolbar's tiny 7-dot icon is promoted to a labeled "Vector Index" nav pill (with a padlock cue) next to Wiki.
Padlock on the "Second Brain" nav item across 61 pages (Insights dropdown) — signals root-only.
Fixed
Login modal on SecondBrain rendered with no visible fields — the gate z-index:100000 covered the auth modal (99999) and the page-blur rule blurred the modal into mush. Gate lowered to z-index:9000 and the blur now excludes #rzModalOverlay, so the standard Email+Password Sign-In modal renders cleanly on top.
Similarity Threshold slider had no visible effect — it now fades every point less similar than the threshold to the selected point / query, with a live "N / total ≥ threshold" caption.
Vector Visualization looked bare — cluster hulls + category labels now default ON, larger points with halos, a faint reference grid, brighter selected/neighbor states.
Dead controls wired — topbar bell/cog/avatar; added a "← SecondBrain" back button.
three.js pinned to a single version (0.128) for the 2D/3D toggle.
Verified
Headless (puppeteer): root unlocks both pages; all 8 tabs render real content; slider + back + tabs + zoom work; 0 console errors. Non-root context → index.html locks and the Sign-In modal shows visible Email + Password fields (z 99999, unblurred). Padlock present on 61 pages. Viz data (vector-index.json) stays local/gitignored — 0 competitor-brand rows in committed files.
v1.106.0MINOR
Ship-D: DC Corpus / KB dive-deep — new-domain sources + rack-power metric
Added
DC Corpus deepened — 12 new authoritative public sources appended (append-only) to tools/dc-corpus/sources.yaml: microfluidic/advanced cooling (Corintis, Microsoft in-chip, IMEC, arXiv power-stabilization + microfluidic-survey), AI-arch power (OCP ORV3 HPR/Mount Diablo, NVIDIA GB200 NVL72, arXiv AI-DC-power), supply-chain (US BIS EAR, WTO Tariff Profiles 2024, DCD supply-chain lead-time, T&T DC Cost Index). 10/12 fetched (2 bot-blocked, handled gracefully — remain as provenance citations).
New corpus metric rack_power_kw (extract.mjs, bounded 5-400 kW/rack) — AI-architecture rack power draw, distinct from rack_density_kw. Yielded 4 source-quoted facts (20/40/50/100 kW; IMDA, NVIDIA DGX H100, Uptime 2024). Distribution n=3, monotonic, 2 independent sources.
Verified
engine 708/0 · parity 155/0 · calibration 19/0 (constants still in-band vs deepened corpus) · bindings 85/0 (catalog 216fn/118src) · dc-corpus 2678→2700 asserts, all green · export 44/0 · tsc 0 + next build. corpus-facts 649→653 (+4). Engine ?v -h→-i. HONESTY: heat_flux_wcm2 + import_duty_pct patterns DROPPED (0 matches / trade-vs-utility-tariff false positives) rather than ship noise — arXiv PDFs converted poorly (binary), so W/cm² + GPU-TDP-W + import-duty-% don't exist in extractable form; sources stand as citations. Ship-D COMPLETE — the full CAPEX+BOQ program (Ships A/B, BOQ P1-P3, C, D) is shipped.
v1.105.0MINOR
Supply-Chain per Country: landed cost + export-control + customs lead-time
Added
RZEngine DATA.supplyChain + models.supplyChain — per-country landed-cost / export-control / customs lead-time (the pending original CAPEX ask "supply chain per selected country"). equipmentShareByCategory (import-exposed fraction per cost category: ups 0.70, generator 0.75, network 0.80, cooling 0.55, electrical 0.40; labor-heavy ≈ 0), importDutyBands (fta 0 / low 3% / med 7.5% / high 17% / punitive 30%), exportTiers (1/2/3 PROXY), frontierArchKeys, customsLeadWk. models.supplyChain.landedFactor(country, category) = 1 + duty×equipmentShare (floored 1.0 — split-rule: duty on imported-equipment fraction ONLY; local labor/civil already in constructionIndex, never double-discounted), exportControl(country, archKey), leadTimeCustomsWk, dutyRate, exportTier. Standard: standarization/SUPPLY_CHAIN_STANDARD.md.
Dangling engine references (HIGH) — DATA.dossier Design Basis + Engineering Calculations pointed structural floor-loading at DATA.requirements.floorLoadingKnM2, which doesn't exist; the key lives in DATA.architecture.floorLoadingKnM2. Corrected both (the "Engine Reference" column an engineer follows to validate the calc now resolves).
Executive-summary timeline via native engine path (HIGH) — models.dossier.executiveSummary read only result.metrics.timelineMonths; a native models.capex result (no .metrics) silently rendered "— mo". Added || result.timeline.totalMonths fallback.
Section-title single-source drift (HIGH) — dossier sections 5/6/7 hardcoded their titles instead of reading SECTION_TITLES, defeating the single-source ToC/header design. Now all sections read SECTION_TITLES[n-1].
Verified
engine 697/0 (new: timeline fallback + floorLoading-path guards) · parity 155/0 · calibration 19/0 · bindings 85/0 · trace-parity 117/117 · export 44/0 · tsc 0 + build · script-tags CLEAN. Engine ?v -f→-g. Found by an adversarial review of the P3 dossier (0 CRITICAL, 3 HIGH resolved; MED/LOW were non-issues or false positives — models.maintenance does exist).
DCMOC full dossier (BoqDossier.ts) — the BOQ is now section 6 of a numbered 10-section "Technical Project Dossier — Bill of Quantities & EPC Reference" with a Table of Contents (shared SECTION_TITLES single-source so numbers/titles can't drift): 1 Executive Summary (KPI grid, live CAPEX) · 2 Regulatory & Permitting Matrix (risk chips) · 3 Design Basis · 4 Engineering Calculations · 5 Equipment Schedule · 6 Bill of Quantities (reconciled + disclaimer/margin/safety-factor) · 7 Procurement Packages · 8 Risk Register · 9 Operations Readiness · 10 Document Register. Dossier-level STANDARD-PRACTICE banner ("permitting durations/risks indicative; validate against AHJ + full design; not a quotation/tender").
Verified
engine 695/0 · parity 155/0 · calibration 19/0 · bindings 85/0 (catalog 211fn/117src) · dc-corpus 2678/0 · trace-parity 117/117 · walk 24/0 · synergy 6/0 · export 44/0 · tsc 0 + next build · js-syntax/script-tags/dark CLEAN. Exec summary pulls live CAPEX; 6 static datasets resolve; no </script> hazard. Engine ?v -e→-f. Grand plan COMPLETE: BOQ P1 (reconciled line items + margin/safety) + P2 (equipment + procurement) + P3 (full EPC dossier) all shipped.
v1.103.1PATCH
BOQ correctness fixes — adversarial review
Fixed
BOQ cooling-quantity overcount (HIGH) — models.boq.drivers.coolingKw (and equipmentSchedule) used IT×PUE, inflating the displayed CRAH count + glycol/coolant volume by a factor of PUE (~1.4-1.6×). Cooling duty ≈ IT heat load (the (PUE−1) overhead is the cooling plant's own fan/pump/chiller power, not extra white-space heat), so coolingKw = itKw. Gensets/transformers correctly stay sized to IT×PUE (they power the whole facility). Line-item $ were already correct (reconciled); this fixes the physical quantities an engineer reads off the schedule.
Margin-override label (HIGH) — models.boq.summary.marginPctGross returned the constant (10%) even when opts.epcMarginPct overrode it, contradicting the embedded-margin $ on the same disclosure. Now reflects the actual margin used.
Popup-blocked feedback (MED) — CAPEX Engine + Construction Engine BOQ buttons ignored openBoqDossier's false return; now alert the user when the dossier popup is blocked (TraceValue already handled it).
Vacuous reconcile (MED) — summary.reconciles read true at capexTotal === 0; now requires capexTotal > 0.
Zero-load reconciliation guard (LOW) — generate now carries the category $ as a lump line if bottom-up is 0 while the category $ is nonzero, preserving Σ(lines)===categoryTotal.
Verified
engine 686/0 (new: coolingKw=IT-heat, margin-override, zero-capex reconcile guards) · parity 155/0 · calibration 19/0 · bindings 85/0 · trace-parity 117/117 · export 44/0 · tsc 0 + build · script-tags/dark CLEAN. Engine ?v -d→-e. Found by an adversarial code-review pass (0 CRITICAL, 2 HIGH + 2 MED + 1 LOW — all resolved).
RZEngine models.boq.equipmentSchedule — major-equipment schedule from DATA.boq.equipmentSizing: UPS modules / gensets / MV transformers / PDU / CRAH / CDU sized by IT load × PUE with N + redundancy counts (N+1 → +1, 2N → ×2, 2N+1 → ×2+1) + lead times reflecting the 2026 long-lead reality — MV transformer/switchgear ~120 wk dominant, UPS 26, genset 40, chiller 32. Informational (indicative, not reconciled to $).
RZEngine DATA.boq.procurement + models.boq.procurementPackages — 12 standard mission-critical DC EPC packages (P01 civil → P12 permits) with scope, tender method, lead time, FAT/SAT, warranty; est value = Σ mapped CapexResult category $. Flagged non-additive (electrical/mechanical categories span several packages — indicative scope envelopes for procurement planning, NOT a cost rollup). DATA.boq.laborBurdenPct (45%, informational) added.
BOQ dossier (BoqDossier.ts) +2 sections after the discipline tables: Equipment Schedule (Equipment · Spec · Capacity · Qty N · Qty Installed · Redundancy · Lead Time · confidence; long-lead ≥52 wk amber-highlighted with LONG-LEAD tag) + Procurement Packages (Pkg# · Name · Scope · Tender · Lead · FAT/SAT · Warranty · Indicative Value · confidence; prominent non-additive-overlap caption).
DCMOC surfaces — Requirements cooling picker +3 options (immersion_1p/2p/microfluidic) with amber "EMERGING · pilot/riset" tag + microfluidic TRL~6-7 advisory note (selectable for scenario modelling, clearly not-production; true 7-key on CAPEX store drives real coolingMult/PUE, sim store gets thermally-equivalent downcast). New Data Library "Cooling Tech" tab (vendor table + COMMERCIAL/EMERGING chips + honesty caption). CDU dashboard "Advanced & Emerging Cooling" subsection (TRL badges). Microfluidic NEVER encoded as an NVIDIA architecture fact.
Fixed
Region dropdown EMEA/MENA/Africa overlap (owner report "EMEA kok ada 2") — regions.ts REGION_LABELS relabels the EMEA bucket "Europe, Middle East & Africa" → "Europe". All 14 EMEA-tagged countries are European; Middle East (MENA) + Africa (AFR) are already separate buckets, so the old label double-listed those geographies. Bucket key + CapexEngine region==='EMEA'→europe mapping unchanged; label only.
DCMOC CAPEX Ship-A: AI reference-architecture profiles + peak-provisioning
Added
RZEngine DATA.requirements.archProfiles — 5 AI reference architectures ber-sumber + confidence-tagged: NVIDIA H100 SuperPOD (~30/45kW air, OFFICIAL), GB200 NVL72 (120 nominal / 132 TDP OFFICIAL / 192 peak = EDPp ANALYST est ~1.5×, liquid), GB300 NVL72 (140/192, cooling-kit ~$50k/rack = Morgan Stanley BOM ANALYST, hybrid), Vera Rubin VR200 NVL72 (~200/300kW, ANALYST — NVIDIA belum publish rack-kW), OCP ORV3 HPR (92-140kW→roadmap 750/1MW, OFFICIAL). Plus peakProvisionFactor:1.5, baselinePeakRatio:1.2, interconnect (IB $4900 / Ethernet $2600 per GPU, ANALYST — SemiAnalysis 512-GPU cluster ÷512). Model fns models.requirements.archProfile / provisionedRackKw / powerProvisionUplift / interconnectCost. Semua leaf punya DATA.sources entry dgn confidence tag di string. Fact-checked vs NVIDIA RA / OCP spec / SemiAnalysis / Morgan Stanley — semua angka firm + label jujur.
DCMOC Reference-Architecture picker (WorkloadProfileSection) — pilih arsitektur AI mutakhir; tiap opsi tampil peak/nominal kW + GPU + cooling + Tier-floor + confidence chip (emerald OFFICIAL / amber ANALYST + sub-note "estimasi analis, bukan datasheet vendor"). applyArchProfile set density=nominal + cooling + tier-floor bump, mirror applyUseCaseProfile. Store additive archKey (default null).
CapexEngine peak-provisioning — arch terpilih menerapkan MARGINAL power-plant uplift (powerProvisionUplift = (peak/nominal) ÷ baselinePeakRatio 1.2, floored 1.0) HANYA di electrical/UPS/generator (bukan raw peak/nominal — CPU-era base $/kW sudah harga ~1.2× headroom via NEC 125% + Google power-provisioning literature, jadi raw ratio = double-count). GB200 → ×1.333. GB300 tambah cooling-kit line ($50k×rack) ke total. Interconnect IB/Ethernet = field terpisah, EXCLUDED dari infra total (fabric IT di luar scope CAPEX). 2 value-trace node baru (req.provisionedRackKw, capex.provisionedPowerUplift).
Hover tooltip sitewide (owner "banyak tooltip tidak ada"): ~85 chart Chart.js di 21 halaman user-facing (article-2/3/17/18/24/25/26/27, tier-advisor, carbon-footprint, dc-market-tracker, tia-942-checklist, water-system, geopolitics-3, FF-1/2/3, spares-readiness — melengkapi v1.99.7 tco/cx/pue/roi) dapat interaction:{mode:'index',intersect:false} (line/bar/scatter/radar) atau {intersect:false} (doughnut/pie/bubble) → tooltip muncul saat hover-dekat, bukan hanya di titik persis. Callbacks existing dipertahankan; chart yang sudah ber-interaction dilewati.
DC Hub calculator charts — fix stretching + hover tooltips
Fixed
Chart "continuous stretching" (owner screenshot tco-calculator): .tco-chart-canvas-wrap + .tco-pro-chart-wrap pakai flex:1 + height:100%!important di flex-parent tanpa tinggi tetap → Chart.js responsive membaca clientHeight → set canvas → resize loop → canvas meledak vertikal saat hover. Fix: tinggi deterministik height:340px/300px + max-height (loop putus, mirror pola terbukti .chart-container{height:280px}). Verifikasi headless: canvas 340→340 STABLE setelah 5× hover (was: tumbuh tak terbatas).
Tooltip value tidak keluar saat hover ("banyak tooltip tidak ada"): Chart.js 4.4.x default intersect:true → tooltip hanya saat kursor TEPAT di titik. Fix: interaction:{mode:'index',intersect:false} di 17 chart (tco 3, cx 6, pue 4, roi 4) — line/bar mode 'index' (semua seri di X), doughnut/pie intersect:false saja (nearest). Verifikasi: tooltip muncul on-hover-near (pue/roi ✓ headless). opex-calculator TIDAK disentuh (standing order).
Adversarial verify v1.99.5 (deploy): Diagnostics Center struktur/deep-link/skip-chip + Trace English + 3 enhance (Copy chain/Expand all/leaf summary) semua PASS; menemukan 5 fragmen ID sisa di Center → difix di ronde ini.
DIAGNOSTICS CENTER + Trace notes → English + enhance
Added
Diagnostics Center (tab SUPPORT baru): agregasi 12/13 collector diagnostik ke satu permukaan — semua indikator merah/amber aktif lintas engine, dikelompokkan per severity (N critical · M warning), tiap temuan → kartu (surface chip, metric, angka value/threshold, detail) + tombol "Open →" setActiveTab ke halaman sumber. lib/diagnostics.ts type kanonik DiagFinding + normalizer (menyatukan 4 bentuk Finding berbeda antar file) + registry (tiap run membangun model dari store + adapter/engine chain SAMA dgn halaman; try/catch per collector). MonteCarlo di-skip JUJUR (10k iterasi sinkron = jank) + chip "skipped (honest)". Empty-state jujur "No active red indicators — everything within limits ✓". Cakupan 12/13 dinyatakan.
Trace enhance (owner): tombol Copy trace chain (seluruh rantai formula+nilai+sumber tree text → clipboard), Expand all (buka seluruh pohon dep sekaligus + filter saat >6 dep), leaf composition summary ("3 inputs · 2 engine constants · 1 screening").
Changed
Trace notes DCMOC → English (owner "change it in english"): 147 node value-trace.ts (label + formulaTemplate kata penghubung + provenance) + seluruh UI string TraceValue.tsx ("Your input / Engine (sourced) / Computed / Screening estimate", "This number is computed from:", "Open / edit in tab", "External source", tooltip) → English natural. traceId/deps/sourceKey/external.href/nilai UTUH → trace-parity tetap 117/117. Diagnostics Center juga English.
Verified
tsc/build · walk 24/0 · synergy 6/0 · trace-parity 117/117 (teks-only, nilai popover tak berubah) · export 44/0 · bindings 85/0 (catalog regen — Center consumer baru) · screenshot Center + trace-EN.
v1.99.4PATCH
Diagnostics Tier-2/3/4 — SEMUA indikator merah kini ber-guidance
DisasterRisk High/Extreme → dekomposisi per hazard engine-live + trade-off asuransi vs hardening vs EAL (basis ROI mitigasi diverifikasi reproduksi angka engine).
FuelGen CO₂ → HVO swap delta live + abatement $/t vs harga carbon negara (DATA.envCosts) + offset compliance/voluntary + lever run-hours (re-run grid engine).
Strategic bottleneck → formula feasibility di-extract single-source + lever grid/land marginal (MW per MVA/acre); Talent Very-Difficult → matriks premium via re-run TalentAvailabilityEngine nyata (temuan jujur: time-to-staff engine tak sensitif salary → note).
ScenarioComparison arah warna delta kontradiktif (CAPEX naik = hijau di KPI vs merah di tabel halaman sama) → merah = lebih buruk konsisten (higherIsBetter IRR/NPV + invertDelta CAPEX/OPEX/PUE/Staff).
Bug #333 kambuh: revenue $120 hardcode di ScenarioComparison → DEFAULT_REVENUE_PER_KW_MONTH 280 single-source (IRR/NPV komparasi kini konsisten dgn MC & surface finansial lain).
ReportDashboard insight statis ("NPV positif <3yr payback" bisa bohong) → rule-derived dari financialResult live.
Beberapa dashboard: useEngineReady() dep fix (recompute saat engine telat load).
Verified
tsc/build · walk 24/0 · synergy 6/0 · trace-parity 117/117 · export 44/0. Diagnostics program COMPLETE (kontrak collector siap Diagnostics Center — DEFER sadar).
DCMOC: lib/cloudProjects.ts (size-guard 240KB pesan jujur tanpa auto-trim; validasi bundle version+10-slice; error tabel-belum-ada → link Owner Action Board), section "Cloud (opsional)" di Projects (label jujur "localStorage tetap primer, tanpa sinkron otomatis"; badge ☁ tersinkron; Share modal salin URL + Cabut), view-only ?share=TOKEN: backup viewer ke sessionStorage DULU → fetch anon → restore → readOnly lock (guard store) + banner "Mode lihat-saja · Keluar & kembali ke data saya" (restore utuh + strip query); share view bypass login (pseudo-viewer), StrictMode-safe.
Added — Diagnostics Tier-1 (standarization/DIAGNOSTICS_STANDARD.md baru; kontrak collectDiagnostics per file)
Financial NPV/ROI merah klik → lever bisection nyata ("Revenue +117% ke $357/kW/bln utk NPV=0; capex-saja unreachable — jujur").
Benchmark grade F/D klik → kontributor terburuk + lever PUE diskret live ("1.50 p50 → RDHx 1.18 p23, overall D→C") + lever kontinu ke band p75; chip drift → panel nilai vs band + arah koreksi + kebijakan drift + ↗ Model Calibration.
Compliance <80 klik → item mandatory yang belum + biaya per item + lever "79→93 · $10,000 initial" (formula runtime-verified vs engine).
CapacityDashboard At-Risk ≥60 klik → atribusi kontributor via neutralize-rerun + lever defer-IT/kompresi-jadwal ter-bisection.
Maintenance riskExposure >50k klik → dekomposisi ber-chip parity "≡ engine" + lever upgrade SLA (net hemat/tambah vs DATA.omContracts) + lever desain Tier IV.
Semua: threshold satu konstanta dgn pewarnaan; honest-unreachable; ekspor collector.
Guided tour DCMOC (TourOverlay.tsx): 8 langkah spotlight (context bar → 13 engine → angka ƒx → panel "kenapa?" → chip rec: → Data Library → Knowledge Base → Export PDF) — teknik spotlight spares di-port ke React portal, positioning clamp viewport + fallback center, Esc/klik tutup, a11y dialog+focus; auto-launch sekali (flag dcmoc.tour.v1), tombol ? replay di header. Probe di-seed flag agar tak terhalang.
Chip global "⚠ perubahan tidak tersimpan" di Shell header — usePersistHealth() subscribe flag persistFailed 5 store tracking; tooltip menyebut store yang gagal.
Changed
Responsive 10 titik (17 baris, 8 file, desktop tak bergeser): KPI grid clamp xl (Executive 7-col, GridReliability 6-col), nested grid md:, min-w semua tabel lebar (AssetLifecycle 720px, ScenarioComparison/Report/Investment 420-640px ×11 tabel), gap sm: FuelGen, label utilization w-24 sm:w-28 truncate.
Verified
tsc/build · walk 24/0 · synergy 6/0 (probe seed diperbaiki — sed arrow-body) · trace-parity 117/117 · export 44/0 · screenshot tour step-3 spotlight ƒx.
v1.99.1PATCH
SHIP 2 Arc-2: KORPUS LOOP HIDUP + sumber APAC/Indonesia
Added
3 extractor pattern baru (+unit test positif/negatif di gate, guard "$/kWh"): construction_months, capex_usd_per_kw, rack_density_kw — fakta jadwal konstruksi NYATA pertama masuk (EIAR Google Dublin 27 bln, Vantage 30/60 bln) + rack density Uptime (7/9/50 kW).
9 sumber APAC/Indonesia (7 ter-fetch): DCI Indonesia (IDX), NeutraDC Nxera Batam 18→54 MW (Telkom), Komdigi PDN 2024, IMDA Green DC Roadmap (300 MW, PUE ≤1.3) + factsheet, NEXTDC FY24 (capex A$1B, 172,6 MW contracted), YTL Kulai 500-600 MW, METI Energy Plan 2025. Gagal jujur: BKPM (TLS chain rusak — dicatat), 1 lainnya. fetch.mjs +flag per-source insecure/referer.
corpus-facts.json kini output pipeline deterministik (aggregate.mjs; sort stabil, quote 240 char) — langkah manual dihapus; diff migrasi diverifikasi (+fakta baru saja, 0 hilang). Konsumen Data Library dibetulkan (count → facts.length).
README recipe refresh lengkap + kebijakan drift.
Verified — LOOP TERBUKTI
Regen penuh (fetch→extract→aggregate→terser→catalog) → dc-corpus 2491/0 (fakta 597→649, research docs 35→39) → kalibrasi tetap 19/0 (band bergeser mengikuti korpus, engine tetap in-band — inilah loop-nya).
Catatan metodologi jujur: mapping-4 di-downgrade dari rencana "fleet-PUE inferens" ke "koherensi capacity-factor" — PUE tak dapat diinferensikan sahih tanpa energi IT per dokumen.
IT Load (Requirements) tidak bisa dihapus/ketik ulang digitnya (owner): NumInput fully-controlled + konsumen hanya commit bila valid (≥100 kW) → tiap ketikan di bawah ambang snap balik. Fix: draft buffer saat fokus di NumInput (berlaku semua field angka Requirements) — bebas kosongkan/ketik, commit tetap tervalidasi, blur mengembalikan nilai valid terakhir bila input tak sah. Probe headless: clear ✓, 1 digit ✓, komit desimal ✓.
v1.98.5PATCH
null-honesty 16 family + hardening 5 store tracking + smoke produksi
Fixed
Sweep null-honesty narasi (16 family): 3 klaim palsu lagi terbunuh — Operations "Healthy · 100% PM Compliance" saat belum ada log PM → "Plan Mode — belum ada tracking"; Site Intelligence "0/100 · Challenged" + false HIGH sebelum analisis → "Not Analyzed"; Financial NPV null → "Not Evaluated" (bukan "$0 · NPV negatif"). 13 family lain diverifikasi aman (metric selalu terhitung); jalur nilai-nyata byte-identical. Konsumen PDF + panel on-page ikut jujur.
Hardening 5 store tracking (sisa audit A 19-Jul): opsLog/cxTracking/financialTracking/constructionTracking/sustainability — payload localStorage versioned {version:1} + migrasi legacy transparan; quota-fail tidak lagi senyap (flag persistFailed non-persisted + warn dev sekali + auto-clear saat pulih); cap FIFO 500 pada semua array unbounded (alarms/incidents/tickets/issues/punch/revisions/initiatives); seed hygiene diverifikasi (seed hanya first-run, tidak pernah menimpa data user).
Verified
Smoke produksi live (resistancezero.com): manual hub tampil 33 manual · pill "Technical Manual ↗" live di kalkulator · dcmoc serve normal.
Theme manual/ sinkron sitewide: 41 halaman manual membaca key theme sitewide (fallback rz_theme legacy) dan menulis keduanya — pilihan dark/light kini konsisten antara manual/ dan seluruh site (pre-existing sejak halaman manual dibuat).
3 straggler biner .html di tools/dc-corpus/raw/ (SEC 10-K pre-hardening) dihapus → audit-version-stamp strict kembali exit 0.
Changed
llms.txt regen: +41 entri manual (170 halaman).
Allowance ShiftEngine didokumentasikan eksplisit sebagai shift-pattern-based BY DESIGN (bukan gap country-data): benefits per negara sudah di benefitsOverheadRate/socialSecurityRate — tabel negara di allowance = double-count. Menutup item MEDIUM terakhir inventori DM.
Section manual/ (41 halaman technical manual) de-orphan: sitemap 115→156 URL (generator build-sitemap sudah include, tinggal stale — regen); 39 inlink baru (footer+dropdown index.html "Technical Manuals" + pill "📖 Technical Manual & Methodology ↗" di 37 halaman tool, theme-aware AA; opex-calculator TIDAK disentuh per standing order); search-index 98→139 entri (command palette otomatis). Dark-coverage 116 CLEAN.
Benchmarks "Corpus Distributions": 12 baris distribusi (PUE/WUE/Capacity/Investment $B/Renewable × segmen finance/hyperscale/pm/research/spec) — n + p10/p50/p90 + bar interkuartil + marker proyek live via interpolasi persentil piecewise ("PUE 1.50 = ~p81 di hyperscale n=30"); chip "n kecil — indikatif" (n<5); nilai proyek dari calculateCapex live (bukan lookup statis).
Data Library DC Corpus: filter segmen dinamis + drill-down klik baris → 597 fakta mentah (nilai, company, tahun, link source_url, kutipan verbatim) via artifact baru corpus-facts.json.
Fixed
Narasi assessment Commissioning: readiness null tidak lagi dirender "0% Early Stage" → "Not Started — plan mode" jujur (kelas bug advisory phantom v1.98.1).
Catatan follow-up: manual/ pakai theme key rz_theme vs sitewide theme (pre-existing, tidak sinkron) — kandidat penyatuan.
v1.98.2PATCH
USER_MANUAL DCMOC dirombak total — mandat owner
Changed
dcmoc/USER_MANUAL.md 191 → 355 baris: daftar LENGKAP 49/49 engine (26 dikonsumsi DCMOC — kegunaan detail + fungsi kunci + halaman pemakai; 19 site/artikel via KB; 4 no-consumer dijelaskan jujur) diverifikasi programatik ≡ engine-catalog.json; tabel 20+ keluarga DATA bersumber; fitur platform per menu (trace ƒx, panel Kenapa?, prefill, Env Costs, Data Library, KB, peta, PDF, kalender, kejujuran presentasi); gate kualitas; changelog ringkas v1.89→v1.98.1. Deskripsi "untuk apa" bersumber DATA.sources katalog, bukan karangan.
v1.98.1PATCH
wave M/L audit tuntas: kontradiksi Architecture + kejujuran plan-mode + Asset MTBF + air ganda + LCC hidup
Fixed
Architecture 3 kontradiksi: profil default "ai-liquid" tak pernah ter-apply → chip drift amber "Profile recommends D2C Liquid — not applied · apply"; Power Topology bind sim.powerRedundancy (bukan tabel tier) — "2N — two fully independent active paths"; ASHRAE compliance kini baca density ceiling (air+60kW → 40% + "liquid/D2C required", rail Review + PDF ikut).
Asset MTBF/MTTR "—" semua baris: field mismatch mtbfHours vs engine mtbf — kini terisi IEEE-493 per class; "—" tinggal class yang memang tak ada di tabel (header jujur).
Dashboard LCC (15yr) "—": guard memanggil models.tco.totalCost yang tidak ada — fn engine sebenarnya lifecycleNPV → KPI hidup (TCO diskonto 15 thn + siklus refresh); node trace fin.lcc15 mirror (parity 117/117).
Readiness advisory phantom (readiness null dianggap 0) + Ops PM Compliance sub jujur; insight duplikat Requirements dedup (rule lokal vs flag engine).
Plan-mode tidak lagi "berprestasi": Construction Progress/AC/SPI/CPI + Financial health "A (baseline)" + Results dimensi Construction/Financial ber-chip "baseline" + footnote komposit "2 dimensi masih baseline plan-mode".
4 IRR diberi basis eksplisit + tooltip rekonsiliasi lintas halaman (Dashboard "unlevered after-tax · ramp · 15 thn"; Results "screening — unlevered 15y flat"; angka memang sah berbeda — kini dijelaskan, bukan disamakan).
Air Sustainability disatukan berlabel: KPI = volume engine pre-climate (≡ node trace), kartu Env menampilkan kedua baris (engine → ×climate ×1.2 → biaya) — tidak ada lagi dua angka tanpa penjelasan.
Site single-site tidak lagi "Recommended/Best" (komparatif butuh ≥2 site); basis note outage blended vs SAIDI mentah. Fire CAPEX "—" → tombol "run the CAPEX Engine →". Jahitan bahasa EN/ID dirapikan ("Penilaian: …").
16 halaman manual/*.html (technical manual per kalkulator, sudah lama ada tapi untracked) ikut tercommit.
EA-2 KORPUS BESAR 47 dokumen + audit visual 2 ronde: 4 bug kalkulasi HIGH fixed
Added — EA-2 Big-Project Corpus (owner: "crawl semua — PM, tech spec, tender, research, calculation")
47/48 dokumen publik baru ter-ingest (markitdown-only, biner dihapus, 60MB cap): spec 15 (OCP ORv3/immersion/manifold, ASHRAE TC9.9 liquid/water-cooled, NVIDIA DGX H100/GB200/GB300 RA, Green Grid PUE/WUE, Schneider WP110 AI) · pm/tender 10 (EIAR Google Dublin/Vantage/CyrusOne, Loudoun 5.33GW brief, ADB IFB, PJM/Dominion DC load, RFP colo) · finance 9 (10-K Equinix/DLR SEC EDGAR, deck Equinix/IronMountain/NTT ¥1.5T, Keppel DC REIT, Digital Core) · research 14 (LBNL 2024 176→580 TWh, IEA Energy&AI 415→945 TWh, arXiv AI-thirsty/POLCA/cooling-RL, EPRI, DOE, Uptime Survey 2024 PUE 1.56).
Korpus: gate 389 → 2.451 assert hijau — fakta PUE 69 · investment $B 134 · renewable 104 · WUE 12 · kapasitas MW lintas segment finance/hyperscale/pm/research/spec; Research Library 9 → 35 dokumen (KB) + Data Library DC Corpus persentil baru. Provenance (source_url + kutipan) wajib per fakta.
Fixed — audit visual 2 ronde (4 HIGH + 2 M + 3 kecil)
H1 CAPEX basis salah 1.0 MW/USA di sesi baru (menular Dashboard/Results/Financial): default capex store tidak pernah direkonsiliasi dgn simulation (2.5 MW/ID) sampai user mengedit requirement → $9.07M @1MW USA → $14.75M @2.5MW Indonesia (constructionIndex 0.65 kini aktif); rekonsiliasi on-rehydrate tanpa loop; $/kW dibulatkan.
H3 Pro-Forma Energy ~1000× off: pembagi /1000 salah unit ($/kWh diperlakukan $/MWh) di 3 lokasi FinancialDashboard → Energy $3K → $2.96M (NPV/IRR/sensitivity pro-forma kini sebasis).
H4 Commissioning racks 417 & "Tier IV": programRich tanpa rackDensity (fallback 6kW) → bucket ai_hpc (racks 34, konsisten fleet); label tier kini bind sim.tierLevel (Tier III), IST diatribusikan ke config 2N. Trace programRich ikut dibetulkan.
M: kolom HEADCOUNT proyeksi 5-thn kosong (field headcount vs totalHeadcount) · deep-link Monte Carlo tidak pindah tab (initialTab useState stale → effect sync) · breadcrumb FAQ.
Env costs benar KOMPUTASI per negara (SG $520K vs ID $116K vs US voluntary $185K, chip+basis+band flip). ƒx popover angka live + drill. NO-GO lever bisection nyata ("Revenue +22% ATAU CAPEX −29%"). Kalender per-class ×N tanpa baris #unit. Capacity chip forecast-aware. Site axis explain. Coverage 98.3% render.
1 FAIL environmental: CORS /fx dari origin localhost (worker allowlist — fallback snapshot jalan; bukan bug kode; produksi origin resmi tak terdampak).
Capacity banding forecast-aware — keputusan owner "go ahead"
Changed
Capacity Utilization status kini forecast-aware: chip OK/Watch/At-Risk dihitung dari puncak forecast pertumbuhan (share kapasitas design) bukan util saat ini yang struktural ≈1/(1+margin); tiap sistem dapat estimasi tahun exhaust ("At Risk ·~2029") — semua baris scale ∝ IT MW (power/cooling via facility, rack/space/network via racks), overlay di capacity-adapter.utilization() (field additive forecastPct/exhaustYear, fallback aman bila forecast kosong). Hover % = "Sekarang X% · puncak forecast Y% · exhaust ~tahun". Basis note diperbarui.
Results dimension explain (DL final): skor dimensi < 60 → chip Poor/Fair ⓘ klik → alasan formula live + lever solved atas fungsi dimensi yang SAMA dgn render (extract single-source dimension-explain.ts): capex bisection $/kW, sustainability upgrade cooling via pueMatrix ("air→inrow: skor 20→66"), financial revenue/capex bisection lewat rantai IRR engine, tier upgrade, requirements field-fill (+16.7/field) — + catatan jujur per dimensi (renewable TIDAK di formula PUE, hurdle 10% = konstanta scorecard). Grade < B → ringkasan 2 dimensi terlemah + apakah lever cukup mencapai 70 (dihitung).
Data Library "Country Coverage": matriks 40 negara × 30 field ✓/— dihitung live dari countries.ts — 98.3% (1.180/1.200); satu-satunya gap = HVO price kosong 20 negara (pasar belum ada); panel klik per negara daftar field kosong (actionable).
Fixed
Staffing "Cost per MW" benar-benar dibagi MW IT (sebelumnya ÷1 = total payroll); trace node diperbarui.
Capacity Utilization dapat basis-note jujur (util current ≈ 1/(1+margin) by construction; exhaustion ada di Forecast).
DL Site Intelligence: chip band Poor/Fair klik → panel alasan (kontributor terbesar weight×value live) + lever DIHITUNG via scoreSite nyata ter-bisection ("SAIDI ≤407 → axis 60 Good"; "cable landings 1→4 → 66"; unreachable → catatan jujur) + tombol Edit Criteria / deep-dive tab.
DL Capacity: baris utilization dapat chip OK/Watch/At-Risk klik → panel reason + lever solved atas rantai adapter yang sama (defer IT load → keluar band; margin honesty note) + tombol Phase Plan/Requirements di remediation rows.
Notes (kandidat pagi, terdokumentasi jujur)
Staffing "Cost per MW" ternyata dibagi 1 (bukan MW aktual) — node trace mendokumentasikan basis; fix menyusul.
HVO jadi opsi bahan bakar NYATA (FuelGen): baris perbandingan HVO vs diesel (harga negara × pajak, litres ×1.03 EN 15940, CO₂ −90%, delta $/thn) untuk 20 negara ber-harga HVO.
Changed (DM fase 3 — dead data dikonsumsi, scoring dinormalisasi; fallback verbatim semua)
Commissioning explainers (DE): chip L1-L5/FAT/SAT/IST bernama penuh + tooltip (label live engine + definisi ASHRAE Gl.0/BCxA berlabel); grup checklist derived dari equipScale (×N live; sistem 0 unit tidak dirender) — totals/readiness ikut config nyata.
CDU page rebuild (DE): 5 section engine-real — Sizing (models.cdu.size), tabel Hydraulics (Darcy/Haaland + Magnus dew point, bar + head), PUE mini-bars air vs liquid (pueMatrix), Refrigerant 9-baris GWP/COP + selected-summary (user vs auto), Deep-Sea advanced (intake @depth, seawater flow, pump, PUE ≤1.15) gated tick + teaser 1.6.
DA3 snapshot jujur: kartu ACTIVE Projects = live sim (chip hijau); kartu tersimpan "saved snapshot · tanggal"; badge amber "differs from current project" (Projects/Scenarios/Portfolio; portfolio +savedAt additive).
Monte Carlo revenue basis divergen: hardcode $120/kW/bln → DEFAULT_REVENUE_PER_KW_MONTH ($280 single-source) — MC sebelumnya menghitung distribusi di basis revenue berbeda dari semua permukaan lain.
2 tooltip escalation Pro-Forma salah copy-paste dikoreksi.
Changed
#333 Financial dedup: Lease Term input → derived dari Requirements contractDuration (+link ↗); Tax Rate sync reaktif negara; MC dapat kartu "Base Case" provenance (project chip + ↗); input analisis lokal diberi hint eksplisit.
EB trace wave-3: 19 node baru (staff.monthlyCost, constr.spi/cpi/ev/progress/forecast, cx.readiness/tests, results.score+4 dimensi, asset.fleet/health/replacement) + wraps di Staffing/Construction/Commissioning/Results/Assets; coverage gate permanen di walk probe (trace-coverage x/y per halaman, floor ≥30% halaman ter-instrument) — walk kini 24/0.
DN consumers: biaya kontrak SLA Maintenance dari DATA.omContracts (NBD≈on-call/4hr≈preventive/2hr≈comprehensive × IT kW; toggle third-party ×0.65 + aging ×1.5; fallback verbatim); Spares tab band harga; Data Library +3 dataset live (O&M Contracts, Spares Pricing, Env Costs sortable 40 negara).
Site Intelligence full trace (ƒx): 30 node site.* (5 engine terintegrasi: grid/disaster/tax/talent/compliance + leaf panel SAIDI/tarif/AQI/WRI/PGA/tax) — semua angka kartu Integrated Analyses + Detail Panels klik-to-trace.
Edit Criteria PREFILL: field kosong menampilkan nilai efektif — baseline negara (chip cyan, dari tabel COUNTRIES) atau screening typical (amber); store tetap unset (= semantik baseline). Preset Grid Voltage site +11/20/66 kV.
1.6 prefill DINAMIS (CB): lib/recommended.ts — 13 rule rekomendasi dihitung live dari parameter (substation by band MW, deep-sea dari poster spec engine, refrigerant lowest-GWP by cooling, fuel by tier, fee AACE band, solar 10% IT + BESS 2h); chip rec: X klik-apply + "Terapkan semua rekomendasi"; tanpa overwrite diam-diam. Field Fuel Autonomy ditambah di 1.6.
Decision explainability (DL, Phased Financial): verdict GO/NO-GO klik → alasan angka live + lever TERUKUR via bisection 36-iter pada model cashflow yang sama (Revenue +X% ke $Y/kW/bln, CAPEX −Z%, catatan hurdle); KPI merah klik → panel fase terburuk; PDF dapat section "Decision Rationale & Required Changes".
Staff Model (sim): slider AQI predefined dari environment.baselineAQI negara + turnover dari DATA.attritionFactors engine (chip baseline/override + reset); Monthly Hidden Loss klik "▸ kenapa?" → formula live + breakdown per-hire + lever terukur (reset baseline → hemat $X/bln, shift 12h, link lokasi).
Kalender maintenance dirombak: baris per SYSTEM ×count (222→19 baris; 500MW: 34.815→20), toggle Week/Month, blok ×N warna per tipe, hover panel kaya, klik → detail event di bawah grid.
Fixed (3 bug nyata)
IRR tax incentives ×100 dobel (site-intel — sumber "5780%" di screenshot owner): engine sudah mengembalikan persen; render & trace disinkronkan.
PI "0x" (Phased Financial): formula meng-nol-kan PI saat NPV ≤ 0 — kini (NPV+investasi)/investasi benar (mis. 0.65x), propagasi ke narasi + PDF.
Kalender maintenance blank di fleet besar: filter nama per-unit #idx tidak pernah match event batch (batch N — X units) sejak v1.91.4 — baris kosong ratusan; agregasi per-class baru menghilangkan akar masalah.
Changed
Dedup input (owner): Staff Model Config Region & Country → display derived "project" + "Edit di Requirements ↗" (satu sumber); Simulation Year diberi keterangan jujur (kontrol proyeksi analisis); dead code region dibuang. Substation dilabel "Dedicated 20–33kV (5-20MW)" (PLN 20kV, basis biaya sama) di 1.6 + Capex.
?v engine loader UNIFIED 2026-07-20-b (57 loader; sebagian stale 2026-07-16 ikut tersapu).
Value Trace instrument wave-2 (5 halaman engine): KPI ber-ƒx di Reliability (Tier Target → rel.tierTarget), Financial (Total Budget → capex.total), Operations (Availability Target + Energy Cost 24h → node baru ops.energyCostDaily), Sustainability (PUE Design → engine.pueMatrix, Carbon Annual → carbon.annualEmissions), Architecture (IT Load, Facility Load, Availability Target). Node ops.energyCostDaily ditambah ke graf value-trace.ts (acyclic, gate 73/0 tetap hijau).
RZExplain coverage sweep (CC): 42 explainKey di 4 file section Requirements (was 4, satu rusak mv-switchgear→voltage); 23 istilah baru di tools/explain-extra.json (budget-usd, design-margin, p50/p80, aace-class, newsvendor, spi/cpi/evm, saidi, pga, water-stress, deep-sea-depth, dll) → DB regen 802 entri (js/rz-explain-db.js), test-explain-db.mjs 10/10, layout dcmoc ?v=2026-07-19-cc.
Spares on-page assessment: panel Executive Assessment + prioritized actions (ReportNarrative spares family) di atas grid metrik Spares Optimization — fill-rate & classes-at-risk dinilai on-page, bukan hanya di PDF.
Changed
Staffing Cost Structure Decomposition dirapikan (owner: "chart aneh"): bar vertikal menggantung diganti 100% composition bar + baris waterfall horizontal kumulatif (offset Σ berjalan, nilai + persen + deskripsi per kategori) — geometri deterministik, tidak ada bar melayang/overflow.
TRACE VISIBLE everywhere it exists + final-audit fixes
Added
Trace "ƒx" badge (owner: "fitur trace kok belum ada" — it existed but was invisible): every traceable number now shows a small violet ƒx badge; graph +3 nodes (capacity committed MW, annual emissions chain, NPV screening chain); Capacity page KPIs (IT Load, Facility Load) now click-to-trace like Dashboard + CAPEX.
Fixed (final total audit, 2 agents)
localStorage migration: capacityPhasesCustomized gets a safe default on old persisted blobs (phase derivation can no longer break on reload).
New candidate sites + portfolio sites now seed from the LIVE project country + real capital coordinates (were hardcoded 'ID'/schematic).
Dead code from the tab deletion cleaned (DashTopBar props/ChevronDown, dashboard tab state); projects console.warn dev-guarded.
v1.94.9PATCH
on-page guidance — Site Intelligence; pattern complete
Added
Site Intelligence rail gains the on-page guidance panel (5th and final planned adoption): site band chip + narrative (leading risk factor named) + prioritized actions. The assessment pattern now covers Carbon, Reliability, Results, Commissioning and Site — every major grade/score in the app explains itself and says what to fine-tune.
v1.94.8PATCH
on-page guidance — Commissioning
Added
Commissioning Engine gains the on-page readiness guidance panel (4th adoption): readiness band chip + narrative (level composition, failed tests / open issues gating) + top-3 prioritized actions — live above the progress view.
v1.94.7PATCH
on-page guidance — Results
Added
Results Engine gains the on-page guidance panel (3rd adoption of the pattern after Carbon + Reliability): composite grade chip + narrative (why the band, strongest/weakest dimension) + top-3 prioritized actions naming the dimension to work first — live beside the score hero.
Strategic Planning explains itself and stops re-asking (owner: "bingung untuk apa; parameter tidak perlu diisi karena sudah ada"): a purpose banner states the three analyses (Feasibility fit / Expansion timing / Acquisition buy-vs-build) and the rule — inputs that already exist elsewhere are LOCKED here (land/grid/climate from Site Intelligence, target PUE from the engine, current footprint ≡ Requirements IT load, capex $/MW ≡ CAPEX engine) with edit-at-source hovers; only analysis-local parameters (growth %, horizon, comparables) stay editable.
v1.94.5PATCH
At-Risk remediation guidance — capacity equipment
Added
Every non-OK capacity status now says exactly WHAT to fine-tune (owner: "At Risk itu apa maksudnya — harus ada guidance per equipment"): computed from the same scaling divisors — e.g. "Utilisasi 122% > 85%: tambah +45 unit (250→295) untuk target ≤80%, ATAU turunkan beban 12.5 MW (phase plan/IT load), ATAU naikkan rating unit" — hover on the status chip AND remediation lines under the table (Watch rows get a plan-ahead variant). Pattern rolls out to reliability SPOF / spares / sustainability chips next.
v1.94.4PATCH
Cx level explainers
Added
Commissioning levels explain themselves (owner: "L1 itu apa kepanjangannya, misal FAT"): every readiness key renamed with its full meaning (L1 — Factory Witness, L2 — Site Inspection, L3 — Startup/Pre-Functional, L4 — Functional Performance, L5 — IST Integrated, SAT/FAT expanded) + hover explainer describing what each level covers (industry Cx Level 1-5 basis) on both the checklist accordion and the slider rows.
v1.94.3PATCH
site score explainers — every criterion self-explanatory
Added
Site Score & Compare rows explain themselves (owner: "score ini tidak menunjukkan apapun — get it done"): every criterion label carries a hover explainer (what it measures, the deriveFactors formula behind it, and an honest note when a low value just means site data hasn't been entered — country baseline in use) + each score cell gains a band chip ("91 · Excellent", "10 · Poor"); Natural Risks band inverts (lower = better). Explainer map lives in AXIS_EXPLAIN (site-intel types) — one source for table + radar.
v1.94.2PATCH
dashboard integrity — fake selector + fragment tabs deleted
Changed
The fake dashboard project dropdown is gone (owner-caught: "Indonesia DC Campus — 3MW" hardcoded label vs 2.5 MW context bar): the header now reads LIVE from the shared stores — real project name + "2.5 MW IT · design 3.0 MW" so the IT-vs-design distinction is explicit, one source with the context bar.
Engineering/Construction/Operations/Financial/Analytics/Reports dashboard tabs DELETED (owner: "isinya cuma potongan Executive Overview"): they were honest subsets; the full Executive Overview is now the single dashboard view (engine deep-dives live in their own sidebar pages).
v1.94.1PATCH
map integrity — real styles, no fakes, clean fallback
Changed
Map style tabs are now REAL (owner: "3D terrain dll cari repos, kalau tidak ada delete"): Map = OpenFreeMap vector · Satellite = Esri World Imagery raster (free tiles, mandatory Esri attribution rendered); fake Hybrid + 3D Terrain buttons DELETED (3D needs a DEM terrain source — backlogged, not faked).
"Wave aneh" fixed: the schematic SVG fallback now renders ONLY when the real map fails to load (onStatus callback) — no more decorative coastline under a working map.
Context-bar hover cleaned: the raw black title box is gone; the data-vintage note lives on the small "data 2026-Q1 ⓘ" chip only.
v1.94.0MINOR
Pro-Forma predefined visible + lease-term dedup
Changed
Financial & Revenue parameters announce their predefined basis (owner: "dibuat predefined, saya tidak perlu isi, tapi bisa override"): a violet banner states every parameter auto-derives from data (country tariff/tax/inflation · tier multipliers · CAPEX-linked NRC) with manual edits preserved (the auto-derive + userEdited guard already existed — now visible). Lease term dedup: contract years now follows Requirements 1.1 Contract Duration (one source) instead of a hardcoded 10.
v1.93.9PATCH
EA6b — Research Library in the Knowledge Base
Added
Knowledge Base "Engine Models" tab gains a Research Library card (owner: "journal/research masukkan ke knowledge base"): the corpus document index rendered from auto-generated research-library.json — each source document (Google Environmental Reports, operator sustainability hubs, Uptime/IEA pages) with a direct link, segment, fact count and metric coverage. The aggregate pipeline emits this index; corpus refresh updates the card automatically.
v1.93.8PATCH
Benchmarks — project position vs the live corpus
Added
Benchmarks page opens with "Posisi Proyek vs Korpus Publik Multi-Sumber" (owner: "jangan cuma JLL/CBRE; benchmarks kurang detail"): per metric (PUE, site capacity, renewable share) a distribution bar shows the corpus p10-p90 band (interquartile shaded, median tick) with YOUR project's marker + percentile chip ("proyekmu: 1.22 (p25-50)") and full provenance line (n facts · document count · companies). Data = live DATA.benchmarksCorpus; corpus refresh updates the page automatically.
v1.93.7PATCH
EA6 — DC Corpus surfaces in the Data Library
Added
Data Library gains a "DC Corpus" tab (first tab): the multi-source public-data distributions rendered live from DATA.benchmarksCorpus — per metric × segment rows with n, p10/p25/p50/p75/p90, company list and document count; auto-generated chip + provenance note (every underlying fact carries source_url + verbatim quote). Regenerating the corpus pipeline updates this tab with zero manual edits (auto-link chain).
v1.93.6PATCH
corpus round 2 — 92 facts, 5 metric families
Changed
Corpus extractor round-2 patterns (phrasings actually present in hyperscaler reports): water (million gallons), investment ($B), energy (GWh), WUE loose-form — corpus now 92 facts across pue/capacity/renewable/water/investment; DATA.benchmarksCorpus regenerated (5 metric families with p10-p90 distributions), full auto-link chain green.
v1.93.5PATCH
Financial input dedup — batch 1
Changed
Tax Rate is no longer a manual input on Financial (owner: "input duplikat di menu lain hapus"): it renders as a derived read-only value from the project country's economy.taxRate (emerald "country" chip, single source — change the Country in Requirements and it follows). Revenue/escalation stay editable but are already country-predefined (auto-derive with manual-edit guard). More dedup (project life ↔ contract duration, Pro-Forma fields) continues per task #332/#333.
v1.93.4PATCH
tooltip coverage — Field primitive cascade
Added
Every Requirements form label is now hoverable (CC toward 100%): the shared Field primitive gains built-in hover — explainKey renders the RZExplain ⓘ panel (779-entry DB), otherwise the hint doubles as the hover title, otherwise a generated fallback — ONE primitive change cascades to all ~67 Field usages; cursor-help affordance on labels. First explain keys wired (tier, SLA, rack density, contingency/margin, grid voltage).
v1.93.3PATCH
predefined values batch 1 + cascading country select
Added
Predefined values, editable (owner mandate, first 4 fields): Peak IT Load auto = IT Load ("menyamakan"), Avg IT Load auto = 75% of peak (AI-cluster utilization band, screening), SLA Target auto = tier availability target (Uptime engine), Budget auto ≈ CAPEX P80 (P50 × AACE band factor, "budget-commitment basis") — each auto-fills ONLY when empty, violet "predefined" chip shows the basis, any manual edit is preserved.
Cascading Region → Country dropdown (owner: "40 negara kepanjangan"): CountrySelect now picks the REGION first (EMEA/APAC/AMER/MENA/AFR/LATAM), the country field lists only that region's markets; region auto-syncs on external changes (project restore). One shared component — Requirements, Site editor, Settings inherit.
v1.93.2PATCH
REAL MAP — MapLibre GL + OpenFreeMap
Added
Site Location & Overview now shows a REAL geographic map (owner: "harus map tampilan beneran, jangan kolot"): MapLibre GL + OpenFreeMap vector tiles (no API key, production-allowed, researched 2026 stack) — dynamically imported so it loads only on the Site tab. Candidate-site pins at REAL WGS84 coordinates (legacy schematic 0-1 values auto-migrate to the country/city anchor), click-to-select, popups, fit-bounds, nav controls. The old schematic SVG demotes to an honest offline fallback below. Headless-verified: maplibre canvas + markers render, 0 errors.
v1.93.1PATCH
DC-hub city dropdown
Added
City / Region is no longer free-typing (owner: "kok isi manual"): new DC_CITIES table — 2-4 real DC-hub cities WITH coordinates for all 40 countries (Ashburn/Dallas/Phoenix/SV, Jakarta/Bekasi-Cikarang/Batam, Madrid/Barcelona/Zaragoza, Muscat/Salalah/Duqm, Hamina, Johor/Cyberjaya, NEOM, Luleå…). Requirements 1.1 City field becomes a dropdown-first combobox (datalist — custom entry still allowed) filtered by the selected country. City coordinates feed the upcoming real-map pin placement.
v1.93.0MINOR
COUNTRY EXPANSION — 32 → 40 markets
Added
8 new country profiles (owner: "tidak ada Oman, Finland, Madrid dll yang penting"): Oman (Duqm/Salalah zones, $0.07/kWh), Finland (Hamina, grid 0.08 kgCO2), Spain (Madrid/Barcelona/Aragón), Canada (Toronto/Montreal, Québec hydro note), Italy (Milan), Norway (hydro 0.03), Denmark (Copenhagen/Odense), Switzerland (Zurich, highest labor rates) — full profiles (economy/labor/compliance/environment/risk/tax-incentives/disaster/grid/talent/fuel) in the authoring source, screening-labeled per value; SITE_AUGMENT rows (Aqueduct/ASHRAE/SAIDI/PGA); COUNTRY_GEO anchors (DC-hub cities). DATA.currency +OMR/CAD/NOK/DKK/CHF (gateway /fx-verified). Engine regenerated: parity gate now 155/0 (was 126), engine gate count pin 32→40. All country dropdowns, site engines, fuel/tax/talent/grid analyses cover the new markets automatically.
v1.92.6PATCH
on-page guidance — Reliability
Added
Reliability Engine gains the on-page guidance panel (pattern from Carbon v1.91.6): availability status chip + plain narrative (why the band, β common-cause note) + top-3 prioritized actions naming exactly what to fix (redundancy on the weakest chain, SPOF elimination, MTTR levers) — rendered live beside the KPI row.
v1.92.5PATCH
trace graph batch 3
Added
Trace graph extended: staff.fte (sub-linear headcount chain, formula shows the MW^0.65 calibration), rel.tierTarget (engine tier availability, glossary-linked), opex.totalAnnual (engine dcContract basis, deps on IT load + FTE). CAPEX Engine page KPI grid is now trace-aware — Total CAPEX (P50) is click-to-trace; the render pattern (trace key on KPI rows) rolls out to every KPI grid next.
v1.92.4PATCH
trace cross-surface links + trace standard
Added
Trace links beyond DCMOC (owner: "trace bisa beda page atau data lain, tidak terbatas DCMOC"): trace leaves gain external cross-surface links — "🌐 Sumber eksternal" beside "↗ Edit di menu" (PUE → site glossary; tariff → DC market tracker; corpus facts → the original public document URL as the graph grows). Audit chain now runs DCMOC number → parameter → engine constant → public source document.
standarization/VALUE_TRACE_STANDARD.md — the trace system codified (registry schema, traceId ≡ data-bind ≡ value-bindings id, provenance colors, adoption RULE: every newly rendered number MUST register a traceId — added to the Definition of Done in ENGINE_UNIFICATION.md; USER_MANUAL §9).
v1.92.3PATCH
Trace Angka v2 — intuitive visual formula
Changed
Click-to-trace popover redesigned for clarity (owner: "sangat bagus, intuitive, mudah dimengerti"): big live value header + provenance chip in plain language ("Input kamu" / "Engine (bersumber)" / "Dihitung") · formula rendered as VISUAL PILLS — each operand is a clickable provenance-colored card showing its label + live value, with large operators between, ending in "= result" · clicking a pill drills into ITS formula with a breadcrumb path back · leaves show a plain-language source card ("Ini titik ujung: angka yang KAMU isi") + an "Edit di menu" button. Headless-verified: "[IT Load 2,500 kW] × [Design PUE 1.5] ÷ 1000 = 3.75".
v1.92.2PATCH
live FX + shared currency list
Added
Live FX rates via the existing gateway /fx (owner: "auto update pakai gate yang sudah ada"): src/lib/fx.ts — 1h-TTL cached live rates (verified live: GBP 0.744, CNY 6.78, IDR 17,945 /USD), offline fallback = engine DATA.currency snapshot with an honest source flag. Project-context bar shows the project currency's live rate chip ("IDR 17,945 /USD · live") when the project currency ≠ USD; fmtInCurrency ready for display-conversion adoption.
Shared CURRENCY_LIST — Settings default-currency dropdown gains GBP/CNY/AUD/INR (one list with Requirements 1.1, drift impossible).
v1.92.1PATCH
VALUE TRACE INDEX — Excel-style formula field, batch 1
Added
Click-to-trace "Formula Field" (owner: "seperti Excel — klik X tahu rumus A+3=X, trace sampai titik paling ujung"): new live trace graph (src/lib/value-trace.ts — deps + formula templates + live getters, acyclic) + TraceValue popover component: click an instrumented number → its formula with LIVE numbers substituted ("3.75 = IT Load [2500] × Design PUE [1.5] ÷ 1000"), every source row expandable (recursive drill-down to leaf inputs/DATA constants with provenance chips + DATA.sources ref) or jumpable (↗ opens the value's home menu). Batch 1 instruments the Dashboard IT Load + Facility Load tiles; the graph seeds from the value-bindings catalog and grows per page. Headless-verified (popover renders live formula + expand controls, 0 errors).
v1.92.0MINOR
DC DATA CORPUS — multi-source benchmarks land in the engine
Added
DATA.benchmarksCorpus (owner mandate: "jangan cuma JLL/CBRE — crawl banyak DC operator/developer"): the engine now carries REAL multi-source distributions (n, p10-p90, company list, source count) per metric x segment, GENERATED from the new public-data corpus pipeline tools/dc-corpus/ — sources.yaml (20 curated public sources: Digital Realty, NTT, Vantage, STACK, AirTrunk, Princeton Digital, CoreWeave, Oracle, Microsoft, Google, Meta, AWS, Uptime, IEA + Google Environmental Report 2024/2025 PDFs) → fetch (rate-limited curl + markitdown, PDF-aware) → extract (every fact REQUIRES source_url + verbatim quote — rejected otherwise) → aggregate (percentiles → engine between @@CORPUS markers, DATA.sources entry, auto-link chain). Seed corpus: 82 facts incl. 30 per-site fleet PUE values from Google's published site table. Append-only growth; refresh = rerun pipeline.
v1.91.6PATCH
on-page grade guidance — Carbon/ESG first
Added
Carbon/ESG Rating now explains itself on-page (owner: "Rating F itu kenapa, apa yang di-finetune"): a guidance panel under the header renders the SAME deterministic assessment the PDF uses — profile chip + narrative (why this band, which inputs drive it) + prioritized HIGH/MED/LOW actions naming the parameter to adjust (PUE via cooling class, renewable share, etc.). Pattern (buildAssessment/buildActions live on-page) rolls out to every grade/rating chip app-wide in the next slices.
v1.91.5PATCH
scale-aware maintenance fleet ratings
Fixed
Absurd fleet counts at hyperscale (owner-caught: x15,625 CRAC 100 kW + x600 gensets on 500 MW): unit ratings now step up with facility scale the way real designs do — air handling 100 kW CRAC (≤5 MW) → 300 kW CRAH (≤50 MW) → 900 kW fan-wall (>50 MW); chillers 1 MW → 2.5 MW; gensets 2.5 → 3.0 MW above 100 MW. 500 MW now shows ~667 air units / ~500 gensets (engineering-plausible campus counts). Screening tiers, labeled in-code; schedule/SLA/spares consumers inherit automatically.
Maintenance Schedule tab hung ~3 minutes at 500 MW (owner-caught): the generator enumerated EVERY unit (x600 gensets, x15625 CRAC → ~300K+ event objects). Units now batch per class (≤26 stagger slots — beyond that duplicate slots add no scheduling information), each event carries units and hours multiply by batch size — ~130x fewer objects, totals identical, every task × frequency × stagger week still present. Batch labels show "(batch N — X units)".
v1.91.3PATCH
calc hygiene — Financial hardcodes bound to canonical sources
Fixed
Financial energy cost hardcoded PUE 1.4 + $0.10/kWh in 3 places (audit finding): now reads the live pueMatrix[cooling][tier] + the project country's economy.electricityRate — changing country or cooling moves the Financial opex/budget lines automatically (owner mandate: tarif listrik auto per negara). Maintenance $50/kW/yr literals bound to the shared sourced screening constant. Revenue screening bases (Disaster/Tax) unified to lib/screening.ts in the prior commit.
v1.91.2PATCH
sub-linear staffing algorithm
Fixed
Auto headcount scaled LINEARLY per MW (owner-caught: 834 FTE at 500 MW vs the Uptime benchmark 204 shown on the same page). calculateAutoHeadcount now applies economies of scale: effective MW = 10 x (MW/10)^0.65 above the 10 MW reference anchor — 10 MW reproduces the reference model exactly; 500 MW Tier-4 lands at ~203 FTE (benchmark 204). Per-role formula strings show the scale curve. Consumers (Staffing, Portfolio, Capacity engines) inherit automatically.
v1.91.1PATCH
chart tooltips dark-theme sweep
Fixed
21 chart hover tooltips rendered as WHITE boxes with invisible text in dark mode (owner-caught on the CAPEX forecast curve): every bare contentStyle={{ fontSize: 10 }} recharts Tooltip across 13 files (capex, capacity, construction, commissioning, financial, operations, reliability, results, sustainability, assets, architecture rail, site radar, scenarios) now carries the dark panel style (#1e293b bg, #334155 border, explicit light text) — hover values readable everywhere.
Capex + simulation stores now PERSIST (audit-critical): reload no longer wipes IT load/tier/cooling/phases/deep-sea tick while requirements survived — the root of the owner-caught cross-page contradictions ("Requirements 1.6 enabled, CDU says off"; "Indonesia · 1,000 kW" headers on a UAE 500 MW project). selectedCountry rehydrates by id against the live COUNTRIES table.
Candidate site follows the project country: syncToCountry updates the pristine scenario site (name/country/coords from the new COUNTRY_GEO table) on every writeSharedCountry + stale-localStorage rebase on load; user-customized sites keep manual values. Kills "Indonesia Site" on a UAE project.
Capacity phases derive from IT load: hardcoded 2/5/10 MW seeds replaced by derivePhases(itLoad) (4-phase staggered split); pristine phases re-derive on itLoad change, manual edits set capacityPhasesCustomized and stop derivation. Kills "22 MW plan on a 500 MW project".
Engine-ready signal: useEngineReady() hook + rz-engine-ready event fix the deferred-engine race (Spares memo never re-ran; CDU deep-sea stuck "off"); CDU banner now distinguishes "engine loading" from "off".
Changed
Data-vintage banner → live project-context bar (project name · country · MW IT · tier · cooling from the shared stores; vintage note moved to hover) — every page shows which project it is working on.
maplibre-gl dependency added (real-map slice upcoming); src/constants/geo.ts country geo anchors.
v1.90.2PATCH
article-calculator engine sweep, batch 4
Added
models.mttr + DATA.mttrResponse (promoted from article-4): vendor-vs-inhouse MTTR phase model (per-category base durations, skill 1.5..0.55, coverage mobilization, spares gap factor) + full annual economics compare (downtime delta x effective cost/hr, callout + 55% retainer recovery, ROI, break-even); 7 engine asserts. Article-4 phases + savings block delegate.
Engine catalog AUTO-regenerated by the ship gate: 34→37 namespaces, 160→173 functions, 85→88 sourced tables — Knowledge Base + FAQ reflect the new models with zero manual doc edits (the auto-linking chain doing its job on its first real engine change).
Owner Action Board: backend-deploy-v163 marked DONE + VERIFIED (owner ran wrangler deploy, Version 06e8006; /calc POST live with engineVersion 2.5.1, NPV test ≡ engine).
v1.90.0MINOR
DCMOC revision round 3: PDF executive-assessment engine · architecture visual depth · auto-linking system
Added
PDF Executive Assessment engine (dcmoc/src/modules/reporting/pdf/ReportNarrative.ts): every DC-OS export now ends with the min-standard algorithmic conclusion — a deterministic profile chip chosen by per-family threshold rubrics (16 families: capex $/kW bands, financial NPV/IRR-vs-hurdle, reliability nines-vs-tier-target, sustainability PUE bands, capacity stranded %, construction SPI/CPI, operations PM-compliance, commissioning readiness, results grade, site score, assets health, spares fill, staffing, requirements completeness, architecture layer validation, cooling) + a narrative built from the page's live numbers + prioritized HIGH/MED/LOW action items. Wired into ALL 16 export call sites (11 verified headless by the new tools/_dcmoc_export_probe.mjs — 44/0: Executive Assessment + actions + summary band present in every captured print document, 0 page errors).
AUTO-LINKING system (owner: "harus auto, tidak perlu Claude"): tools/build-engine-catalog.mjs generates dcmoc/src/lib/engine-catalog.json from rz-engine.js — 34 model namespaces · 160 functions with parameter lists · 85 sourced DATA tables · consumers GREP-DERIVED from real usage (site articles + DC-OS modules). Knowledge Base "Engine Models" tab + FAQ "Engine & Data Reference" section render the catalog LIVE (auto-generated, never hand-written). New SHIP GATE tools/test-value-bindings.mjs (73/0): value-bindings coherence (unique ids, engine-provenance entries carry engineFn, every engineFn resolves against the live engine, pages ⊆ real tab ids — caught 3 stale tab aliases + 1 missing engineFn on first run) + catalog STALENESS (regenerates in-memory and diffs vs committed — an engine change cannot ship with stale docs). Chain codified in ENGINE_UNIFICATION.md + CLAUDE.md + USER_MANUAL §8.
Requirements 1.1 "Data Center Image (Dashboard hero)" upload card restored where the owner expects it: same store as the CAPEX card (useCapexStore.heroImage, WebP-compressed client-side) — custom/default chip, preview ring, "Reset to default"; the Executive Dashboard falls back to the default DC-OS render when nothing is uploaded.
Changed
Architecture diagram visual depth (reference-parity round): transformer glyph now draws primary/secondary winding coil pairs; genset = engine block + shaft + alternator; UPS carries an integrated battery-cell stack; chiller = evap/cond barrels + compressor; rack = dense U-unit grid; fuel = saddle tank + level line; cooling tower gains fan + fill louvers; pylon + switchgear + PDU + CRAH/CDU detail passes. Stacked-unit CASCADE rendering with real ×N count badges (layout units field), group containment headers rendered as count-badged chips ("COOLING PLANT — 8 chillers"), A/B bus bars emphasized (3px, junction dots, tap points).
Owner Action Board (setup-supabase.html): github-token-rotate updated — rotation reported DONE 2026-07-19; remaining owner-typed step documented (store the new PAT in ~/.git-credentials, verify git push --dry-run).
v1.89.1PATCH
article-calculator engine sweep, batch 2 — includes a REAL unit-bug fix
Added
models.water.aiQueryFootprint + DATA.aiWater (promoted from article-20 "AI Water Footprint" + "AI Vs Human" tabs): per-query water model — 19 AI models (mL/query, sourced), complexity/cooling/region multipliers, upstream factor (×3 grid-power water), scale multipliers (personal → global 1e10), bottle/shower/drinking-day equivalences + CO₂ + cost; 12 engine asserts. Both article-20 tabs now delegate to the engine (inline fallback kept), headless-verified page ≡ engine.
models.aiFactory.gpuBuild + DATA.aiFactory.gpuBuild (promoted from article-23 Colossus calculator): GPU capex, annual power at PUE, build-speed vs the 122-day Colossus benchmark, $8M/MW infra screening, 5-yr TCO; 4 engine asserts. Article-23 delegates (verified: defaults $4.59B TCO ≡ engine; reactive input change $7.59B ≡ engine; 0 console errors).
Fixed
article-18 annualEnergy ×1000 unit bug (REAL "angka ngawur" caught by the sweep): the article computed facPowerMW × 8760 × elecRate × 1,000,000 — MWh→kWh conversion needs ×1e3, not ×1e6, so the annual-energy cost line was overstated 1000× (a 1 MW facility at $0.08/kWh showed ~$700.8M instead of ~$700.8K). Fixed in the engine promotion AND the article's inline fallback; correction documented in DATA.sources.aiFactory; accuracy assert added proving the sane value. Headless-verified: page now shows $1.1M-scale energy ≡ engine.
Changed
Engine min rebuilt (terser); ?v=2026-07-19-a23 bumped sitewide (59 files + engine pdf scriptTagsHTML). Sweep tracker updated in ENGINE_UNIFICATION.md — 5 calculators engine-bound, 18 article pages remaining.
v1.89.0MINOR
article-calculator engine sweep, batch 1
Added
models.gridImpact.residentialBillImpact + DATA.gridImpact (promoted from article-11): the SEA citizen-bill screening model — DC annual GWh at 90% CF, household equivalence, tariff pass-through (40% screening) × IEA 15%/yr growth, monthly/annual bill impact in local + USD, grid-load share — 6 SEA countries with sourced tariffs/grid data; 5 engine asserts.
models.water.facilityFootprint + DATA.waterFootprint (promoted from article-20): full facility water screening — WUE base per cooling class × climate multiplier + upstream-power water factor (1.5 L/kWh non-renewable), household/city/pool equivalences, water cost per source type, AI share, hyperscaler benchmarks (Google/Microsoft/Meta/AWS per-MW) — 6 engine asserts.
Changed
article-11 & article-20 calculators now ENGINE-BOUND: page computations delegate to the shared models (inline math retained as no-engine fallback); headless-verified page ≡ engine on both. ?v=2026-07-19-a11 bumped sitewide. Sweep progress: 2 of 22 pages; tracker in ENGINE_UNIFICATION.md.
v1.88.3PATCH
Owner Action Board pindah ke setup-supabase.html
Changed
Owner Action Board dipindah dari rz-ops ke setup-supabase.html (halaman "Yang harus kamu lakukan" — rumah aslinya, per owner): 8 item hidup (rotasi GitHub token, deploy backend v1.63.0 + tes, SQL hardening Supabase, Edge Function admin-users + Migrate, reset password terekspos, naikkan spend limit Claude, musik intro CC0, taste-skill Dunia) dengan alasan owner-only + langkah + tick persist (rz_owner_actions_v1) + badge terbuka; item merujuk step detail ber-perintah di halaman yang sama. Nav rz-ops kini link eksternal ke halaman ini.
v1.88.2PATCH
accuracy program M3-full: truth coverage across EVERY model family
Added
Accuracy gate expanded 27 → 40 green asserts covering every remaining model family with formula-grade independent truth: fire NFPA-2001 clean-agent mass (m = V/s·C/(100−C)) + inert flooding (V·ln(100/(100−C))), water WUE·kWh, partial-load-PUE exact identity, deep-sea ACCURATE-mode energy balance (TEOS-10 ρ=1025/cp=3985), commissioning readinessIndex weighted-sum identity, newsvendor Q* end-to-end replication (erf-Newton Φ⁻¹ ≡ engine Acklam at ceil grain), compound-growth FV, plus recomputed-from-DATA identities: staffing round(hc·salary·load), carbon scopes Σ s1+s2+s3 ≡ total, construction schedule bounded by Σ phases, tier classify bounds, LCOE ≡ CRF reconstruction ($48.82/MWh exact match). Engine formulas certified — zero code changes needed this batch (all families already exact).
v1.88.1PATCH
rz-ops: Owner Action Board
Added
Owner Actions section in the rz-ops admin console: the standing list of actions only the OWNER can perform (Supabase SQL hardening, admin-users Edge Function deploy + migrate, exposed-password resets, Claude spend-limit raise, CC0 intro-music file, Dunia taste-skill install) — each with why-owner-only, exact steps and an added-date; per-item done-ticks persist locally and an open-count badge sits on the nav. The list is maintained: any future task that ends blocked-on-owner lands here.
v1.88.0MINOR
ENGINE ACCURACY PROGRAM M1-M3: independent truth harness + precision core
Added
Independent accuracy truth harness (permanent gates): tools/verify-engine-accuracy.py computes 27 worked examples at 50-digit Decimal / machine-erf precision from the AUTHORITATIVE formulas (never engine output) — annuity/CRF, NPV/IRR bisection, Weibull, exact binomial k-of-n, availability chains, Φ⁻¹, exact Poisson sums, Magnus dew point, Colebrook iterated to 1e-14, deep-sea poster identities, opex accounting identity — into tools/fixtures/accuracy-truth.json; tools/test-engine-accuracy.mjs asserts every engine value within its documented per-family tolerance. Baseline run MEASURED the defect list before any fix.
Deep-sea poster-floor spec (owner baseline): DATA.deepSeaCooling.spec + deepSea().spec now carry every poster field — intake 800-1,000 m @ 4-6 °C, return 9-11 °C, three loops (TCS/FWS/seawater), Ti Grade-2 PHE (approach 1.5-2.5 °C @ 10 bar), 4-stage filtration (50 mm → 5 mm → 200 µm → 50 µm backwash), full materials + redundancy map, facility 20-21/28-32 °C, trim-chiller basis. Poster identities gate-locked (8.625 m³/s exact, 4+1 pumps).
Changed (measured error, before → after)
Reliability chain de-saturated (MATERIAL): availability/parallel/series/kOutOfN/downtime no longer round 6 dp INSIDE the chain — two-path parallel of a 0.999968 item was collapsing to exactly 1.000000 (downtime 0.0 min/yr, the engine-level origin of the fake "100.0000%"); now full double precision end-to-end (err 1.0e-9 → < 1e-15; downtime 0.0 → 0.00243 min/yr on the reference chain).
Inverse normal CDF upgraded: Beasley-Springer-Moro (1977, |ε|<4.5e-4) → Acklam (2003, |ε|<1.15e-9) — spares newsvendor quantiles now precision-grade; kernels exposed as models.spares.normInv/normCdf/poissonCdf (one implementation, gate-verified at 2e-9).
IRR solver scale-aware: Newton/bisection tolerances now scale with cashflow magnitude (1e-12·max|cf|) — $M-scale series converge to machine precision (measured err 5.6e-17) instead of a fixed $1e-7 cutoff.
Provenance: sources updated (spares.acklam, pue.partialLoad screening basis); ?v=2026-07-19-accuracy bumped across all 58 engine-loading pages + DCMOC.
Verified unchanged (parity locks held)
capex golden EXACT (<$1), commissioning cx golden exact, opex default ≡ dcContract bit-identical, reference-parity 126/0, engine suite 426/0, datahall 57/57, conv DoD, DCMOC walk 23/23 + synergy 6/6. Whole-dollar opex line rounding kept as the documented ACCOUNTING convention (components+overhead ≡ total identity now gate-asserted; max error ≤$3 on multi-$M totals).
Value Binding & Sync catalog (src/lib/value-bindings.ts): machine-readable documentation of every bound number in DC-OS — 38 rich entries across 7 groups (shared canonicals, CAPEX chain, PUE/OPEX, reliability & Cx, site & sibling engines, staff/assets/spares, financial & results). Each entry: source parameters → exact formula/engine function → every consumer page, with provenance (input/engine/derived/tracking/screening) and trap notes (pueMatrix-direct, EVM single-source, margin→contingency).
Knowledge Base rebuilt as the manual: searchable/filterable "Value Bindings" tab — expandable per-value detail with source-param chips, engine fn, consumer list and REAL page links; engine-models tab retained. FAQ gains a "Where does every number come from?" section linking to it.
Cross-page synergy probe (tools/_dcmoc_synergy_probe.mjs, permanent): drives the app and ASSERTS the same bound number renders identically everywhere — CAPEX total identical on 4 pages (Dashboard/CAPEX P50/Financial baseline/Investment), design PUE (Dashboard = Architecture), Ops partial-load PUE ≥ design (documented separate basis), IT load, dashboard design-capacity ≥ IT (margin basis), tier labels. 6/6 green.
Changed
Strategic Planning "Estimated CAPEX" now states its basis explicitly: CAPEX-module rate × the land/grid-CONSTRAINED MW — not the project MW (the audit's one confusable pair, now labeled + documented in the manual).
v1.86.2PATCH
DCMOC audit round 2, batch AG: visual-audit fixes
Fixed
Data Library "Latest As-Of" KPI showed prose text ("const") — now filters to date-like source entries.
Spares "Recommended Stock Value $0" now explains itself honestly when every class' newsvendor Q* is zero at a small fleet (ROP covers the low failure demand).
Audit Trail was an explainer-only page — now shows the REAL local change history (settings change log + scenario saves + project updates, newest first) with KPI chips; server-side audit_log note retained.
Fire Suppression enriched: suppression-zones section (engine equipment scaling), volume per zone, fire-system CAPEX from the capex engine (fire + detection keys, same source as CAPEX page), shared fire/alarm type basis.
Asset Lifecycle depreciation-by-year strip capped with internal scroll (was dominating the page).
Audit verification: 43-tab full-page screenshot walk reviewed one-by-one; Grid/Talent "empty chart" appearances confirmed false positives (recharts entrance animation vs screenshot timing — bars verified present at runtime).
v1.86.1PATCH
DCMOC batch 7: universal hover sweep
Changed
Every KPI card across the 10 engine pages is now hoverable (Phase X): native title with the exact label/value/basis on Architecture, Capacity, CAPEX, Construction, Commissioning, Site Intelligence, Operations, Financial, Sustainability and Assets — on top of the hover titles already carried by the new diagram symbols, div-bars, comparison cells and radar/donut recharts Tooltips shipped in batches 1-6.
v1.86.0MINOR
DCMOC batch 6: Platform group rebuilt to the reference UIUX
Added
Scenarios page (was only a side panel): platform header, KPI chips (total/in-comparison/countries/baseline), search filter, rich table with per-scenario CAPEX/OPEX DELTA columns vs baseline (oldest save), ★ comparison selection, detail rail (full saved config + restore/compare quick actions), pagination footer.
Template Library to reference: platform header + KPI chips, engine-profile cards with what-it-sets checklists; "Use This Template" now runs the FULL shared profile writer (density + cooling + tier floor + workload-mix preset — same path as Requirements 1.2).
Data Library + Projects: platform header + KPI chip rows (datasets/provenance/DATA version/as-of; projects/active/engines-ready/scenarios/countries) over the existing honest data.
Fixed
Settings/Integrations deep link now switches tabs correctly (React instance reuse keyed).
v1.85.4PATCH
DCMOC batch 5: Settings & Integrations to the reference UIUX
Changed
Settings rebuilt to the reference layout (setting.png): Overview tab with KPI chip row (projects/scenarios/integrations/engine-data/storage — all real counts), Quick Settings rows (organization, default country w/ Apply, currency, theme — icon + subtitle + inline control), System Preferences card grid (Data Management, Integrations, User Management, Audit Trail, real JSON Backup & Export, Security note), right rail with Platform Information (site version, engine DATA version, real storage estimate bar), Recent Activity (real settings change log, additive store field) and Need Help (FAQ/Knowledge links).
Integrations rebuilt to the reference layout (integration.png): KPI row (total/reachable/configured/errors), search filter, integrations TABLE (name, kind, status chip, last test, test/delete row actions), selected-integration DETAIL RAIL (name/kind/URL/secret-ref editors + Test Connection), Add Custom Integration; the sidebar Integrations tab now opens this same surface (old card grid retired). Honest scope labels retained (CORS-limited reachability, secrets never stored).
v1.85.3PATCH
DCMOC batch 4: Architecture dynamic symbol palette + all-parameter diagram
Added
Dynamic symbol palette (diagram/palette.tsx): data-driven registry of 16 parametric equipment glyphs (utility pylon, IEC transformer, switchgear/breaker, genset, fuel tank, UPS AC/DC, battery plates, PDU, rack, spine-leaf fabric, chiller, cooling tower, CDU pump+HX, CRAH, BMS, IT load) with logical + SLD skins — the layout engine composes ONLY from the palette; a new equipment kind is one registry entry.
Diagram plots from ALL requirement parameters: transformer stage (MV/LV bank counts), A/B bus bars (+ dashed spare trunk at 2N+1), battery-autonomy annotation per tier, utility blocks named from the 1.1 utility provider, IT hall split into CELLS sized by the workload mix (AI/GPU, storage, general+network with per-cell rack counts), growth-phase boxes from the capacity plan (NOW/PLAN badges), design-margin + SLA + use-case annotations, group containment boxes (POWER TRAIN / IT HALL / GENERATION / COOLING PLANT / NETWORK & CONTROL) with live counts. Every block carries an exact-value hover title. Param-reactivity probe: changing grid voltage (incl. new 20 kV) re-labels the diagram live — 7/7.
v1.85.2PATCH
DCMOC batch 3: Monte Carlo folded in + Strategic dedup + CDU rebuild with deep-sea advanced
Changed
Monte Carlo folded into Financial: sidebar sub-menu removed; MonteCarloDashboard now lives as the "Monte Carlo Risk" tab inside the Financial Engine (deep links land there). Its analysis-local inputs (iterations/seed/variables) stay.
Strategic Planning is now an OUTPUT page: feasibility land/grid/climate auto-derive from the active candidate site (Site Intelligence) with source chips + edit-at-source links; target PUE from the engine matrix; expansion footprint/growth/horizon/capex-per-MW derive from Requirements growth, contract duration and CAPEX results. Acquisition comparables stay local (genuinely market data).
Added
CDU / Liquid Cooling rebuilt: loop hydraulics (Darcy-Weisbach ΔP, Reynolds, pump power, dew-point margin), PUE-impact panel (air vs D2C vs current from the engine matrix), refrigerant selection table (engine GWP/COP database, click-to-select shared capex field), and the Deep-Sea Water Cooling ADVANCED section — gated on the shared capex deep-sea tick: seawater flow, intake temp @ depth, pump station N+1, chiller-less PUE, marine CAPEX/OPEX + screening warnings, with depth/pipeline/ΔT parameters editable in place. Full-standard PDF export.
v1.85.1PATCH
DCMOC batch 2: Site Intelligence full integration + Requirements 1.6 Infrastructure
Added
Site Intelligence FULL INTEGRATION: the five sibling analyses (Tax & Incentives, Disaster Risk, Grid Reliability, Talent, Compliance) are now COMPUTED PER CANDIDATE SITE inside the unified page — integrated analysis panels with real engine outputs (grade/outage-minutes/required gen, composite risk/insurance/EAL, incentive value/NPV uplift/rank, time-to-staff/recruitment cost, mandatory items/costs), radar axes remapped to the integrated engines, comparison table gains 7 engine rows, deep-dive links preserved.
Requirements 1.6 Infrastructure & Site Options: the capex-calculator questionnaire surfaced in Requirements writing straight to the SHARED capex store — Deep-Sea Cooling (tick + depth/pipeline/ΔT), Front-of-Meter (substation/transformer lead/utility rate), Building & Site (building type, seismic zone, site condition, floor), Systems (UPS/gen/fire/alarm/refrigerant from the engine GWP database), Distribution & Security, Sustainability & Renewables (solar MWp + BESS MWh), Delivery basis (year/market/method/fees; contingency stays bound to Design Margin). Deep-sea tick in the CAPEX drawer now shares the same store field.
v1.85.0MINOR
DCMOC revision round batch 1: PDF standard renderer + Cx checklist + spares engine + shell UIUX + reliability depth
Fixed
PDF export crash ("Invalid arguments passed to jsPDF.text") on Operations → Staffing: shiftComparisons shape mismatch mapped correctly in StaffingPdf; systemic hardening at the initDoc chokepoint — doc.text/splitTextToSize now String-coerce and NaN-guard for all 12 legacy jsPDF generators.
Sidebar tooltip clipped at the sidebar edge: React Tooltip now renders via a body portal with fixed positioning + viewport flip — no overflow ancestor can clip it.
Reliability fake precision: "100.0000%" availability and "0.0 min/yr" downtime replaced by nines-formatted availability (with nines chip), seconds-scale downtime, and a β=5% common-cause screening factor on parallel paths.
Added
Standard PDF renderer (print-window): new PrintReport.ts renders every DCMOC export in the OPEX-report min-standard visual language (header + headline chip, KPI cards, configuration table, colored section titles, navy-header tables, SVG donut, tinted callouts, executive assessment, prioritized actions with priority chips, dark summary band, disclaimer). generatePillarPDF now routes through it; jsPDF path retained as fallback.
Export PDF on every engine page: Capacity, CAPEX, Operations, Financial, Sustainability, Results (composite), Assets — plus full-standard upgrades (config/callouts/actions/summary band) for Architecture, Construction, Site rail and Requirements summary.
Commissioning Cx Checklist (DC Hub cx-calculator port): 20 real test-procedure templates (NETA/IEEE/ASHRAE basis — IR/megger, CT ratio, relay, breaker, transformer, UPS, genset load bank, chiller/DLC startup, balancing, IST scenarios, doc review, training, closeout) mapped per readiness level and system; PASS/FAIL tri-state ticks with witness badges (Hold/Witness/Review), expandable procedure detail (steps, acceptance + standard, tools, safety, logsheet fields); checklist-derived completion now drives the engine readinessIndex (sliders become coarse fallback with a "from checklist" chip); FAIL → one-click issue logging.
Spares Optimization rebuilt engine-real: fleet from engine equipment scaling × IEEE-493 MTBF → per-class newsvendor optimization (Q*, ROP, safety stock, fill rate, annual cost) with country lead times, criticality fill targets, per-class overrides and provenance chips — single spares model aligned with Maintenance.
Resizable sidebar: drag the right edge (200–420 px, persisted; double-click resets).
Requirements calculator-basis parity: use-case pick auto-applies the engine profile (density/cooling/tier floor/mix preset) with reference-guidance card; single use-case picker (Industry/Use-case/Workload-category duplication removed); Total Racks overridable (auto chip ⇄ manual); workload-mix manual-override tick; rack-class descriptions on density presets; Grid Voltage gains 20 kV (PLN MV standard).
Reliability Engine deepened: Tier Classification folded in as an in-page tab (sidebar sub-menu removed); per-component RAM table with fleet counts and unavailability contribution; downtime-budget waterfall; MTTR/path-loss sensitivity; maintenance-basis honesty note; Export PDF.
v1.84.3PATCH
DCMOC: Data Library provenance browser
Added
Data Library "Provenance" tab: searchable read-only browser over the engine's DATA.sources ledger (80 entries) — data key, source citation, method, as-of date. Surfaces the gate-enforced provenance discipline (every economically-material DATA value carries a source) directly in the platform UI.
Promised bindings now real (no-placeholder mandate): Settings default currency actually seeds the Requirements currency default; Settings orgName auto-resolves into every PillarPdf export ("Prepared for …" footer) without per-caller changes.
Project breadcrumb per the DC-OS reference: header now reads Projects / {active project} / {page} — Projects clickable, active project name violet, wired to the projects store.
Input-vs-generated visual convention (owner UX rule): editable input labels now carry a 2px violet left-accent (shared Field primitive) — derived/generated values keep the flat tinted read-only panels with provenance chips. Instantly distinguishable.
Sidebar labels aligned to the new engine pages (CAPEX Engine, Architecture Engine, Construction Engine, Commissioning Engine, Asset Intelligence, Reliability Engine, Sustainability Engine, Capacity Planning, Results Engine).
RZExplain tooltips wired on technical KPI labels (PUE, WUE, MTBF, MTTR, Redundancy) via the shared knowledge DB — renders nothing when a key is absent (safe), per the resistancezero tooltip standard.
v1.84.0MINOR
DCMOC Phase O: PDF export standard
Added
PillarPdf standard extension (additive — every existing call keeps working) per the owner's reference PDF (opex-calculator report): PillarReport gains optional config (full configuration table), callouts (tinted analysis boxes info/good/warn), actions (prioritized HIGH/MEDIUM/LOW table), summaryBand (footer mini-KPI band), orgName (from Settings → "Prepared for" footer). All DCMOC page exports can now compose the full standard: header+KPI cards → configuration → sections → callouts → actions → summary band → disclaimer.
v1.83.0MINOR
DCMOC Phase R: Settings & Integrations
Added
DCMOC "Settings" (Phase R): 3 functional menus — General (org name → PDF exports; default country w/ real Apply through the shared writers; default currency seeding Requirements), Data (per-store reset actions clearing the REAL stores — requirements/sites/trackings/ledger/log/sustainability), Integrations (versioned IntegrationConfig schema: webhook/REST/BMS/export-schedule kinds, real reachability Test via no-cors fetch with honest CORS-limited labeling, secretRef labels only — secrets NEVER stored, backend-bound kinds show their true status). Scalable schema (versioned persist) ready for future backend wiring — zero placeholder controls.
v1.82.0MINOR
DCMOC Phase N: Projects — the workflow spine
Added
DCMOC "Projects" page (Phase N): full-state PROJECT BUNDLES — every program store snapshotted (sim/capex canonicals + requirements + candidate sites + architecture + construction/cx/financial trackings + ops log + sustainability), versioned payloads, ordered sanitized restore (sim → capex auto-recalc → requirements → sites → arch → trackings), max 10, active-project tracking with Update. Lifecycle strip (9 engine dots w/ live completion booleans + deep links — the wired workflow). Template Library = engine cx scenario presets (enterprise-2MW … AI-factory-100MW) applied through the shared writers (itLoad/cooling/redundancy) → lands on Requirements. Scenarios remain the lightweight input-only snapshots (cross-linked).
v1.81.0MINOR
DCMOC Phase I: Results Engine
Added
DCMOC "Results Engine" (Phase I, tab report): final scorecard/verdict for the current configuration — 8 dimension scores as DOCUMENTED deterministic composites over live engine data (Requirements = intake completeness; Site = engine site score; Architecture = 100−0.35×complexity; CAPEX = $/kW vs the cx reference band; Construction = SPI/CPI blend from tracking EVM; Ops readiness = tier availability positioning; Sustainability = PUE band; Financial = screening IRR vs 10%% hurdle via models.roi on the dcContract opex basis). Weighted overall + grade, performance radar, dimension ranking, key financial outcomes (NPV/IRR screening, "not investment advice" note), deterministic recommendations, HONEST validation chips ("computed successfully", not a fake audit). Distinct role vs the Executive Dashboard (live cockpit); full ReportDashboard kept as "Full Report" tab.
v1.80.0MINOR
DCMOC Phase M: Sustainability Engine
Added
DCMOC "Sustainability Engine" (Phase M, tab carbon): engine-real core — GHG Protocol scope 1/2/3 donut (models.carbon.scopes), monthly energy (MW × PUE-matrix × 730h), annual water (engine WUE), energy mix DERIVED from the capex renewable + certification inputs (labeled), documented scorecard composites (PUE band / grid-carbon × mix / WUE band / waste diversion) → grade A–D. Initiatives (progress sliders) + certifications (user-attested status selects) via the sustainability store (EXAMPLE seeds). Old CarbonDashboard kept as "Carbon / ESG Detail" tab.
v1.79.0MINOR
DCMOC Phase L: Reliability Engine
Added
DCMOC "Reliability Engine" (Phase L): per-system availability chains composed ENGINE-REAL from IEEE-493 component MTBF/MTTR (models.reliability series/parallel at the current redundancy paths — chains documented per row), composed overall availability vs the Uptime tier target (BELOW-target flagged), downtime budget, composite MTBF (harmonic) + avg MTTR, SPOF list from single-path components at the config, documented reliability-score composite (availability-margin 40 + redundancy 30 + maintainability 15 + SPOF 15), component-MTBF chart, active failure events from the SHARED ops log (no duplicate ledger). Old ReliabilityDashboard kept as "RAM Detail" tab; risk/tier children deep-linked.
v1.78.0MINOR
DCMOC Phase J: Asset Intelligence
Added
DCMOC "Asset Intelligence" (Phase J): fleet GENERATED engine-real — per-class unit counts from equipment scaling, health from the engine Weibull healthIndex at a user-set fleet age + condition slider (dropdown-first combobox for age), wear-out risk (Weibull CDF), MTBF/MTTR from the engine IEEE-493 component data. Honest per-CLASS aggregates (no fabricated per-unit registry). Category donut, class health/reliability table, health-distribution chart, KPI buckets (excellent→critical, at-risk ≥25%% CDF). Old AssetIntelDashboard (replacement schedule + failure risk) survives as "Lifecycle Detail" tab; asset-lifecycle/cbm/spares children untouched + deep-linked.
v1.77.0MINOR
DCMOC Phase K: Commissioning Engine page
Added
DCMOC "Commissioning Engine" (Phase K): planned plane = the v1.68.0 RICH cx engine (L0–L6 staffed-duration timeline bars, equipment-scaled systems list, IST scenarios/tier); actuals = cxTracking store — per-level completion sliders feed the ENGINE readinessIndex (real linkage, engine weights), test counters, issues & punch CRUD (EXAMPLE seeds, Plan-Mode banner); tests-per-system = labeled screening (unit counts × tests-per-unit table); program-cost-share donut (engine fixed proportions). The rich cost + Monte-Carlo band + tornado cards survive as the "Program Cost & Risk" tab (absorbed by composition).
v1.76.0MINOR
DCMOC Phase H: Financial Engine
Added
DCMOC "Financial Engine" (Phase H): budget baseline = engine capex P50 + approved change orders (checkbox revisions); committed/paid from a fraction-scaled tracking ledger (transactions + AR/AP invoices CRUD, EXAMPLE seeds, Plan Mode); CPI/SPI passthrough from the Construction EVM — single source, never recomputed; Forecast-at-Completion = AC + remaining/CPI; PV budget curve from the engine CPM schedule; Annual OPEX donut via models.opex.totalAnnual on the Phase-Q dcContract preset (labeled basis); Financial Health grade A–E (documented composite 0.3 budget-var + 0.35 CPI + 0.35 SPI); deep-dive links (Investment/Monte-Carlo/Portfolio/Benchmarks/Strategic all preserved); full legacy FinancialDashboard as "Pro Forma (Full)" tab — EditableCell workflow intact.
DCMOC "Operations Engine" (new 'ops' tab, Engine-8 group [ops, sim, staff, maint]): derived plane engine-real (availability shown as the Tier design TARGET, PUE-at-load via partialLoadPUE × occupancy, active IT load, 24h energy cost at the country tariff, Weibull asset-health distribution, shift overview from the staffing model), ops-log plane user-entered (opsLog store: alarms/incidents/tickets CRUD with EXAMPLE-chipped seeds + Plan-Mode banner + PM-week compliance log). 24h load curve = documented deterministic diurnal cosine (±5%%, peak 14:00) — labeled SIMULATED, no Math.random. Shift & People + Maintenance sub-tabs absorb the existing dashboards by composition; quick actions wired to real targets.
Changed (owner UX corrections — retroactive)
CreatableCombobox v2 — dropdown-FIRST: the control now renders as a normal select-style button listing preset options (the simple dropdown experience preserved exactly); "Custom value…" row switches to validated typed entry; "Use default" row clears overrides. Applies everywhere the primitive is used (Requirements density/margin, Architecture, Capacity, Construction).
No fabricated sites: Site Intelligence now seeds exactly ONE candidate site bound to the active scenario country; multi-site compare appears only after the user adds sites.
v1.74.0MINOR
DCMOC Phase F: Construction Engine
Added
DCMOC "Construction Engine" (Phase F): Plan-Mode-first tracking — planned plane engine-real (CPM schedule, PV S-curve, engine milestones, long-lead procurement w/ derived PO-by/ETA + example-labeled vendors, screening trapezoid manpower), actuals user-entered (constructionTracking store: status month, per-phase actual %% combobox, AC spend, risks/issues CRUD w/ EXAMPLE seeds). Deterministic EVM SPI/CPI (baseline 1.00 in Plan Mode; single source for Financial), forecast = planned/clamp(SPI), documented health composite (SPI 40 + CPI 30 + schedule 15 + issues 15). L2 Gantt reused.
Fixed
Latent zero-schedule bug: old dashboard fed capex.timeline (no duration keys) into models.construction.schedule → durations 0, "Total Build" 0. Durations now mapped explicitly from timeline phases + engine long-lead months.
Phase-editor write-back bug: legacy CapacityDashboard kept phases in local React state and never wrote back to inputs.capacityPhases — Strategic/Phased-Finance/Report only ever saw defaults. Debounced store write-back added (same schema).
v1.71.0MINOR
DCMOC Phase C: Architecture Engine + dynamic system diagram
Added
DCMOC "Architecture Engine" (Phase C) — per the DC-OS reference with a fully DYNAMIC system-architecture diagram (owner mandate: not a static image): pure layout engine computes blocks + orthogonal edges from live requirements — utility feeds/trunks follow redundancy (N+1 single trunk + spare · 2N dual A/B · 2N+1 dual + spare), UPS/generator/switchgear/PDU/rack counts from the engine equipment-scaling model (per-module MW from the redundancy factor), cooling chain follows coolingType (liquid/rdhx → CDU loops + heat rejection · air/inrow → chiller + CRAH), voltage labels from the requirements grid voltage, network fabric leaf estimate, BMS control edges. Logical + Single-Line-Diagram skins of the same graph, pan/zoom, 5-line-type legend. Changing ANY upstream requirement re-renders the diagram.
Header selects: Design Standard (Uptime III+/IV, TIA Rated-3/4 → writes shared tier) + Architecture Profile presets (multi-param registry writes: AI-liquid/AI-rdhx/colo/enterprise). KPI row (IT/facility MW, design PUE via pueMatrix direct lookup — defaultFor key-trap avoided, availability target, redundancy). Absorbed old dashboard cards: complexity → KPI/BOM, ASHRAE thermalCheck → Cooling card, topology + floorLoading → Power card, disciplines + designFee → BOM section. Rail: Architecture Summary, Load-Breakdown donut derived from LIVE PUE (labeled screening split), per-layer Design Validation. Bottom: Key Design Decisions, Reference Design id (pattern template), HONEST compliance (Uptime/TIA/ASHRAE engine-derived %; NFPA + ISO/IEC 22237 = SCREENING chips, no fake bars), Next Steps → Capacity.
Fixed
zustand v5 object-selector infinite-render trap (React #185) in the diagram pan/zoom state — selectors split per-field.
v1.70.0MINOR
DCMOC Phase B: multi-site Site Intelligence Engine
Added
DCMOC "Site Intelligence Engine" (Phase B) — the single-country Site Score tab becomes a MULTI-SITE comparison engine per the DC-OS reference: up to 5 candidate sites (3 illustrative EXAMPLE-labeled seeds, fully editable), schematic SVG site map (4 style skins, no map lib), 8-axis radar overlay (recharts), site cards with rank badges, 6 detail panels (Power/Connectivity/Environmental/Risks/Land/Cost with per-value provenance dots: site attribute vs country baseline), Site Score & Compare table (+ absorbed legacy 10-factor engine breakdown), sticky rail (Run Analysis stamp, ranking bars, deterministic key takeaways, Next: Architecture), Edit Criteria drawer — every numeric attribute a CreatableCombobox (preset or custom, clamped; clear → country baseline).
site-adapter (src/lib/site-adapter.ts): site attributes → engine factor overrides MIRRORING models.site.deriveFactors formulas exactly (SAIDI/PGA/tax/flood/water/power); PARITY INVARIANT verified — a site with no overrides reproduces the engine country score exactly (42.1 ≡ 42.1). Engine remains the authoritative Total Score; the 8 axes are a documented presentation decomposition. "View Power/Risk/Tax Analysis" links route to the existing grid/disaster/tax tabs — all 6 Engine-2 children preserved.
Changed
Sidebar label 'Site Score' → 'Site Intelligence' (tab id unchanged). Old SiteIntelDashboard export dormant (content absorbed).
DCMOC "Requirements & Workload Engine" page (Phase A of the DC-OS UIUX program) — replaces the thin Requirements intake dashboard with the full reference design: 1.1 Project Overview (project/customer/COD/grid-voltage/project-type/contract/currency/use-case), 1.2 Workload Profile (IT/peak/avg load w/ MW|kW toggle, rack density CreatableCombobox, computed Total Racks, auto-normalized workload-mix sliders, AI chip, cooling approach), 1.3 Growth Plan (Y0 always = shared IT load; linear/step/custom; 5-yr CAGR), 1.4 Availability Target (engine tier fraction → %, downtime budget min/yr, SLA check), 1.5 Business, Margin & Priority (budget, Design Margin wired into CAPEX contingency per owner mandate, rankable priorities), 1.7 Summary (engine-real models.requirements.validate checklist + flags + PDF). Sticky right rail: deterministic AI Insights, Requirement Score ring, Active Scenarios, per-section Input Quality bars, Next: Site Intelligence.
Phase 0 state layer now live: requirements store registers as the registry's req.* provider; shared fields (itLoad/cooling/country/tier/rack-density) written through mapping helpers into BOTH simulation + capex stores (no drift).
Changed
Sidebar: Engine-1 children → [Requirements & Workload]; Staff Model Config moved to Engine-8 Operations (sim/staff/maint) per owner. Old RequirementsDashboard deleted after absorption.
v1.68.1PATCH
fix: root-tier lockout on premium gates, 22 pages
Fixed
Root accounts were locked out of premium actions site-wide (owner report: "export PDF capex-calculator tidak bisa"). Supabase profiles.tier legitimately holds 'root' (rz-supabase.js allow-list), but 22 pages carried LOCAL strict gates (tier === 'pro' / userTier !== 'pro') that treated a root session as free → login modal instead of the action. Patched every local gate to accept pro || root (capex-calculator gatedAction/isFullPremium/exportCapexCSV path, dc-market-tracker, roi/carbon calculators, FF-1/2/3, geopolitics-3, dashboard tier badge, 13 article premium blocks). Headless regression probe: seeded tier:'root' session → gatedAction('pdf') passes on capex + market-tracker (was BLOCKED). Note: js/rz-feature-flags.js + auth.js getTier() were already root-aware — only page-local checks had drifted.
v1.68.0MINOR
DCMOC 7·Commissioning wired to the RICH cx engine
Added
Rich commissioning engine promoted into rz-engine.js (models.commissioning.*, DATA 2.4.0→2.5.0) — a faithful port of the DC-Hub cx-calculator.html model so DCMOC and the standalone calculator share ONE brain. New pure models: equipScale (26 equipment counts scaled from IT load + rack density), levelDurations/levelCosts (per-level L0–L6 staffed durations + costs at 30 regional day-rate cards — cxDay/fieldDay/oemDay/witnessDay + per-diem + diesel), programRich (gm-normalized ^0.45 base-vs-level-sum blend + campus adders + 15%% contingency → grand total, per-level + per-discipline splits, pctCapex, tier/availability), monteCarlo (N=10000 Box-Muller, itLoad ±7.5%% + pricing ±5%% → P5/P50/P95/CVaR95, seedable for tests), sensitivity (7-param tornado), and mapInput (DCMOC store → rich schema; liquid→dlc, ISO-2 country → nearest CX region, redundancy passthrough). New DATA.commissioning.cx.rich tables (30 rates + cooling/redundancy/building/seismic/substation/BMS/delivery/scope/fire/UPS/gen/density/base + fixed display proportions + 8 scenario presets). Compact programCost/programSchedule kept for back-compat.
DCMOC CommissioningDashboard (Layer 7) rewired to the rich model — equipment-scaled program cost + %%-of-capex, per-level L0–L6 duration+cost bars, equipment-count grid, discipline breakdown, a Monte-Carlo cost-uncertainty band (P5–P95 + P50 marker + CVaR95 + CoV), and a sensitivity tornado. All values move with IT load / cooling / redundancy / country. Graceful fallback to the compact estimate if a stale engine build lacks programRich.
Changed
Engine gate tools/test-rz-engine.mjs 395→421 asserts (rich-cx worked examples with golden values computed from cx-calculator.html's own cxCalcTotalCost — exact grand/subtotal/contingency/duration/equipment parity across enterprise_2mw/hyperscale_50mw/colo_10mw). Reference-parity 126/0 unchanged. rz-engine.min.js rebuilt; DCMOC engine cache-bust dcos2→cxrich; bridge header 2.4.0→2.5.0.
v1.67.6PATCH
audit fixes: LOW polish D11/G3
Changed
Pillar Metric cards (NewEngineDashboards + DesignTools) now lift + glow on hover (intuitive). OPEX KPI labels its basis ("DC-contract rate · 100%% util") + the tooltip explains why it can differ from the standalone OPEX calculator (retail rate + partial utilization) — the documented dual-rate made visible.
v1.67.5PATCH
audit fixes: honesty D8/D9
Fixed
Integrations panel no longer shows hardcoded true — RZ Engine status is a real check (rzModels().capex resolved), gateway labeled honestly "configured" (no fake live-ping). Report availability fallback used Tier-3 (99.982%%) for all tiers → now tier-correct (Tier 2 99.741 / 3 99.982 / 4 99.995) or "—".
v1.67.4PATCH
audit fixes: gaps + docs D7/D10/G1/G2
Fixed
Site data completed: the 7 countries missing from SITE_AUGMENT (CL/CO/KE/NG/NZ/PL/PT) now carry real WRI-Aqueduct/ASHRAE-zone/SAIDI/PGA values → all 32 countries score on real data (was neutral fallback). Missing tooltip: added tab-dashboard to the RZExplain DB (779 entries) + bumped the DCMOC ?v. Dead tab: removed the unrouted scenarios id from the store union. Doc drift: ENGINE_UNIFICATION.md test count 299/0→395/0; rz-engine.ts bridge header v2.3.0→DATA.version 2.4.0. Engine 395/0, parity 126/0, JS-audit clean.
v1.67.3PATCH
audit fixes: DCMOC integrity D1-D6
Fixed
Executive Dashboard site score was hardcoded (constant 68.4 regardless of country, diverged from the Site Intel pillar) → now uses models.site.deriveFactors(country) — the two surfaces agree. Revenue default diverged across 5 surfaces ($120/$150/$280 → contradictory IRR) → single shared DEFAULT_REVENUE_PER_KW_MONTH (constants/finance.ts) in Executive/Financial/Report. SimulationDashboard hardcoded $20,000 salary + $1,126.30 overtime → real selectedCountry.labor + derived OT premium. TierDashboardnetwork sub-score was fixed 70 → derived from tier. StrategicPlanning acquisition ROI used fixed 5MW → real inputs.itLoad. PortfolioDashboard "Grid Intensity" column showed PUE values → relabeled "Site PUE". tsc+build clean.
v1.67.2PATCH
audit fixes: engine correctness E1-E6
Fixed
CRITICALmodels.decision.recommend crashed when objectives passed as {} (guard only covered null/undefined). MAJORmodels.tax.macrsDepreciation treated discountRate=0 as 0.10 (~30%% wrong undiscounted NPV). models.architecture.topology(1) now returns real Tier-1 (was Tier-3 fallback). Climate free-cooling denom 5500→5800 so ASHRAE zone 8 (subarctic) scores above zone 7. Added DATA.sources['decision'] provenance (was missing, CLAUDE.md rule). +5 edge-case asserts (test-rz-engine 395/0, parity 126/0). From a 3-agent adversarial audit.
v1.67.1PATCH
DCMOC: surface AACE accuracy band on CAPEX
Added
CapexDashboard shows the AACE 18R-97 Class-4 budgetary accuracy range (−30%%/+50%%) under Total CAPEX (models.capex.accuracyRange) — honest estimate confidence band. (Carbon scope 1/2/3 already computed locally in CarbonDashboard.)
v1.67.0MINOR
engine research-deepening: pillars 5/10/13 — all 13 covered
engine research-deepening: Pillar 2 Site + per-country data
Added
Pillar 2 Site deepened with real per-country data (25 major DC markets augmented): WRI Aqueduct 4.0 water-stress (0-5), ASHRAE 169-2021 climate zone → free-cooling hours, IEEE-1366 SAIDI grid reliability, USGS PGA → seismic design category. models.site.deriveFactors now uses them: water is no longer hardcoded (0.65→real, e.g. Saudi 0.00 desert), a climate free-cooling factor was added (Sweden 0.91 vs Singapore 0.15), grid prefers SAIDI, seismic prefers PGA (Japan 0.10). Site scores now discriminate realistically (Sweden 80 B prime vs Indonesia 42 D). DATA.sources provenance; test-rz-engine 376/0, parity 126/0. 7 of 13 pillars research-deepened.
v1.65.6PATCH
DCMOC: surface facility-load + lease-up S-curve on Capacity
Added
CapacityDashboard shows PUE-adjusted facility load (models.capacity.facilityLoad) + year-2 lease-up occupancy by market type (occupancyScurve: hyperscale/wholesale/retail). Completes surfacing of all 6 research-deepened pillars (1/3/4/6/8/9) in the DCMOC UI.
v1.65.5PATCH
DCMOC: surface Uptime staffing benchmark on Staffing
Added
StaffingDashboard shows the Uptime Institute critical-facilities staffing benchmark (models.maintenance.staffingBenchmark) vs the configured headcount — FTE/MW by tier, with an aligned/above/below variance chip.
v1.65.4PATCH
DCMOC: surface density-safety + rack-count on Requirements
Added
RequirementsDashboard now feeds the real cooling type into models.requirements.validate and surfaces the density band, implied rack count, and a CRITICAL red flag when the workload density exceeds the cooling ceiling (e.g. AI on air) — ASHRAE-TC9.9 safety check.
v1.65.3PATCH
DCMOC: surface Weibull wear-out risk on Asset
Added
AssetIntelDashboard surfaces models.asset.failureProbability — per-class Weibull cumulative failure probability at 60%% of design life (battery/ups/generator/crac/chiller/transformer), color-coded, driving condition-based replacement timing.
v1.65.2PATCH
DCMOC: surface research-deepened engine outputs
Added
ConstructionDashboard surfaces models.construction.longLeadRisk — critical long-lead gear (transformer/switchgear/genset/UPS/chiller) vs power-on month, flagging what to pre-order (rose = critical). ArchitectureDashboard surfaces thermalCheck (ASHRAE TC9.9 compliance) + topology (Uptime/TIA-942-C rating, power/cooling paths, floor loading). Makes the v1.65 research depth visible.
v1.65.1PATCH
engine research-deepening: pillars 6/8/9
Added
Pillars 6/8/9 deepened (research pass cont.): L6 construction long-lead procurement risk (transformer 60-120wk, switchgear, genset, UPS, chiller — the dominant AI-era schedule driver, models.construction.longLeadRisk); L8 Uptime critical-facilities staffing benchmark (FTE per position × tier + per-MW techs, models.maintenance.staffingBenchmark); L9 Weibull wear-out failure curves per asset class (models.asset.failureProbability, β/η from IEEE-493 + manufacturer MTBF). DATA.sources provenance + asserts (test-rz-engine 373/0, parity 126/0). 6 of 13 pillars now research-deepened.
v1.65.0MINOR
engine research-deepening: pillars 1/3/4, cited
Added
Deep-research enhancement of the DC-OS engine (pillars 1/3/4) from authoritative standards. L1 requirements: coolingMaxRackKw density ceilings (ASHRAE TC9.9 5th ed. — air ~20kW/rack; NVIDIA GB200 132kW), rackCount, densityBand + a CRITICAL density-vs-cooling safety flag. L3 architecture: ASHRAE thermal-envelope check (A1-A4/H1 supply-temp + ΔT), Uptime/TIA-942-C tier topology, floor loading, design-fee. L4 capacity: logistic S-curve lease-up by market type (CBRE H1 2025), PUE-adjusted facility load, stranded-capacity (Uptime 2024), power-vs-space binding. All with DATA.sources provenance + worked-example asserts (test-rz-engine 366/0, parity 126/0). Pillars 2 (site) + 5-8 + 9-13 research follow.
v1.64.4PATCH
DCMOC: engine tooltips + hover on the KPI efficiency row
Changed
KpiEfficiencyRow (PUE/WUE/CUE/Availability/LCC) now has hover-lift + engine-sourced <Explain> tooltips (glossary keys pue/wue/cue/availability/tco). Completes the metric-tooltip sweep across the dashboard's main panels.
LifecycleStrip "7. Commissioning" chip was dead — it (and several others) had a missing/stale tab mapping, so the button was disabled. All 13 chips now route to their real pillar tab (Requirements→requirements, Site→site, Architecture→architecture, Construction→construction, Commissioning→commissioning, Assets→asset-health, AI→dashboard).
Changed
Lifecycle chips are now color-coded per engine, identical to the sidebar submenu numbers (ENGINE_COLORS 1–13) — intuitive alignment; each chip's icon + number + border + hover glow use its engine accent. The Engine Dependency Graph matches.
Metric tiles get hover highlight + engine-sourced tooltips (CapacityArchOverview): each tile lifts on hover and carries an <Explain> tooltip pulled from the RZExplain knowledge DB (glossary keys — NOT hardcoded strings). Tiles also made responsive.
Data Flow & Digital Thread is now LIVE: each stage shows its real count (configured inputs / 13 engines / N model namespaces / computed outputs / AI decisions / actions) with a flowing pulse + a "live" indicator — no longer a static image.
StrategicPlanningDashboard acquisition ROI now derives annual OPEX from models.opex.totalAnnual (engine-real, country + PUE aware) instead of a flat 8%-of-CAPEX heuristic — with a graceful fallback to the heuristic when the engine is unavailable. Backend /calc redeployed with the v1.64 rich models live.
v1.64.1PATCH
DCMOC: Platform surfaces made real, engine-backed
Added
The 8 DCMOC sidebar Platform/Support entries (were dead "coming soon") are now functional, engine-backed surfaces (PlatformDashboards.tsx):
Data Library — read-only browser of the canonical rz-engine.js DATA (32 countries with electricity/tax/grid-carbon/constr-index, 25 markets, PUE matrix, refrigerants) — proves the single source.
Templates — one-click apply of the engine use-case profiles (models.requirements.useCaseProfiles: AI/HPC/cloud/colo/enterprise/edge) to the project.
Projects — saved-scenario list (restore/compare/delete).
Knowledge Base — live engine-model list + glossary link. Integrations — real connection status (Supabase/engine/backend/AI). Audit / User Management — honest status tied to the Supabase audit_log + rz-ops admin.
DC-OS: promote the RICH standalone models into the engine
Added
The richer models that lived only in the standalone tools are now full-fidelity in rz-engine.js (extending, not replacing, the simple functions — tier.classify/fire.agentQuantity/cdu.size/spares.eoq unchanged):
models.tier.advise() — 6-band Uptime grading (I / I+ / II / II+ / III / IV) + per-category scores + canT3/canT4 floor constraints + recommendations (from tier-advisor.html).
models.cdu.hydraulics() — water/glycol properties + Darcy-Weisbach ΔP (Haaland friction) + Reynolds + pump power + Magnus dew-point margin (from js/cdu-engine.js).
models.spares.newsvendor() — critical ratio, Q* via Φ⁻¹(CR) or Poisson (low-demand), fill rate, annual cost curve (from spares-readiness-calculator.html).
Constants moved to DATA.{tier,fire,cdu,spares} with DATA.sources provenance. tools/test-rz-engine.mjs355/0 (+34 asserts). cf-worker/src/calc.js allow-lists the new data.
DCMOC pillars upgraded to the rich functions: Fire (NOAEL margin + CO₂e + cylinders), CDU (ΔP + Reynolds + pump kW + dew-point margin), Spares (newsvendor Q* + fill rate). Tier pillar stays on classify (the 6-band advise needs the full topology dropdowns the tool collects). All with graceful fallback to the simple function.
standalone tools: engine as fallback where the dedicated engine is richer
Changed
fire-calculator.html + cdu-calculator.html now load rz-engine.min.js and delegate to models.fire/models.cduonly as a fallback — their dedicated FIRE_ENGINE/CDU_ENGINE (NOAEL safety, GWP, thermohydraulics, NPSH) are richer, so they stay primary. tier-advisor.html (6-band I/I+/…/IV grading) + spares-readiness-calculator.html (newsvendor/fill-rate) kept inline — their models are richer than the engine's 4-band tier / EOQ; promoting the richer logic into rz-engine.js is the honest next step, not downgrading the tools. Audits CLEAN; no calculator behavior changed.
setup-supabase.html repurposed → DC-OS manual-action checklist (token rotation, backend deploy for the v1.63.0 /calc models, tests).
models.decision.recommend() — Layer-13 deterministic decision brain in the engine (mirrors the DCMOC provider): always-on, never-empty, explainable recommendations (PUE/availability/cost-band/financial/density/schedule/site rules + objective ranking + disclaimer).
All 5 registered on the cf-worker /calc allow-list (backend-served, anti-theft). tools/test-rz-engine.mjs +10 asserts → 321/0.
DCMOC surfaces the 4 new pillars (DesignToolsDashboards.tsx): Tier Classification (Reliability group), Fire Suppression (Architecture), CDU/Liquid Cooling (Capacity), Spares Optimization (Asset) — each engine-real + input-driven, in the sidebar under its lifecycle engine.
Retail vs DC-contract electricity rate documented, not force-merged:DATA.countries.economy.electricityRate is the retail/display rate (single-sourced across calculators + DCMOC); models.opex keeps its calibrated DC-contract blend (what the cockpit-accuracy gate is validated against) — the two are intentionally distinct (a DC gets PPA/wholesale rates), with ppaRate override available. This resolves the user-visible rate divergence without degrading OPEX accuracy.
models.site.deriveFactors(countryId) — derives the 0-1 site factor vector (power/grid/seismic/talent/tax/carbon/flood/latency/water) from DATA.countries, so site.score() is REAL and country-varying (SG 70, US 72, DE 61, ID 55). Was: a hardcoded factor vector that produced a constant score regardless of country.
models.commissioning.programCost() + programSchedule() — full L0–L6 commissioning PROGRAM cost + schedule (discipline $/kW base, level cost/schedule shares, discipline split, cooling/redundancy multipliers, region scaling via DATA.countries.constructionIndex), promoted from cx-calculator.html's inline logic into the shared engine (DATA.commissioning.cx + DATA.sources row). Cost/schedule now move with itLoad/cooling/redundancy/country.
DCMOC integrity fixes ("de-fake"): the Commissioning module fed a hardcoded {L1:1,L2:1,…} completion vector to a real engine function (constant output); it now shows the LIVE models.commissioning program cost + schedule + discipline breakdown from the current project inputs. The Site Intelligence module fed a hardcoded factor vector; it now derives factors from the selected country via models.site.deriveFactors. (DCMOC static app — versioned separately; noted here for the engine wiring.)
DCMOC Layer-13 AI Assistant: the "AI Assistant" button now opens a config modal — plug in an OpenAI/Anthropic/custom API (key stored only in-browser) and the decision layer routes through it; empty ⇒ the built-in deterministic RZ engine runs everything (auto-fallback on any error). remoteApiProvider is now runtime-configurable.
Root calculators single-sourced:opex-calculator.html, carbon-footprint.html, tco-calculator.html now resolve country electricity rate / grid-carbon / PUE from RZEngine.data.countries + pueMatrix (inline fallback), killing the per-tool divergence (Singapore rate now 0.22 everywhere; air PUE 1.50). Also fixed: carbon-footprint.html previously referenced the engine only inside a PDF template string — it now actually loads rz-engine.min.js. Audits (audit-js-syntax, audit-script-tags) CLEAN.
DCMOC Phase C wiring: 9 modules (GridReliability, DisasterRisk, Compliance, Capacity, AssetLifecycle, CBM, FuelGen, TaxIncentive, Carbon) now delegate to their existing rzModels() engine models with local fallback; Carbon reconciled to models.carbon.annualTonnes (agrees with the Executive Dashboard). DCMOC module KPI grids made mobile-responsive (12 files); dashboard 0 horizontal overflow at 390px + 768px.
Still to build (deferred, need judgment): models.decision (promote the DCMOC deterministic provider into the engine), and consolidate fire/cdu/tier/spares (standalone js/*.js + inline tool logic) into shared models.*.
Dual DC electricity-rate (DATA.regions.powerKwh DC-blend vs DATA.countries.economy.electricityRate retail) intentionally NOT force-merged — needs a deliberate accuracy decision on the canonical DC power rate.
v1.61.0MINOR
DC-OS engine unification P-A: single-source country reference
Added
rz-engine.js DATA.countries — THE single source of truth for region/country economics
Phase 1 (shared system + article-26/article-27 reference pair). The shared CSS/JS lands on all 27 editorial articles; the bespoke calculator/chart work is the reference pair, with the rest to roll out in batches.
Fixed
Article body alignment ("nggak lurus"). Paragraphs were individually centered with a ch-based measure, so the 1.14rem lead paragraph and 1rem body paragraphs computed different widths and landed on different left edges. Replaced with a single left-aligned, justified reading column at a fixed rem measure (css/rz-article-dark.css) — every prose block (incl. the drop-cap lead), the hero figure, and tables now share one left+right edge.
Callout AI-design-slop..info-box(+variants), .ws-insight-box, .ws-engineer-note, .ws-evidence-block carried saturated gradient fills + 3–4px borders. All adopt the editorial language: flat color-mix tint + 1px hairline + a 2px semantic accent rail + one restrained shadow + capped radius — scoped to the editorial register so it lands site-wide. Inner callout text re-aligned to the heading edge.
Article number badges washed out (articles.html): white text on saturated category gradients → one legible dark-glass instrument chip (hairline + mono tabular number) that reads on any thumbnail in both themes.
Added
Section dividers — a tier-3 hairline above each <h2>, aligned with the hero/reading column.
Related-articles rail — a compact sticky right rail built by js/rz-article-editorial.js from each page's existing "Continue Reading" cards (thumbnail derived from the href). 2-col on wide screens, collapses to the bottom grid < 1024px. Zero per-article markup edits.
Calculator tooltip compliance (article-27.html.ws-calc-wrap): all 12 inputs + 9 KPI cards now carry .tip help affordances (formula / denominator / source) per TOOLTIP_STANDARD.
Interactive, validated-data charts — new js/rz-article-chart.js renders on-brand Chart.js charts (CNBC-style hover crosshair + tooltip, instrument tokens, theme-aware, finding-titled) from an inline sourced config, with a visible Source caption + basis chip. Reference charts: article-26 dual-axis fluid-loss line (from the verified data/article-26/worked-model-scenarios.csv) and article-27 aging-workforce bar (new data/article-27/workforce-stats.csv, AFCOM 2024). New gate tools/audit-article-charts.mjs blocks any chart without a source+basisTag. See standarization/ARTICLE_DATAVIZ_STANDARD.md.
Changed
tools/audit-dark-coverage.mjs + tools/audit-responsive-layout.mjs hardened (--disable-dev-shm-usage, resilient page close) to survive headless browser crashes when cycling many pages.
v1.50.0MINOR
index hover fixes + WhatsApp removal + Contact box enhancement
Fixed
Hover wobble (index.html cards) — the company-logo hover used a springy easing
cubic-bezier(0.34, 1.56, 0.64, 1) (Y=1.56 overshoots) that made the logo bounce/oscillate. Replaced
with a smooth cubic-bezier(0.4, 0, 0.2, 1) + smaller scale(1.02) → a subtle micro-movement.
Hover blink/disappear (index.html cards) — the global hover rule set animation: none !important
while the entrance animations (swingIn on .oe-card, bentoRise on .bento-exp-card) stayed bound
to the cards, so every mouse-leave re-applied and replayed the entrance from opacity:0. Several
reveal observers also kept re-adding .visible. Fix: the IntersectionObserver now unobserves after the
first reveal, an observer-independent init-time handler freezes each card's entrance animation
(animation:none) once it ends so it can never replay, and the animation:none was removed from the
hover rule. Verified by instrumented hover-testing: 0 entrance-animation restarts on hover.
Removed
Personal WhatsApp number removed from index.html — the visible contact card
and all four structured-data spots (Person telephone+sameAs, ProfessionalService telephone,
Organization sameAs+contactPoint). The generic "Share on WhatsApp" share-bar button (shares the page
URL, not the number) is kept.
Changed
Contact box enhanced (index.html.contact-info): availability line with a pulsing status dot +
a primary "Email me" CTA; the methods are grouped under labels (Direct contact · *Find me
elsewhere · Based in*); each email has a copy-to-clipboard button with "Copied!" feedback; refined
card styling with matching dark-mode overrides. Re-minified styles-index.min.css + script.min.js,
cache-busts bumped.
v1.49.10PATCH
cdu-calculator — slashed-zero KPI numerics
Changed
cdu-calculator.html KPI value displays (.kpi-val, .cc-pp-kpi .v) now use
font-variant-numeric:tabular-nums slashed-zero, matching the instrument-grade numeric style
already used on fire-calculator.html and the checklist tables. Closes the last item of the
CDU-hub uiux back-port (theme-key was already unified in v1.49.5; dark chips already present).
v1.49.9PATCH
Add root favicon.ico — kill the site-wide /favicon.ico 404
Fixed
Added a multi-resolution root favicon.ico (16/32/48 px, generated from assets/Favicon.png).
The site declared a PNG favicon via <link rel="icon">, but browsers still auto-request
/favicon.ico from the web root — which 404'd on every page (a console error visible on all
100+ pages). A health probe across the CDU + Fire pages confirmed this was the only remaining
console artifact; with the file in place, those pages now report 0 console errors and 0 4xx.
v1.49.8PATCH
Responsive reading column + responsive tables
Fixed
Articles no longer sprawl or sit left-stuck on wide screens. Root cause: there was no site-wide
reading-width rule on .article-content / .article-body (only inside @media print), so the body grew
with the viewport while the 68ch paragraph cap held text left-aligned and tables filled the full body width
— at a 2400px viewport article-27's body reached 1520px with a 1520px table beside a 930px text column
("table lebih lebar dari text, berantakan"). Added a shared reading-layout system in styles.css:
.article-body is capped to 1180px and centered, and its direct prose children (`p, h2–h4, ul, ol,
dl, blockquote, figure, pre, table) share one centered 760px` reading column. Because only direct prose
children are capped, the interactive widgets embedded in .article-body (calculators, strategy grids,
gantt charts) keep full width — verified article-27's Workforce calculator still renders at 1116px.
Tables scroll instead of overflowing on phones..article-body table becomes an overflow-x:auto
scroll container ≤900px. Fixes article-1 (4 data tables, was +210px horizontal page scroll) and
tia-942-checklist (.gap-table, was +38px).
EPMS_Telemetry no longer scrolls sideways on mobile. The position:absolute.ui overlay held a
fixed-height toolbar wider than a phone; the page's overflow-x:hidden guard was defeated by the
overflow-x:hidden+default-overflow-y:visible scroll-promotion quirk. Switched the guard to
overflow-x:clip and constrained the toolbar to 100vw with internal horizontal scroll (cockpit stays a
desktop dashboard).
Editorial skin (css/rz-article-dark.css) reconciled: the 68ch measure is now centered
(margin-inline:auto) and extends to direct-child tables, so editorial articles match the new column.
Added
tools/audit-responsive-layout.mjs — new ship-gate. Renders every content page at 390 / 768 / 2400px
and FAILS on real user-facing horizontal scroll (measured via actual scrollX, not the scrollWidth
artifact) or an article prose table wider than the reading column. CLEAN across 113 pages. Added to the
CLAUDE.md audit suite. See standarization/RESPONSIVE_STANDARD.md "Article reading column".
v1.49.7PATCH
Cooling pillar — link the CDU toolkit
Added
pillar-cooling.html now includes a Liquid-Cooling CDU Toolkit resource card (→ cdu-hub.html),
placed next to the Air-vs-Liquid comparison. The pillar's own intro promises it "links every
cooling-related resource," but the CDU suite — the biggest recent cooling addition — was missing.
Found via a cross-linkage evaluation (search-index coverage for all 9 new pages is complete; the
fire pillar↔tools links were already in place; this was the one real pillar→tool gap).
v1.49.6PATCH
DC Solutions hub — surface the Fire suite
Added
datacenter-solutions.html Engineering & Compliance Tools now lists the **Fire Suppression
Calculator (fire-calculator.html) and Fire Safety Checklist** (fire-checklist.html) alongside
the CDU Toolkit — previously only the Fire-Safety pillar was linked, not the interactive tools
(the same discoverability gap just fixed on tools.html). Tool-count badge 9 → 11.
Fixed
Corrected a stale <!-- coming soon --> code comment on the PLN Sumatra card (the card itself has
been published/linked since v1.46.6).
v1.49.5PATCH
Theme persistence — unify the localStorage key site-wide
Fixed
Cross-page theme persistence. 19 pages stored the theme under non-standard localStorage keys
(rz_theme on the cdu-, compare-, pln-java-grid-*, dc-market-tracker, tier-advisor pages;
rfs_theme on rfs-readiness-workbench) while the rest of the site + script.js + the FOUC guard
use 'theme'. Result: setting dark on one page didn't carry over when you navigated to one of
these — it reverted to its own key's default. Unified all of them to 'theme' so the theme
persists across the whole site (the admin panel's separate rz_admin_theme is intentionally left).
Verified: the affected pages now honour the shared key on load and write it on toggle; dark-coverage
gate CLEAN in both modes; JS audits CLEAN.
Known (pre-existing, separate)
tier-advisor.html has a chartjs-plugin-annotation console error (chart-lifecycle, unrelated to
the theme-key change) — tracked for a separate fix.
v1.49.4PATCH
CDU Mini-BMS — P&ID instruments light up by alarm state
Changed
cdu-mini-bms.html P&ID schematic now colours each tagged instrument by its live alarm state
(amber WARN / red ALARM), mirroring the tiles and the active-alarm banner — so a fault reads
consistently across all three views instead of the schematic only showing the leak marker. Example:
on Filter clog, PDT-01 (filter ΔP) and PDT-02 (loop ΔP) go red and FT-01 (flow) goes amber. The
LT-01 reservoir-level instrument is now also link-interactive (click ↔ its tile) and state-coloured.
v1.49.3PATCH
Both-mode enforcement — fix stuck-dark-in-light + gate covers both themes
The both-mode audit's structural check (body must match the active theme) caught the
inverse of the v1.48.1 bug: pages with a light palette that stay dark in light mode.
Fixed
changelog.html stuck dark in light mode — it had [data-theme="light"] rules for the
content cards but no [data-theme="light"] body, so light mode showed a light navbar over a
dark body (a broken middle state). Added the light body/surface rule to the generator template
(tools/build-changelog-html.py) and regenerated; the hero band stays dark intentionally.
(achievements.html is intentionally dark-only — no light palette — so it's correct as-is.)
Changed — enforcement now covers BOTH modes
tools/audit-dark-coverage.mjs now also fails STUCK-DARK-IN-LIGHT: a page declaring a
light palette ([data-theme="light"] / :root:not([data-theme="dark"])) must render a light
body in light mode. Pages with no light palette are dark-only (cockpits, dark trophy pages) and
skip the light check. Gate CLEAN across 114 pages in both themes.
v1.49.2PATCH
Tools hub — surface the missing Fire suite + PLN Sumatra grid
Added
tools.html now lists the recently-shipped tools that were missing from the hub:
Fire Suppression Calculator (fire-calculator.html) and Fire Safety Checklist
(fire-checklist.html) in Compliance & Standards Tools, and the **PLN Sumatra Interconnected
Grid** (pln-sumatra-grid.html) in Market & Grid Monitors. (All three pages already existed and
were in sitemap/search-index/llms — only the hub listing was stale.)
Fixed
tools.html tool count corrected 18 → 23 across the hero copy, hero badge, meta/OG/Twitter
descriptions, and the JSON-LD CollectionPageItemList — which was also completed to 23 entries
(it had been missing the CDU toolkit and Spares Readiness calculator in addition to the new tools).
v1.49.1PATCH
Nav cleanup — Tools out of the Insights dropdown
Removed
index.html Insights navbar dropdown — removed the "Tools & Calculators" sub-item (it mixed a
tools link into a reading/insights menu). The dropdown now reads Engineering Journal · Global Analysis
· Future Forward · Glossary · — · Second Brain · All Insights, matching the Insights dropdown already
used on the rest of the site. tools.html stays reachable via the global search and page footers;
the calculators remain in the DC Solutions dropdown.
v1.49.0MINOR
CDU checklist spares-planning view + Mini-BMS tile reading-guide
Added
cdu-checklist.html §07 — "Spares planning" subsection turning the parts register into an
actionable stocking plan: per-item min stock on site, typical service life (real replacement
cadence, not just the inspection interval), lead time (the real driver for holding a critical
spare), and estimated annual spend per CDU. Closed with a budget roll-up: recurring
Opteon 2P50: lifetime ~22 d (EPA SNAP), GWP ~2 (was ~10) — propagated to the calculator (runFluidCost) and the worked-model generator; worked-model-scenarios.csv regenerated.
AGAGE portal URL updated to the NASA LaRC archive; GHGRP format corrected to Excel/zip.
Galden (PFPE) GWP flagged unverified/disputed (~10,000 commonly cited but PFPEs are nearly involatile) — marked in the CSV, the model constants, and a Model C footnote in the article.
README.md gains a Verification section documenting the above.
v1.43.71PATCH
Article 26 — GenX extract + the TFA-absence finding
Added
data/article-26/measurements-usgs-wqp-genx-hfpo-da.csv — 4,832 rows of real GenX
(HFPO-DA) monitoring results from the USGS Water Quality Portal since 2020. GenX is the
current-generation "safer" PFAS replacement, now itself EPA-regulated (10 ppt MCL) — the direct
analogue to the article's HFO→TFA replacement argument. Reproduce via fetch-usgs-wqp-genx.sh.
"The dataset that doesn't exist" finding box in the Data & Downloads section: querying the
USGS portal for trifluoroacetic acid (TFA) returns zero records under every name variant —
the breakdown product central to the article is unmonitored in US water-quality systems, a literal
confirmation of the "nobody is measuring it" thesis (this is a finding, not inline data).
Now ~35,000 rows of real measurements across four extracts; grid → 10 cards; README + intro updated.
v1.43.70PATCH
Article 26 — real USGS PFOS water-measurement extract
Added
data/article-26/measurements-usgs-wqp-pfos.csv — 9,230 rows of real PFOS monitoring
results from the USGS Water Quality Portal since 2020 (both controlled-vocabulary terms ORed:
"Perfluorooctanesulfonate" + "Perfluorooctane sulfonic acid"), across water/groundwater/sediment/
tissue. PFOS is the second compound under the EPA 4 ppt MCL. Public domain. Reproduce via
fetch-usgs-wqp-pfos.sh.
Now ~30,000 rows of real measurements across three extracts (atmospheric HFC-134a + aqueous
Water-quality Tech Spec sections — DC AI dry-only + DC Conv cooling-tower
PATCH ship: closes owner request "kualitas air masing-masing di dc ai dan dc conventional itu juga masukkan ya full standard parameter dan chemical dan consumption di tech spec."
Added
DC AI Tech Spec (datahallAI.html) — 7 new sections appended to Section 5
Cooling Discipline (after 5.10 Cooling Tower / Condenser Water Sizing):
5.11 Make-up Water Quality — 13 inlet quality parameters vs WHO
gw(path, opts) — fetch helper with credentials:include + CSRF header
loginV2(email, password) — POST /auth/login (cookie set by Worker)
logoutV2() — POST /auth/logout + clear local mirror
hydrateSessionFromWorker() — GET /auth/me on page load
Login modal + logout button + initial-load hydrate all guard if (AUTH_V2)
__rzAuth.getCsrf() public helper (Phase 4 admin UI consumes)
auth.min.js rebuilt with terser (--reserve loginV2,logoutV2,hydrateSessionFromWorker,gw)
worker-auth/test/client-auth-shape.test.mjs — 5 new tests pinning the Worker contract from the client's perspective (cookie shape, expiresAt seconds vs ms, CSRF lifecycle)
fuel-system / water-system / ict) — DC Conventional audit returned
full PASS; no edits needed in this ship.
GAP-6 — Feed-A red on Electrical SLD: already fixed in earlier
v1.20.2 Stage 6 (var CA='var(--b)' blue, Feed A title says
"FEED A (BLUE) / FEED B (GREEN)"). Audit was flagging a stale
reference; current code is correct.
Notes
UIUX audit findings (ict.html + datahall.html P0 redesign, IBM Plex
brand-token system-wide, EPMS_Telemetry mobile overflow) are
separate larger work — queued for v1.23.x with their own plan,
not bundled here (keep scope tight, one concern per ship).
v1.22.7PATCH
Featured Engineering Deep-Dive & Standards grouping — promotes the LTC Lab out of the buried bottom row
Changed
datacenter-solutions.html — new "Engineering Deep-Dive & Standards" featured section inserted directly above "Strategic Analysis & Market Intelligence" with two cards using the same .ds-strat-card bento pattern (gradient top-border, large icon, badge, feature bullets, gradient CTA):
Card 1 — Standards + Liquid-to-Chip Lab (amber gold theme, ROOT lock badge, links to standards-ltc-lab.html, keeps id="rootStandardsCard" + .root-only-card class so the existing amber-tinted lock styling carries over).
Card 2 — Liquid-to-Chip System Modelling Lab (cyan teal theme, links to ltc-system-modelling-lab.html).
standards-ltc-lab.html — lifted the "Liquid-to-Chip Engineering Lab" card out of the 6-sibling Standards Deep-Dive grid into a dedicated "Main Module" hero section above the standards grid; new self-contained .standards-hero CSS block (gradient top-border, 56 px icon, feature bullets, cyan CTA) with light + dark coverage + small-screen responsive collapse.
Removed
datacenter-solutions.html — buried duplicate ds-tool-row#rootStandardsCard row in the "Engineering & Compliance Tools" list (it lived just under "Pillar: Sustainability"). The LTC Lab entry-point is now featured up-page only — no duplication.
standards-ltc-lab.html — the LTC Lab card removed from the 6-card .standards-grid (5 standards-engine cards remain: ASHRAE / ANSI-TIA / ISO / NFPA / Uptime).
Notes
Reuses .ds-strategic-grid / .ds-strat-card / .ds-strat-card::before / .ds-strat-icon / .ds-strat-title / .ds-strat-subtitle / .ds-strat-desc / .ds-strat-features / .ds-strat-cta from the existing Strategic Analysis section — no new global stylesheet rules; the page's existing light + dark coverage applies automatically.
Auth gating unchanged: standards-ltc-lab.html is not in auth.jsROOT_ONLY_PATHS; the lock chip stays decorative (signals "root-only territory").
DC AI + DC Conventional pages untouched in this ship (under independent background-agent audit). Engineering-audit P1s (datahallAI Cooling P&ID ~1.18 / ~0.12 hardcodes + Math.random PUE badge, BMS lifecycle gaps, Feed-A red, MSB/UPS first-paint values) queued for v1.22.8.
v1.22.6PATCH
B-016 part 2: 390px horizontal-overflow fixed — B-016 COMPLETE
Fixed (CSS-only, additive, one idempotent <style id="b016-mobile-overflow-fix"> per page, ≤768px-scoped)
The 6 catalog placeholders (cat_summary_counts, cat_tbody@colspan=13, OEM tbody, facility-types panel, ca_blind_tbody, ca_sc_lane_tbody) only resolved when the user activated the Catalog tab. If the user landed elsewhere and never clicked Catalog, they stayed Loading forever.
Fix: at end of the catalog IIFE (line ~8204), added _eagerInitCatalog() that:
Fires on DOMContentLoaded (or immediately if already loaded)
If window.SPARES_CATALOG is available, calls catInitIfReady() + calcCatalogAnalytics() eagerly
Plus a _loadingFallback() 4-second setTimeout that scans for any Loading… text in catalog placeholders and replaces it with a graceful "Catalog data unavailable — refresh the page" message
This honours feedback_basic_feature_discipline.md rule #6 (Loading placeholders must always resolve).
See commit 73338de. 33 HTML pages had og:image pointing to non-existent files OR missing og:image:alt. Fixed in batch via NEW tools/fix-og-meta-tags.py. Audit went from 66 PASS / 33 FAIL → 99 PASS / 0 FAIL. Plus 99 NEW per-page OG images at assets/og/*.webp (1200×630 editorial cards).
v1.18.5PATCH
index.html — replace tacky marquee with lean credentials band
User: "running text ini jelek sekali, kurang lean, kurang professional look. norak"
The engineering-keyword marquee at index.html:423-454 was a 60s linear infinite
scroll with mint diamond bullets (◆), 3rem gap, 24 duplicated items, gradient
overlay background, dual borders, and edge-fade-out masks — every decoration
working against signal density.
Replaced with .rz-cred-band: static, dense, editorial credit line.
JetBrains Mono 10.5px (engineering numerics font)
Uppercase, letter-spacing: 0.08em (half the prior 0.16em → denser)
Pipe | separator (replaces ◆ diamond)
One hairline top border (no bottom, no gradient, no fade masks)
12 unique items (no duplication, no animation)
Left label CERTIFICATIONS · STANDARDS · OUTCOMES in muted signal-amber
Mobile: horizontal-scroll on overflow, scrollbar chrome hidden
User mandates this turn (verbatim per feedback_log_every_user_comment.md):
"agar saya tidak bolak2 balik request... di rz-ops admin. dan jangan hanya 2 free dan pro tapi buat 3. free (tanpa login apapun), demo (login account demo) dan pro root (pakai account bagus@xxx atau admin@xx) jadi saya lebih mudah atur di kondisi tertentu saya bisa disable atau enable per specific feature."
"padahal saya sudah minta ke memory anda jika setelah/sedang membuat suatu apps/calculator atau article atau apapun itu selalu buat folder as per nama nya dan draftkan md file untuk post draft di medium, x, linkedin, mastodon dsbnya seperti yang lain"
"termasuk ini /home/baguspermana7/rz-work/standarization/Indexing gconsole/top-urls-request-indexing.txt ini juga tidak diupdate, cek di standarization folder itu harus selalu diupdate."
Shipped (4 commits)
18b56ea Phase A — Feature flag foundation
NEW js/rz-feature-flags.js (315 lines): window.RZ_FEATURE_FLAGS schema for 14 page-keys × 6-13 flags each with {free, demo, pro} booleans; window._rzFeatures = { getTier, has, listFeatures, listPages }
NEW ~/.claude/agents/uiux-reviewer.md (impeccable design eye agent)
Reconciliation follow-up
Agent A's RZ_FEATURE_FLAGS (14 pages) and Agent B's RZ_FEATURE_FLAGS_FALLBACK (14 pages) have asymmetric sets. Union of both = 18 pages. Plan v1.18.1: align both schemas to the same 18-page set.
User context: "tidak perlu ada tailored message draft itu tidak penting, yang penting strategicnya bagaimana bisa come up" + "belum ada alur, flowchart, cards, jadi melihat engine spares-readiness-calculator.html jadi membingungkan alurnya" + "di awal2 kasih high level summary context"
Phase G — Stakeholder strategic refactor
Removed Tailored Message Drafts section from genStakeholder() output — replaced with 4 strategic outputs.
Added Influence × Impact 2×2 matrix (Manage Closely / Keep Satisfied / Keep Informed / Monitor) computed per stakeholder from their role and urgency level.
Added Strategy Heuristics card (8 Cialdini-based influence principles adapted for DC procurement context).
Changed button label from "Generate Plan" to "Build Strategy"; updated placeholder and ops-intro text to reflect strategic focus.
Phase H — Per-module flow cards
Added<div class="module-flow-card"> to all 9 analytical modules (criticality, readiness, stock, meio, hub, supplier, ltb, kraljic, montecarlo) showing Inputs → Computation → Outputs → Connects-To data flow.
Phase I — Top-of-page workflow flowchart
Added<details class="workflow-flowchart-wrap"> collapsible SVG flowchart (viewBox 1200×440) showing all 27 modules across 4 column groups: ANALYTICAL / OPERATING ENGINE / SUPPLY CHAIN / REFERENCE. Module labels are clickable (onclick="switchTab(...)") with tier-1 amber flow lines and tier-2 dashed cross-connections. Respects prefers-reduced-motion.
Phase J — Per-pane high-level summary cards
Added<div class="module-summary-card"> to all 9 analytical modules + catalog module. Each card has Q (what problem this solves) / A (method) / Output (what you get) / Use-when (trigger conditions).
.influence-matrix, .influence-quad, .iq-manage/.iq-satisfy/.iq-inform/.iq-monitor — 2×2 matrix grid for Stakeholder output.
All rules include [data-theme="dark"] overrides and @media (max-width: 768px) responsive behaviour.
## v1.18.0-prep — 2026-05-13 (Brand & design system foundation)
User-mandated work to escape "AI design slop" and establish identifiable brand character. Foundational artefacts created — visual changes to come in v1.18.x releases.
NEW~/.claude/agents/uiux-reviewer.md — local Claude Code agent with impeccable design eye. Enforces anti-pattern list (dot-grid noise, default Tailwind palettes, Anthropic-purple, saturated-emerald-everywhere, glassmorphism, neumorphism, cursor-3D-tilt, lifestyle stock photos, etc.). MUST BE USED on every UI commit going forward.
NEWdocumentation/design.md — comprehensive brand & design system manifest (target 2,500-3,500 lines). Covers: brand essence, visual character (industrial-instrumentation aesthetic), 30+ anti-patterns, typography (IBM Plex Sans + JetBrains Mono), color tokens (signal-amber #FFAA00, oscilloscope green #00FF88, fault-red #FF3030, instrument-cyan #00DDFF), layout patterns, kinetic patterns, iconography, component library map, 7 page archetypes with ASCII wireframes, PDF export design, accessibility (WCAG 2.2 AA), mobile responsiveness, 5-year roadmap (2026-2031), decision log. Authored async via sonnet agent.
NEW~/.claude/projects/-home-baguspermana7/memory/feedback_always_document_everything.md — codified user mandate: every code/content change MUST update CHANGELOG + standardization + relevant docs in the same commit. No exceptions.
User reported across multiple screenshots that "basic feature selalu bermasalah" (basic features always broken). Specific complaints + fixes:
Login button dead on spares-readiness-calculator.html. Root cause: auth.js detects inline .nav-login-btn and skips its own injection, but the inline button had no click handler. Fix: added onclick="if(window._rzAuth)_rzAuth.showModal();". Pattern now codified in feedback_basic_feature_discipline.md.
Active tab indicator invisible ("tidak ada indicative sedang active bisa ada warna kuning"). Root cause: line-724 CSS rule .tab-btn.active { border-bottom: 2px solid var(--amber-light) !important; } was the cascade winner because it came last + used !important, but it only set the underline (no background fill). The line-203 amber-fill rule was overridden. Fix: rule now uses filled amber background + amber border + 700 weight + ::after underline accent, all with !important to lock the cascade.
Negotiation tab horizontal overflow — .leverage-list rendered 12 long pills with white-space:nowrap in a flat row, expanding the pane beyond viewport. Fix: flex-wrap: wrap on the list + max-width:100%; overflow-x:hidden on .module-pane.active.
Negotiation output too thin ("level detail ini sangat2 kurang") — added ZOPA / Walk-Away table, weighted Decision Matrix (3 paths × 5 criteria), Risk Register (5 risks with prob/impact RAG), Role Allocation (5 roles), Cause-Effect Lever Map (5 levers + primary/secondary effects), Communication Cadence (5 time-buckets). Roughly tripled analytical depth.
Loading placeholders stuck ("loading2nya nggak berhenti") — diagnosed 9 Loading placeholders; resolution pattern documented in standardization. Implementation continues into v1.17.3 per orchestration.
Documentation discipline established this session (per user mandate "ingat di memory anda"):
Root cause identified: the v1.8.0 mobile-responsive patch tool injected a raw multi-line CSS block directly into JS html += '...' string literals inside PDF export functions. This created a JS syntax error (Invalid or unexpected token) that silently killed every function declaration in the script block, making all inline onclick= handlers throw ReferenceError.
cx-calculator.html — Fixed CSS injection (line 4125); fixed walk(ganttData) → walk(ganttData.items) bug in cxRenderGanttStats (Calculate button threw items.forEach is not a function); added 25 window exports. Probe: 0 errors, 0 missing, burger+back-link OK.
Three Puppeteer probes created: tools/probe-calc-tia942.mjs, tools/probe-calc-tieradvisor.mjs, tools/probe-calc-cx.mjs.
audit-script-tags.py --strict: CLEAN (149 files). audit-onclick-handlers.py --strict: CLEAN on all 3 pages.
v1.17.1PATCH
Spares Engine — stabilization #3: dead Generate buttons
User reported "Generate Proposal" button (and 8 sibling generators) silently dead on Operating-Engine tabs after v1.17.0 ship.
Root cause: inline handler pattern is onclick="safeGen(genX)" — it requires BOTH safeGen AND genX (the function REFERENCE passed as arg) to be on window. v1.16.2 exposed safeGen but missed the 9 generators. The v1.16.2 audit tool only checked direct call targets (onclick="X("), not identifiers passed as arguments.
9 gen* functions exposed on window — genPMOps, genNegotiation, genContract, genProcessImprovement, genMeetingPrep, genStakeholder, genEOLPlan, genAmbiguitySolver, genSTAR — all added to the export block in spares-readiness-calculator.html near line 9620.
tools/audit-onclick-handlers.py tightened — extract_handlers() now walks the entire event-handler expression (onclick="Y(X, Z)") and reports EVERY identifier, not just the call target. Skips JS built-ins.
Probe SUMMARY (live URL after push): consoleErrors=0, pageErrors=0, all 65+ inline-event handlers exposed.
v1.17.0MINOR
Spares Engine — MEIO optimizer
Added
New tab "4 · MEIO Optimizer" (pane-meio) in the Analytical group, inserted between Optimal Stock and Hub Positioning.
calcMEIO(): 2-echelon METRIC marginal-analysis solver (Sherbrooke 1968 + VARI-METRIC effective-LT expansion, Graves 1985). Iteratively allocates stock units between Regional Warehouse (s1) and Site (s2) to minimise total annual cost given a target fill rate and optional budget cap.
poissonBackorders(lambda, s): Poisson expected-backorder helper used by VARI-METRIC echelon-1 backorder expansion. Normal approximation kicks in for lambda > 200.
exportMEIOPDF(): minimal print-window PDF report for MEIO results.
Crosslink pills in pane-stock (→ MEIO) and pane-hub (→ MEIO).
TAB_ORDER updated to 27 entries; probe TAB_NAMES updated accordingly.
User reported "still many errors" after v1.16.2. Re-probed comprehensively with a deeper Puppeteer audit covering ALL inline event attributes (not just onclick). v1.16.2 only audited onclick and missed 19 functions called via oninput= / onchange= on input elements across 10 modules.
19 more handlers exposed on window — every input-bound recalc function across the Readiness, Stock, Hub, Supplier-Risk, LTB, Kraljic, Fleet, Scorecard, SC-Risk, Catalog-Analytics, Commodity-Defaults, Preset, MC-labels, Fleet-update, and SC-part-from-catalog handlers. Pattern: window.X = (typeof X !== 'undefined') ? X : null; for forward-compat in case any are deprecated later.
tools/audit-onclick-handlers.py updated — now extracts handler names from ALL inline event attributes: onclick, oninput, onchange, onkeyup, onkeydown, onfocus, onblur, onsubmit, onmouseover, onmouseout, ondblclick. Skips JS built-ins (if, for, Math, etc.) via blocklist. Strict mode for CI.
tools/audit-all-handlers.mjs — companion Puppeteer-driven cross-check that loads the live page and reports any handler typeof !== 'function' on window.
Deep probe coverage expanded — tools/probe-spares-deep.mjs now exercises Save/Load/Share, 15 calc functions across all module groups, tour Start → Next×2 → Skip, 189 tooltip elements, and mobile viewport at 375×667.
Spares Engine — stabilization: dead handlers, NaN cards
Runtime-verified stabilization of the 9,302-line calculator (Puppeteer probe green):
All 59 inline onclick handlers exposed on window (critical) — every function used in onclick="X(...)" was defined inside the main IIFE and unreachable from global scope, causing ReferenceError on every user interaction. Added a window.X = X export block before })(); // end IIFE. All 26 tabs now switch correctly.
NaN% on 4 criticality KPI cards (critical) — script.min.jsinitMetricCounters() selected ALL .metric-value elements via querySelectorAll('.metric-value') and wrote NaN% to any card lacking a data-target attribute (the calc engine's KPI cards). Fixed by changing the selector to .metric-value[data-target] so the counter animation only targets landing/article page stats. Rebuilt script.min.js.
Dead switchTab + TAB_ORDER (cleanup) — removed the 14-item TAB_ORDER and the stub switchTab function (declared at the top of the tab-switching section but immediately overridden by the 26-item version 2000+ lines later). Promoted the authoritative declarations with proper var / function syntax.
logoutPremium stub — the nav dropdown had onclick="logoutPremium()" with no definition anywhere; added a safe stub that delegates to window._rzAuth.logout().
tools/audit-onclick-handlers.py — new CI tool that enumerates inline onclick handler names and verifies each has a window.X = exposure. Exits 1 in --strict mode if any are missing. Passes clean on v1.16.2.
Probe SUMMARY (node tools/probe-spares.mjs): consoleErrors=0, pageErrors=0, tabsFailed=0, cardNaN=[], all windowExposure="function".
v1.16.1PATCH
Spares Engine — final QA pass: 7 fixes
Comprehensive code review of the 9,302-line calculator after its ~8 build passes. 7 surgical fixes, no regressions:
TAB_ORDER regression (critical) — the runtime-authoritative TAB_ORDER reassignment (line ~6996) was missing 'sc-lane','sc-risk','sc-sim','sc-expedite' (the v1.16.0 agent added them only to the first assignment), so keyboard Arrow/Home/End navigation + the mobile jump-to-module loop skipped all 4 Supply-Chain tabs. Added them — TAB_ORDER now lists all 26 module panes.
Patched switchTab calcs map missing the 4 SC handlers (critical) — opening a Supply-Chain tab via click/keyboard never triggered its calc on first open (stale/empty output). Added the 4 SC handlers to the patched map (they existed in the pre-patch switchTab but the later reassignment dropped them).
onTimeCnt double-increment in runSCSim — the disruption-sim loop set onTime = true then immediately if (onTime) onTimeCnt++ (always true) → the on-time-% was overstated. Fixed to a single increment.
Duplicate fleet-storage init — the Fleet list was loaded from localStorage twice on startup (two identical IIFEs); removed the second.
Stale chart registry on scsim_output/sc_expedite re-render — setHTML(...) wiped the <canvas> but left charts[...] pointing at a detached node → getOrCreateChart would .destroy() a stale object on the second run. Added explicit registry cleanup before re-injecting the canvas (both charts).
Dead variable demandForSites in calcHub — declared, never read; removed.
NaN guards on lane fields — added || 4 / || 9 / || 30 fallbacks on customsD/airD/oceanD in calcLanePlanner + calcExpedite so a catalog entry missing a field can't propagate NaN.
Confirmed already-correct: all ~26 tabs render, charts re-render on hidden→visible switch, SCENARIO_FIELDS covers the inputs, the 5 presets round-trip, math (Poisson lambdaLT = muAnnual × L, Monte-Carlo Box-Muller + percentiles + readinessRaw[] tornado, NPV picks lower-cost, EOL exposure, supplier-risk weights, hub-LT clamp, normInvCDF), safeGen(), per-module reset, all <\/script> escaped, dark mode, mobile (104/0 responsive), audit-script-tags --strict CLEAN.
Spares Engine: Global Supply Chain & Transport module group — 4 new tabs
Added — "Global Supply Chain & Transport" module group (deep-research-backed)
spares-readiness-calculator.html 7,575 → 9,303 lines (+1,728). Four new tabs driven by SPARES_CATALOG.transportModes (7), .tradeLanes (13), .countryRisk (16) — grounded in the 2026 DC-equipment-shortage research (Documents/Training/spares_supply_chain_transport_research.md):
🚢 Lane & Mode Planner — origin region → destination DC region + part (weight/value from catalog) + Incoterm + urgency → a mode-comparison table (ocean-FCL/LCL · air-standard/express · road · rail · courier — door-to-door days = mode transit + customs + last-mile, freight cost ≈ weight × base-$/kg × cost-index, CO₂ relative), cheapest-feasible vs fastest-feasible highlighted, a chokepoint-reroute what-if (+10-14 d for Suez↔Cape-style), the Incoterm 2020 cost/risk split (who pays export-clearance / main-carriage / import-duty / unloading / last-mile, where risk transfers) + the lane's tariff-exposure note (e.g. China's Section 122 10% + Section 301 + copper +50%), and a days-vs-$ trade-off chart. PDF + ⓘ box documenting the cost/day method.
🗺️ Supply-Chain Risk Map — a part (or the saved fleet) + origin + need-window + hub/consignment/VMI toggles → a composite 0-100 supply-chain risk score weighted across single-source exposure (scaled by # alternates), country-of-origin risk (countryRisk.geoRisk), lane congestion + geopolitical + rate-volatility, lead-time-vs-need-window pressure, tariff exposure, supplier OTIF/financial-health proxy, regional-hub coverage → band (LOW/MEDIUM/HIGH/CRITICAL), a radar of the dimensions, a ranked top-risks list, and recommended mitigations (dual-source / "China+1" / regional hub / consignment-VMI / last-time-buy / Incoterm change / FTZ-bonded-warehouse deferral / qualify substitute) each tagged effort × impact and ordered by impact-per-effort. Fleet mode → a per-part SC-risk table + fleet composite. PDF.
🌪️ Disruption Scenario / Resilience Sim — Monte-Carlo (≥1000 iterations, Box-Muller) over lane delay + σ, tariff-shock probability + magnitude, supplier-commit-slip probability + weeks, demand-spike probability + %, chokepoint-reroute probability → distribution of "% of critical-spares need met on time", P10/P50/P90 of (effective lead time, expedite-$, downtime-$), expected expedite-$ + downtime-$, a tornado of which disruption drives the most variance, and a with-vs-without comparison (regional hub / dual-source / +X weeks safety stock — Δ on-time-% and Δ expected-$). PDF + ⓘ box.
✈️ Logistics Cost & Expedite Calculator — site need-date vs supplier commit (the gap) + part weight/value + lane + downtime $/hr → a costed recovery-options menu: air-freight the critical sub-assembly + ocean the rest · full air-freight (standard or express) · partial shipment · ship from alternate plant (if alternates) · pull from regional hub/consignment/VMI (if a hub toggle) · qualify substitute (if no alternates) · accept downtime/escalate (the baseline) — each with $ + days-saved + closes-the-gap?, recommends the cost-minimizing path that closes the gap (or, if none does, "no option closes it — escalate to supplier exec + accept residual downtime; here's the least-bad partial"), a cost-vs-days-saved chart, and a "→ generate the supplier escalation email" link to the Daily-PM-Ops tab. PDF.
Changed — light touches
Catalog Analytics tab: an "🌐 Supply-chain exposure" panel — tradeLanes ranked by composite risk (congestion + geopolitical + volatility + tariff) + the China-transformer-dependency / 2026-tariff context.
Fleet / Portfolio tab: a "SC Risk" column per part (quick composite from the Risk Map logic) + a fleet-level supply-chain-risk KPI (table colspan 14→15).
Methodology footer note: added "Incoterms 2020 · multi-modal freight · World Bank LPI".
FAQ tab: +5 Q&As under a new "Supply Chain" filter (why transformer lead times are 2.5-5 yr in 2026 · the China-tariff exposure on DC M&E · when to air-freight a spare vs wait · what a "China+1" strategy is · how Incoterms split the duty burden) with citations.
js/spares-parts-catalog.js regenerated to expose transportModes/tradeLanes/countryRisk in window.SPARES_CATALOG (358 KB, 445 curated parts — structure otherwise unchanged); tools/build-spares-db.py's write_js_catalog updated accordingly.
Documents/Training/pm2_spares_sourcing_data_center_engine_prompt.md gained Appendix D — Global Supply Chain & Transport (the 2026 reality, transport-mode/Incoterm mechanics, the emergency-logistics recovery options, the mitigation playbook, the quantitative-companion cross-reference).
Wiring / verification
TAB_ORDER + switchTab calcs map + keyboard nav + mobile jump-to-module selector + SCENARIO_FIELDS + scenarioSnapshot/applySnapshot all updated for the 4 new tabs. IIFE closes exactly once; no duplicate fleet init (uses the _origCalcFleet/_origCatInit patch pattern). node --check OK, audit-script-tags --strict CLEAN, audit-mobile-responsive --strict 104/0, catalog file intact (445/7/13/16).
Wiring: TAB_ORDER + switchTab calcs map + recalcAll + SCENARIO_FIELDS + scenarioSnapshot/applySnapshot (serializes the fleet list as __fleetParts) + the jump-to-module selector all updated. Fixed a syntax error (spurious })(); // end catalog IIFE at EOF). Audits clean.
6 new tables + 2 views: sites (12 DC facilities), suppliers (distinct from OEMs — OTIF / commit-accuracy / quote-turnaround / PO-ack / defect-rate / responsiveness / corrective-action-closure / financial-health / capacity-headroom / geo & geopolitical & lead-time-volatility scores / strategic-importance / review-cadence / consignment & VMI capability), inventory_positions (on-hand / reserved / in-transit / safety-stock-target / reorder-point / max / days-of-cover by part × location), purchase_orders (full lifecycle — creation/ack/commit/need-by/received dates, delivery-status [on-track/at-risk/late/delivered/blocked/cancelled], blocker, recovery-plan, owner, demand-type), consumption_history (actual usage events → demand forecasting), engineering_changes (revision / supersession / EOL-notice / LTB-window / vendor-transition with qualification cost+lead-time + mitigation status); views v_po_at_risk (late/at-risk/blocked POs against critical parts with slip-days) + v_readiness_gap (critical parts where on-hand+in-transit < safety-stock target). build-spares-db.py --platform populates them with synthetic operational data (--scale 1: 12 sites · 122 suppliers · ~1,570 inventory positions · ~250 POs · ~10k consumption events · ~470 engineering changes). The default build (no --platform) leaves them empty so the committed catalog/CSVs are unchanged.
Added — Supply-chain & transport reference data (tools/spares-db-schema.sql — always populated)
3 new reference tables + 3 views, grounded in deep research (ICC Incoterms 2020; World Bank LPI; the 2026 DC-equipment shortage + tariff context — see Documents/Training/spares_supply_chain_transport_research.md):
transport_modes (7) — ocean-FCL / ocean-LCL / air-standard / air-express / road / rail / courier-express, each with intercontinental + intra-region transit days, a relative cost_index (ocean-FCL = 1.0; air-standard ~12, air-express ~25, road ~3, rail ~2, ocean-LCL ~1.6, courier ~18), a co2_index, capacity unit, typical use.
trade_lanes (13) — origin region → destination region (CN / EU / NA / SEA-Vietnam / India / Korea-Japan / MENA / LATAM / Intra-NA / Intra-EU / Intra-APAC) with ocean / air / road-rail transit days, customs-clearance days, last-mile days, and 1-10 scores for congestion / geopolitical / rate-volatility / tariff-exposure + reroute options + notes (e.g. CN-NA: 30 d ocean transit, congestion 6, geopolitical 7, tariff 8; Suez↔Cape +10-14 d; the China-transformer-dependency + Section 122/301 + copper-50% context).
Views: v_lane_lead_time (door-to-door days per lane × mode), v_high_risk_lanes (congestion + geopolitical + volatility composite), v_oem_country_exposure (parts/suppliers by country-of-origin × that country's risk → single-geography concentration).
Docs
Documents/Training/spares_engine_platform.md — the platform overview (the 5 layers, how they connect, the methodology grounding table, the v1.14→v1.16→beyond roadmap).
Documents/Training/spares_supply_chain_transport_research.md — deep-research synthesis: the 2026 DC-equipment shortage + tariff context, freight/port/lane risk, Incoterm/mode mechanics, the mitigation playbook (dual-source / "China+1" / regional hubs / consignment-VMI / component-specific safety stock / control tower / digital twin), with full citations + the design notes for the upcoming "Global Supply Chain & Transport" calculator module (Lane & Mode Planner · Supply-Chain Risk Map · Disruption Scenario Sim · Logistics Cost & Expedite Calculator — coming in v1.16).
MODULE_RESET_DEFAULTS.ltb had ltb_demand_yr: '0.4' / ltb_discount: '0' not matching the HTML input defaults → "↺ Reset defaults" on the Last-Time-Buy module set demand wrong + the discount rate to 0% (making the NPV stock-vs-requalify comparison always zero-benefit). Corrected to '1.2' / '8'.
Added a methodology/data-vintage footer note: "Models: FMECA (MIL-STD-1629A) · METRIC/VARI-METRIC (Sherbrooke/Slay) · newsvendor critical-fractile · Kraljic matrix (HBR 1983) · DMSMS lifecycle · MEIO. Catalog data vintage: 2026-Q1 · {N} curated parts · illustrative — not a substitute for a full supply-chain analysis." ({N} updates at runtime from SPARES_CATALOG.parts.length = 445.)
Confirmed (re-verified): all 21 tabs in TAB_ORDER, Poisson no double-count, hub-LT clamped, normInvCDF accurate, all <\/script> escaped, catUsePart() field IDs all match, dark mode complete, no leftover console.log.
Docs: Documents/Training/spares_parts_database.md (schema, sample queries, regen instructions, how it feeds the calculator, platform roadmap). Master-prompt doc pm2_spares_sourcing_data_center_engine_prompt.md gained Appendices A (methodologies referenced — FMECA/RCM/METRIC/Kraljic/DMSMS with formulas), B (calculator cross-reference), C (citations).
Added — Parts Catalog tab in the calculator
spares-readiness-calculator.html 5,639 → 6,249 lines, 21 modules. New "📚 Parts Catalog — Browse & Search" tab (Reference group): filter by system / DC generation / lifecycle / OEM / criticality ≥ / lead-time ≤ / free-text (150 ms debounce); sortable 13-column results table (capped at 150 visible, "showing X of Y" count) color-coded by lifecycle (active=green / nrnd=yellow / ltb=orange / obsolete=red) and EOL risk; "Use ▸" per row loads that part's attributes into the Criticality / Readiness / Optimal-Stock / LTB / Hub / Monte-Carlo modules + matches the commodity dropdown + shows a "Loaded from catalog: …" banner + recalcAll(); an OEMs sub-view (85-row table — name / HQ / market position / financial health / lead time / OTIF / single-source-risk color-coded / contract models); a DC facility-types sub-view (6 cards — era / IT-load range / PUE / cooling & power architecture / rack density / key equipment); CSV export of the filtered set; a light commodity-defaults hook ("{n} catalog parts — browse →"). Loads js/spares-parts-catalog.js?v=2026-05-12; ~115 lines of new dark-mode-aware CSS + mobile breakpoints.
Share-button overlap (your report) — the floating 5-circle column was position: fixed; z-index: 500 and on mobile sat at bottom: 60px of the viewport, intercepting taps on the "2·Readiness"/"3·Optimal Stock" tab buttons → fixed: #pageShare.share-buttons { display: none !important; } on ≤768px (footer has share/contact links).
Tab navigation hardened — try/catch around every calc/gen call in switchTab() / recalcAll() / on-input handlers / preset apply; montecarlo added to the auto-run map; safeGen() wrapper on all 9 operating-tab generators (catches uncaught exceptions, surfaces via showMsg() instead of silently failing).
134 per-input tooltips (your report) — ⓘ tooltip on every parameter across all 21 tabs — what it means + typical range + how it's used; lightweight data-tip + CSS popup, dark-mode-aware, keyboard-accessible (hover desktop / tap mobile + Enter/Space/Escape).
Criticality NaN cards (your screenshot) — RPN / Effective Severity / Fleet Exp. Failures/yr / Alternates Factor were showing NaN%; fixed to compute proper values with correct units (RPN integer; Eff. Severity X.X/10; Fleet Failures X.XX/yr; Alternates ×X.XX) + Number.isFinite() guards on every metric card site-wide + '—' fallback.
Poisson double-count (Module 3) — annualLambda = installedBase × muAnnual multiplied by installed base twice when muAnnual already is the fleet demand → 4× overestimate of stockout probability; fixed to lambdaLT = muAnnual × L.
Meaningless Monte-Carlo tornado correlations (Module 8) — tornado correlated arrays from independent simulation runs (different seeds) → random noise; fixed by capturing readinessRaw[] (insertion-order, before sorting) and using it for all three tornado correlations (also eliminated a duplicate simulation run).
Enhancements: 3 new Module-3 outputs (days of cover at Q, annual carrying $ at Q, expected stockouts/yr) — in both the per-module and full-report PDFs; per-module "↺ Reset defaults" buttons on all 8 analytical modules; chart axis unit labels (% on criticality, u on hub).
Changed — DCMOC app code review (dcmoc/)
Type-safety + numerical guards (commit 98c963c): typed nav LucideIcon; exported SimulationState/CapexStore interfaces + HeadcountKey union → eliminated all as any casts; useEffectiveInputs subscribes to s.inputs only (perf); Number.isFinite() guards on all 4 format.ts formatters; depreciation/PMT/ROI/PI/IRR div-0 + NaN guards in FinancialEngine/InvestmentEngine.
Error boundary + dead code (commit f5392af): new ErrorBoundary.tsx (class component, friendly fallback + retry) wrapping the dashboard area in page.tsx — a crashing dashboard no longer blanks the whole app; ReportDashboard hardcoded 2025 → new Date().getFullYear().
js/rz-version.js 1.13.0 → 1.14.0 (MINOR — new database + new calculator tab + new tooling).
SW cache name auto-synced → rz-cache-v1.14.0.
v1.13.0MINOR
Spares Engine: 4 more operating tabs — full 20-module engine · DCMOC pass 3
Added — Spares Engine: the last 4 draft modules (now 20 modules / 19 tabs + Summary Dashboard)
spares-readiness-calculator.html 4,107 → 4,997 lines. Completed the master-prompt draft coverage with 4 more deterministic template-generator tabs:
Stakeholder & Communication Planner (draft Module 9) — pick the 8 stakeholders involved + situation + urgency (Routine/Elevated/Urgent/Critical) → a stakeholder-map table (what they care about / communication style / recommended channel / cadence — keyed to urgency), per-stakeholder message drafts in the right register (executive = 3-line status+decision, supplier = specific ask+hard deadline+consequence, finance = cost+options, engineering = spec+decision), and a 3-level escalation ladder with trigger criteria (for Urgent/Critical).
EOL Response Plan (draft Module 11 — complements the LTB math tab) — inputs (part, notice date, installed units, sites, criticality, support years, failure rate, on-hand, open-PO, unit cost, alternates, alt-qual lead time + cost, redesign feasibility, carrying rate) → EOL summary, impact assessment (supply gap + single-source flag + rush warning), an options matrix filtered by input viability (Last-Time-Buy / Qualify Alternate / Redesign / Refurbished Pool / Do Nothing — each with Pros/Cons/When), LTB_Q = ceil(N × λ × yrs × 1.20 − onHand − openPO) (documented in the ⓘ box), a 6-step replacement-qualification plan with timeline, 6 supplier negotiation points, and a stakeholder-comms draft. Cross-links the dedicated LTB tab for the full NPV stock-vs-requalify comparison.
Ambiguity Solver (draft Module 14) — paste an undefined ask + who asked + apparent scope + key themes → 4-6 candidate interpretations (derived by matching the ask against a 13-signal supply-chain term map), a sharpened SMART problem statement, a 6-row hypothesis tree (Inventory / Supplier / Demand / Sourcing / Lifecycle / Process gaps — each with validation method + data needed), 8 clarifying questions tailored to the asker, a 10-item data-request list, a 30/60/90-day Discover → Stabilise → Systematise plan, and risks & assumptions.
Interview & Performance Story Builder (draft Module 15) — pick a competency (Ambiguity / Influence / Negotiation / Risk / Process / Crisis / Strategic / Data) + Situation/Task/Action/Result → a structured STAR narrative + "skills demonstrated", a competency-specific story scaffold, 3-5 likely behavioral interview questions, and coaching notes (competency-specific + 6 universal sourcing-PM interview principles). A "career companion" — clearly labeled.
Wiring: TAB_ORDER 15 → 19 module tabs; SCENARIO_FIELDS +17 input IDs (save/load/share-URL now covers all 20 modules); all 4 use the existing tab-btn-ops styling, .ops-output/.ops-table/.gen-text-block dark-mode classes, and the per-tab PDF pattern (<\/script> escaped); hero stat updated "9 modules" → "20 modules"; checkbox accent-color: var(--amber) in dark mode. FAQ +6 Q&As under "Operating Engine" (undefined-ask handling, 30/60/90 plan, exec-vs-supplier messaging, LTB-vs-requalify-vs-redesign, STAR-story structure, supplier-escalation).
Changed — DCMOC app refresh pass 3 (dcmoc/)
FaqDashboard (commit b23d44b) — 5 new Q&As (why PUE-median ~1.5, Tier-III/IV availability with exact Uptime values, how the wildfire risk factor works, the Capacity headroom analysis, 2026 tax incentives in the Investment module) + updated existing answers (33 countries, JLL/CBRE 2025 citations, 6-factor risk matrix, exact Tier-Standard values).
PDF exports (commit e24a4f0) — disclaimer footer on every page of all 11 generators ("Illustrative model — not a substitute for a full engineering or financial analysis. All figures in USD unless noted."), dynamic generation dates + projection base year via new Date().getFullYear() (no hardcoded 2025), CarbonPdf industry-PUE 1.58 → 1.50.
Dashboards (commit b6599b7) — Carbon tooltip 1.58 → 1.50, Simulation/Staffing dynamic years, + a dismissible "Data vintage: 2026-Q1 · benchmarks Uptime Institute 2025, JLL/CBRE 2025 · USD" banner in the Shell (localStorage-persisted, aria-labeled dismiss).
spares-readiness-calculator.html grew 2,384 → 4,107 lines. Beyond the 9 quantitative modules (v1.11.0), it now has 6 deterministic, copy-ready template generators that turn the day-to-day Program Manager workflow into structured outputs:
Daily PM Operating System — input today's situation (# late POs, supplier-not-confirmed count, critical shortages, severity sliders, free-text site/finance asks) → derives RED/YELLOW/GREEN situation status, a P1/P2/P3 priority stack, critical-follow-ups table, decision log, and an end-of-day status-email draft. Decision logic per the draft (critical spare + need-date <30 d → ≥High; supplier commit > need date → At Risk/Red; no alternate + critical → bump risk).
Contract / SOW Requirements Checklist — toggles (lead-time committed? forecast binding? capacity reserved? EOL notice months? LTB rights? change-notice timeline? consignment/VMI?) → a 15-area requirements table (Scope / Pricing / Lead Time / Forecast / Capacity / Delivery / Warranty / Quality / Documentation / EOL Notice / Last-Time-Buy / Change Notice / SLA / Inventory / Termination) with proposed-language concepts, flagged rows, and an open legal/procurement questions list.
Process Improvement Builder — describe a recurring problem + frequency + per-incident impact + affected stakeholders → problem statement with annualised impact, root-cause checklist, a future-state process keyed to the ticked causes, RACI matrix, KPI table, 30/60/90-day rollout plan.
Meeting Intelligence — Prep mode (meeting name/type/attendees/decision/risks → prep brief with the canonical agenda per meeting type) + Notes mode (structured decisions/actions/risks/open-questions/next-meeting template with add-row buttons).
Each tab: copy-to-clipboard + per-tab PDF export (<\/script> escaped), aria-labels, dark-mode-themed tables/cards, mobile-safe. 35 new input IDs added to the save/load scenario. FAQ 19 → 24 (+5 operating-engine Q&As) with a new "Operating Engine" filter button.
Poisson CDF overflow bug: e^{-λ}=0 underflow for λ > ~200 made P(stockout) return 0 for any stock level → added a normal-approximation fallback (CLT, continuity-corrected).
Inverted NPV decision bug in Last-Time-Buy: both options' NPVs are negative (costs); the code picked the more-negative NPV → recommended the more expensive path. Flipped + corrected chart highlight + documented the direction rule in the ⓘ box.
Added — SEO: 15 more per-page OG cards (commit 4f7d934, F18-01)
Generated 1200×630 WebP OG cards (~52-56 KB each) + patched og:image/twitter:image for: spares-readiness-calculator, chiller-plant, datahall, fire/fuel/water-system, ict, EPMS_Telemetry, asean-dc-report-2026, infographic-dc-cost-breakdown/-sustainability/-pue-global, achievements, insights, glossary. Only 4 utility pages now use the profile-photo fallback (404/dashboard/privacy/terms). tools/build-og-images.py TARGETS extended. (B2-001 double-<h1> audit flag confirmed a false positive — the 2nd <h1> on 41 pages is inside PDF print-window JS template strings, not the rendered DOM.)
New page spares-readiness-calculator.html (2,384 lines) — a comprehensive 9-module calculator for data-center mechanical & electrical (M&E) spare-parts management. Companion to the master-prompt operating doc in Documents/Training/.
User screenshot: rz-ops-p7x3k9m.html (admin console "Data Center Industry Intelligence") — all chart cards empty.
Root cause
v1.10.3 (commit 5c158f6, "perf: defer scripts") added defer to the chart.js CDN script tag on 22 pages. On pages whose inline <script> calls new Chart(...)synchronously during parsing (not inside a DOMContentLoaded/load listener and not behind a typeof Chart guard), Chart is undefined at that moment because the deferred chart.js hasn't executed yet → silent throw → every chart blank.
rz-ops-p7x3k9m.html runs if(checkAccess()){ ...renderDashboardCharts()... } at top level → all dashboard + benchmark charts dead. The bug shipped 2026-05-09, surfaced when the user opened the page.
Fix
Removed defer from the chart.js CDN tag (restoring blocking-load behavior, so Chart is defined before any inline script runs) on the 7 at-risk pages — those with deferred chart.js + no load listener + no typeof Chart guard:
The other 14 pages with deferred chart.js keep defer — they wrap chart init in a load listener or typeof Chart guard, so they work fine.
Trade-off
Blocking chart.js (~70 KB gzipped) costs ~100-300 ms of parse-block on those 7 pages — acceptable for correctness. Pages that already gate their chart init keep the perf win.
Audit-flagged E10-1: Google Analytics fired before GDPR consent on multiple pages. The eager-load <script async src="...gtag/js?id=..."> ran on every page load regardless of cookie banner state — sending pageview data before user could accept/decline.
Action
tools/gate-gtag-consent.py (NEW): walks every HTML, replaces eager gtag pattern with the canonical consent-aware deferred pattern.
63 pages migrated to the new pattern.
New pattern (consent-aware + interaction-deferred)
Default-deny: if rz_cookie_consent === 'declined', window['ga-disable-G-GED7FX8RTV'] = true is set BEFORE any gtag call.
Interaction-deferred: actual GA script only loads after user scroll/click/keydown/touch (not on idle pageview).
Disable-flag respected: even after interaction, the loader checks ga-disable-* and skips the network call if disabled.
gtag commands queued safely: queued before script loads; if disabled, never reach Google.
Impact
GDPR compliance improved: declined users never trigger GA at all.
First-visit users still queue gtag commands but the network call is delayed until interaction (faster FCP, +154 KB saved on bounce).
Cookie banner decline handler in sw.js-style code already sets the disable flag, now it sticks across reloads via localStorage check.
SW
SW cache name auto-synced 1.10.14 → 1.10.15 via tools/sync-sw-version.py.
2. BreadcrumbList schema (Home → DC Solutions → Standards Labs → System Modelling Lab).
Audit cleanup
E3-2 (113 target=_blank without noopener): pre-resolved — all 844 target=_blank links already have rel=noopener. The single remaining is in changelog prose (literal text in <code> block, not an active link).
D3-001 (broken jateng-diy link): pre-resolved — pln-java-grid-jateng.html exists and is correctly linked.
A1-FF-MODAL-01 (FF modal close handlers): pre-resolved — byId('hfxLoginClose').addEventListener wired on FF-1/2/3.
A2-SECONDBRAIN-01 (62 pages broken Apps/second brain link): now only 1 reference in changelog.html prose (legitimate documentation reference, not active nav).
SW
SW cache name auto-synced 1.10.13 → 1.10.14 via tools/sync-sw-version.py.
Bump 1.10.13 → 1.10.14 (PATCH — SEO + audit-driven cleanup).
v1.10.13PATCH
SW cache version-aware via tools/sync-sw-version.py
sw.js had a hardcoded CACHE_NAME = 'rz-cache-v8' that drifted from the actual site version. Manual bumps were forgotten across releases — meaning users on stale caches got mismatched JS+CSS+HTML for hours.
Action
sw.jsCACHE_NAME now reads rz-cache-v1.10.13 (matches site version exactly).
tools/sync-sw-version.py (NEW): reads js/rz-version.js, writes the matching CACHE_NAME to sw.js. Idempotent. Run after every version bump.
Comment in sw.js notes the auto-sync requirement so future maintainers know.
Workflow update
Per CLAUDE.md "Audit before push" section, add a step:
python3 tools/sync-sw-version.py # syncs CACHE_NAME to current RZ_VERSION
Impact
Service worker now invalidates its cache on every version bump → users always get fresh assets after a release.
No more stale cache after CSS/JS deploys.
Bump 1.10.12 → 1.10.13 (PATCH — SW hygiene).
v1.10.12PATCH
Cache-bust normalization across 96 pages
Audit found 8+ different cache-bust strings in active use for the same files (styles.min.css?v=20260324b, ?v=2026-05-09e, ?v=20260509-v1108, ?v=20260509-share-fix, etc.). Different bust strings = different URLs = browser caches the same file under multiple keys.
Action
tools/normalize-cache-bust.py walks every HTML file, normalizes ?v= on script/link tags pointing to: styles.min.css, styles-index.min.css, styles.css, styles-index.css, script.min.js, script.js, auth.js, rz-engine.js.
All normalized to single ?v=2026-05-09-v1 token.
Documentation prose (changelog mentions of old bust strings) is NOT touched — script only matches actual <script src=> and <link href=> tags.
Coverage
222 cache-bust strings normalized across 96 files.
Browser cache now uses single key per file → predictable cache invalidation on next bump.
Future
Next version bump should also bump the bust string (e.g., 2026-05-10-v1 for tomorrow's PATCH). Use this script to keep them in sync.
Bump 1.10.11 → 1.10.12 (PATCH — cache hygiene).
v1.10.11PATCH
Performance — extract 683 KB inline JS from LTC system modelling lab
ltc-system-modelling-lab.html was 914 KB total with 683 KB of inline JS in a single <script> block — blocking initial render and forcing the entire page to re-download every time the JS changed (no caching benefit).
Action
tools/extract-ltc-js.py extracted the 699,063-byte inline IIFE → js/ltc-system-modelling-lab.js.
HTML now references it via <script src="js/ltc-system-modelling-lab.js?v=2026-05-09" defer></script>.
Trailing inline <script> blocks (cookie banner + root auth gate) preserved unchanged — they don't depend on the extracted IIFE.
Audit-driven fix. 659 form inputs (<input>, <select>, <textarea>) lacked <label for=> AND aria-label — invisible to screen readers, fails WCAG 4.1.2 Name/Role/Value.
Action
Bulk script tools/fix-aria-labels.py walks every input with an id attribute, skips inputs that already have a linked <label for=> or aria-label, then injects aria-label derived from:
Input's placeholder attribute (if present), OR
Humanized version of id (camelCase → "Camel Case", abbreviation expansion: pue→PUE, capex→CAPEX, etc.)
Audit-driven cleanup. audit-reports/C-performance.md flagged assets/DC/ as 71 PNG files averaging 9-11 MB each (641 MB total). The original audit assumption (referenced from dc-conventional.html) was wrong — that page references assets/DC_Conventional.jpg (a different file). Zero HTML/JS/MD references the assets/DC/ folder.
User screenshots: "ini gambarnya stretch, need keep aspect ratio, ini juga cardnya saat 100% mobile view kok cardnya ke sisi kiri tidak fill (card area og image) dan card terms dll (akhir) dan card footer navbar tidak responsive full".
Root cause (3 issues):
.brief-hero-img mobile patch had object-fit: cover + max-height: 220px but no defined box-height → browsers couldn't crop properly, image rendered with squashed aspect ratio.
Mobile cards (.brief-card, .calc-disclaimer, .results-card, etc.) had inherited margin/padding from desktop rules — left-aligned with empty right gutter on narrow viewports.
<footer> + .footer-grid inherited fixed-width desktop padding → not full-width on mobile.
Fix
Aspect-ratio preservation: every .brief-hero-img variant now declares aspect-ratio: 1200 / 669; object-fit: cover; height: auto — locks the rendered box to the source image ratio. CSS aspect-ratio is supported in all modern browsers since 2021.
Card width-fill: explicit width: 100% !important; max-width: 100% !important; margin-left/right: 0 !important; box-sizing: border-box on every card class (.brief-card, .results-card, .input-section, .calc-disclaimer, .scenario-card, .model-card, .summary-card, .kpi-card, .tier-card, .feature-card, .terms-card, .info-card, plus prefixed variants).
Section wrappers: .brief-section, .results-section, .calc-section, .scenario-section get full-viewport-width with consistent 1rem padding.
User: "menu dc solution bisa expanded tapi nggak bisa di shrinked/di susutkan, saat mobile view".
Root cause: my v1.8.4-v1.8.5 mobile drawer CSS forced dropdowns to be max-height: 50vh; overflow: visible always — i.e., dropdowns expanded permanently when drawer opened. No way to collapse them. Once "DC Solutions" sub-items were visible, they stayed visible, cluttering the drawer.
Fix
js/rz-mobile-nav.js (cache-bust ?v=2026-05-09c):
Click handler intercepts taps on .nav-dropdown > a (dropdown trigger) inside the open drawer.
Toggles .is-mobile-open class on the parent <li class="nav-dropdown"> instead of navigating to the link.
300ms cubic-bezier ease for the height + opacity transition.
Sub-menu gets a left mint-accent border + indented background tint for visual hierarchy.
Replaces the existing SVG .dropdown-arrow with a ::after+ that rotates 45° to become × when expanded — clearer "tap to toggle" affordance on touch devices.
Plan v18 — Final dark-mode mandate for form widgets
User: "ini masih ada warna putih di calculator opex. astaga, saya bilang audit completely, fix all" (5th dark-mode regression flagged this session).
Root cause analysis
The Country/Region select on opex-calculator was rendering with white background despite [data-theme="dark"] .country-select { background: #1e293b !important } rule existing. Browser-level quirks (especially Firefox/Linux native <select> rendering) sometimes ignore CSS background on form widgets, even with appearance: none.
Fix — multi-layer dark-mode mandate (added to BOTH styles.css + styles-index.css + 7 calc page inline styles)
Layer 1 — color-scheme: dark on [data-theme="dark"] root tells browser native widgets to use dark chrome.
Browser <select> rendering ignores CSS background in some configurations even with appearance: none
Fix requires color-scheme: dark + forced-color-adjust: none
Include <option> element styling, not just the select
Inline style attribute selector defeats style="background: white" leaks
This was the 5th dark-mode regression in one session (v1.2.2 brief-card, v1.2.3 model-card, v1.4.1 input-field, v1.4.2 scenario-card, v1.10.7 select widget). Each had a different root cause but same symptom. The multi-layer mandate above defeats the entire class going forward.
Bump 1.10.6 → 1.10.7.
v1.10.6PATCH
Item 30 — per-page OG cards generated for ~50 more pages
Item 30 — Extended tools/build-og-images.py to auto-discover pages
Audit flagged 1 inbound link as "orphan" but all 5 pillar pages (cooling/power/standards/fire-safety/sustainability) ARE linked from datacenter-solutions.html (a high-traffic hub). Adding more random inbound links would be link-spam-y. SEO PageRank distribution is acceptable as-is. Documenting as resolved.
v1.10.4PATCH
Item 33 — CLS fix: inject width+height on 212 imgs
Item 33 — 208 <img> elements lacked explicit width + height attributes (primary cause of Cumulative Layout Shift / CLS spike on first paint, hurting Core Web Vitals).
Fix: Python helper walked all 76 HTML files, read intrinsic dimensions from local image files via Pillow, injected width="X" height="Y" attributes.
Remaining 49 are external CDN URLs / data: URIs (can't determine dims without HTTP fetch)
Pillow dimension cache: 71 unique local images analyzed
Impact: Browser can now reserve correct image space BEFORE the image loads, eliminating CLS jumps on every page that has <img>. Should improve Lighthouse CLS score significantly.
v1.10.3PATCH
Phase 4 perf batch — defer + minify rz-engine
Item 35 + 38 — Render-blocking script defer sweep
242 script tags across 107 pages gained defer attribute. Previously most were render-blocking.
Targets and counts:
auth.js: +108 defer attributes (was 86 unsafe — now 0)
rz-engine.js: +52 defer
rz-tracker.js: +60 defer
chart.js: +22 defer
rz-mobile-nav.js: already had defer
Item 38 — rz-engine.js minified
Created rz-engine.min.js via terser: 41 KB → 13 KB (-28 KB / -68%)
Switched 51 pages from rz-engine.js → rz-engine.min.js
Saves ~1.4 MB total bandwidth on first-page-loads across calc pages
Audit flagged capex + opex calc pages with 2× auth.js loads. Investigation: the second tag is INSIDE a PDF print-window template literal string (the <\/script> escape gave it away). Top-level DOM has only 1 tag. Print window needs its own script tags — intentional design. No fix needed.
#6b7280 on dark background measured 2.96:1 (WCAG AA requires 4.5:1).
Replaced with #94a3b8 (4.6:1 — passes AA).
327 occurrences replaced across 39 files.
styles.css + styles-index.css re-minified.
Item 43 — Tables without <th scope=>
Screen readers couldn't associate column headers with data cells on 75 files.
2421 <th> elements patched with scope="col" across 74 files.
Idempotent — <th> elements that already had scope= were skipped.
Item 44 — Skip-link sweep
49 pages had no skip-link to bypass nav for keyboard/screen-reader users.
42 pages got <a href="#main-content" class="skip-link">Skip to main content</a> injected after <body>.
15 pages that had skip-link but missing target got <a id="main-content" tabindex="-1"> anchor injected after </nav> (or after the skip-link itself if no nav).
Total skip-link-equipped pages: 49 → 91 (+42).
11 noindex pages correctly skipped, 5 fragments without <body> skipped.
Portrait Scenes 5+6+7 density + hamburger inline-style fallback
User screenshot: tiny "white dot" on capex-calculator mobile navbar that zooms when tapped. Confirms the hamburger button was rendering with no styling on calc pages — the spans inside collapsed to 0×0 dots.
Hamburger inline-style fallback (calc page fix)
js/rz-mobile-nav.js now applies INLINE STYLES on the injected hamburger button as a defensive fallback. Inline styles win over any CSS specificity collision on calc pages (which have their own navbar styling that may not include .rz-nav-burger rules).
Forced on every injected burger:
44×44 px button with 1 px mint border + 8 px border-radius
3 spans @ 20×2 px each, displayed as block flex children
All !important to win cascade
position: relative; z-index: 1001 so it sits above other nav items
This means the hamburger renders correctly on calc pages even if the page's CSS doesn't load .rz-nav-burger rules from styles.css.
Portrait video 2nd render — Scenes 5 + 6 + 7 all densified
This render picks up ALL the v6 source patches:
Scene 5 (Virtual Standards Labs): per-lab descriptions + 4 live audit metrics cards + 12-month compliance bar chart + 5 standards-body logos row. Vertical fill: 30% → 85%.
Scene 6 (DC AI vs Conventional): added stats sidebars filling the empty left ⅔ on each half (AI/HPC metrics top, Conventional metrics bottom) + architectural delta callout at the bottom (25× density, 0.35 PUE delta, 38% energy savings).
Scene 7 (Markets/Grid): added "Global Footprint" panel filling the 680 px empty middle — capacity utilization donut (Used 47% / Available 38% / Reserved 15%, total 2.4 GW) + 5×5 latency matrix (SG / TYO / LON / NV / DXB intercity ms) with color-coded heatmap.
Standards body logos row: 5 pulsing badges (ANSI, ASHRAE, ISO, NFPA, UPTIME) at bottom
Vertical fill: ~30% → ~85%.
Audio mux fix
-shortest was truncating 90s video to 60s (audio length). Replaced with apad,atrim=duration=90 filter complex so audio pads to 90s with silence and full video length is preserved.
Pending in v1.10.1 (next render)
Scene 6 (DC AI vs Conventional): left ⅔ empty fill — add stats sidebars + architectural delta callout
Scene 7 (Markets/Grid): empty middle fill — add capacity utilization donut + 5×5 latency matrix
These edits already in source (my-video/src/compositions/ResistanceZeroIntroPortrait.tsx); 2nd render already triggered in background.
Removed Sitemap: https://resistancezero.com/llms-full.txt directive — llms-full.txt is content not a sitemap; Google Search Console rejects non-XML sitemaps.
Item 29 — ai-content-declaration sweep
Tagged page count: 45 → 89 (+44). Helper walked all main HTML pages, skipped noindex (13) + pages with no description meta (6) + already-tagged (48), patched 55 new pages.
Item 30 (35 pages still using profile-photo as og:image) — extend tools/build-og-images.py TARGETS for ~70 articles + compares + pillars
Bump 1.9.2 → 1.9.3 (PATCH — Phase 2 SEO).
v1.9.2PATCH
Phase 1 broken-functionality fixes — items 9-20
Item 9+10 — subscribeNewsletter() unified to mailto: pattern
Added global window.subscribeNewsletter() to script.js: validates email, opens mailto:bagusdpermana7@gmail.com with pre-filled subject + body, shows inline confirmation message. No localStorage, no fake save.
Removed 18 per-page inline stubs (article-1 through article-17, FF-1/2/3, geopolitics-1/2/3) that used localStorage-only fake sign-up.
Articles 3, 9, 10, 14, 15, 19 (which had the form but no function) now work via the global.
Item 11 — exportToPDF() stub removed from article-10.html
Removed "Download PDF" button (was calling a stub that showed an alert() placeholder).
"Print Article" button (window.print()) remains as the working alternative.
Stub function definition also removed.
Item 12 — FF-1/2/3 modal close buttons (FALSE POSITIVE)
All three close buttons (#hfxLoginClose, #tgsLoginClose, #iecLoginClose) already have addEventListener('click', ...) wired correctly inside their IIFE. No change needed.
opmRegion and opmTier appear only once in the DOM (line 2364, 2377). The second occurrences are inside JS comments: // ── Region data (must match <select id="opmRegion">...). No duplicate IDs exist.
Item 18 — Skip-link targets added
404.html: Added id="main-content" to <div class="scene"> (the first post-nav content element).
datacenter-solutions.html: Added id="main-content" to <main class="main-content">.
Item 19 — _rzAuth null guards (ALREADY FIXED)
dashboard.html, dc-conventional.html, dc-market-tracker.html, datahallAI.html, datacenter-solutions.html: all _rzAuth.* calls already wrapped in if (window._rzAuth && typeof window._rzAuth.X === 'function') guards from a prior session. No change needed.
Item 20 — alert() → showToast() across 35 files
Added window.showToast() utility to script.js: non-blocking bottom toast, 3s auto-dismiss, dark glass style.
Replaced all alert(msg) calls with (window.showToast||alert)(msg) across 35 HTML files (~55 occurrences). Fallback to native alert for pages that don't load script.js (e.g. ltc-system-modelling-lab.html, calc pages).
prompt() and confirm() deferred to v1.9.1+ (need richer modal UI).
User: "Continue, audit total feature, cari celah error, bug terkait functionality atau area improvement. High and medium impact at least 500 item".
6-agent comprehensive audit — 759 items found (target 500)
Agent A (functionality): 157 items
Agent B (a11y): 119 items
Agent C (performance): 124 items
Agent D (SEO): 111 items
Agent E (mobile/consistency/security): 155 items
Agent F (tech debt): 93 items
All 6 reports + master aggregation in audit-reports/.
Top 50 fix candidates documented in MASTER-AUDIT-REPORT.md with phase roadmap (v1.9.0 → v2.0.0).
Phase 1 — Critical security/privacy fixes
localhost:8200 link removed from geopolitics.html:776 — replaced with dc-market-tracker.html.
target="_blank" rel sweep: 962 anchor tags across 96 files now have rel="noopener noreferrer" (was 113 unsafe — now 0).
"Second Brain" broken nav link removed from 67 pages (file path didn't exist anywhere).
Underscore-em markdown emphasis disabled in tools/build-changelog-html.py — was producing malformed target="<em>blank" because target="_blank" matched the underscore-em pattern. Disabled the underscore variant; *emphasis* still works.
Remotion v5 — fill empty space + complete DC Conventional + new VFX
User: "Tidak hanya ini, hampir semua screen remotion videonya kurang optimal penggunaan spacenya banyak ruang kosong... dc conventional kosong... Enhance more vfx dan visual nya".
Scene 6 — DC AI vs Conventional: Conventional bottom half now mirrors AI top half — full 3×2 rack grid with 9 thin server rows per rack, vent grilles, raised-floor scrolling stripe pattern, overhead cable tray, 2 animated CRAC units with rotating fan blades, sub-callout "Single feed · CRAC perimeter cooling", PUE 1.45 badge. AI top half gains liquid-cooling pipe particle flow + PUE 1.10 badge.
Scene 7 — Markets & Grid: Empty middle filled with NEW "LIVE CAPACITY FLOW" animated bar chart (10 bars, sinusoidal MW values, growth arrows, per-market colors) + running stats line "Global capacity: 2.4 GW · YoY growth: 18% · Avg PUE: 1.32". PLN chain compacted.
Scene 8 — DCMOC + Finance: Major compaction — KPI gap tightened, ROI gauge moved up (top:680→390), gauge radius 110→80. NEW NPV/IRR/Payback row ("$42.3M NPV · 22.7% IRR · 4.3 yrs"). NEW monthly OPEX trend mini line chart (12 months, gradient area). NEW live operations alert feed (3 rows with rotating active highlight).
New VFX layers:
GlitchTransitionvariant="vhs" — 30-frame extended glitch with stronger chromatic aberration (18px), 3 VHS horizontal distortion bands (yellow/teal/magenta), tracking noise bar, stronger CRT scanlines, corner vignette intensification. Applied at major scene boundaries (frames 1558, 1888, 2218).
AmbientParticles — seeded deterministic upward-drifting particle dots with sinusoidal drift + fade. Added on scenes 6/7/8.
index.html had TWO hamburger buttons (existing <button class="hamburger"> at line 344 + my new .rz-nav-burger).
calc pages appeared to have NO navbar (visual confusion).
Drawer couldn't scroll to see menu items below the fold.
Drawer wouldn't collapse properly in some cases.
Fixes
js/rz-mobile-nav.js — comprehensive rewrite:
Detect existing hamburger before injecting: .hamburger, .menu-toggle, [data-nav-toggle], .nav-toggle, .mobile-menu-btn, button.menuButton — if found, WIRE UP that button instead of double-injecting.
Mark wired buttons with .rz-nav-burger-bound class so CSS knows.
100dvh for modern mobile browsers (handles floating address bar).
z-index stacking: burger 1002, navbar 1003 when open — burger stays clickable above backdrop.
Smooth scrollbar styling inside drawer.
Cache-bust on js/rz-mobile-nav.js?v=2026-05-09b across 101 pages.
Bump 1.8.4 → 1.8.5 (PATCH — critical UX fix).
v1.8.4PATCH
CRITICAL FIX: mobile hamburger nav menu
User: "Critical bug, menu tidak keluar saat di klik button menu yg hamburger button in mobile view. Please audit properly, fix comprehensive".
Root cause: v1.8.0 mobile responsive sweep added .nav-menu, .nav-links { display: none; } on ≤768px to all 116 pages — but DID NOT add a hamburger toggle button. Mobile users had ZERO way to access the navigation menu after the v1.8.0 ship.
Fix
NEWjs/rz-mobile-nav.js (90 LOC, idempotent):
Injects a hamburger button into the navbar on every page
Toggles .rz-nav-open class on <body> to show full-screen drawer
Closes on link click + Esc + outside click + resize-to-desktop
Locks body scroll while menu is open
Hamburger animates to X on open
CSS in BOTH stylesheets (per CLAUDE.md 2-stylesheet rule — styles.css AND styles-index.css):
.rz-nav-burger button styling (44×44 mint-on-hover, 3-line icon → X morph)
Sitewide rollout: tools/inject-mobile-nav-script.py injected <script src="js/rz-mobile-nav.js" defer> on 116 pages, right after the existing js/rz-version.js script tag.
Cache-bust: styles-index.min.css?v=20260509-hamburger to force browsers to refetch the new CSS.
CLAUDE.md updated
Added "Mobile menu MUST have hamburger toggle" rule to prevent this regression class.
Bump 1.8.3 → 1.8.4 (PATCH — critical UX fix).
v1.8.3PATCH
CLAUDE.md project instructions + service worker v8
User: "All lesson learnt utk diupdate juga di claude.md agar tidak ulangi kesalahan yg sama atau serupa".
9 scenes: Electricity Awakens · DC Awakens · SLD · Calculators · Virtual Labs (NEW: 6 LTC standards labs in honeycomb) · DC AI vs Conventional (NEW: split-screen comparison) · Market & Grid Monitors (NEW: world map dots + PLN SLD) · DCMOC + Finance (NEW: 6-KPI dashboard + ROI gauge + 10-yr TCO chart) · Knowledge Graph + Finale
4 new VFX components: glitch-transition.tsx (RGB aberration + scan-line at 8 scene boundaries), holographic-grid.tsx (animated hex overlay), kinetic-text.tsx (spring-powered slide-in), lens-distortion.tsx (pincushion warp on finale)
Posters synced: agent generated intro-poster-landscape.webp + intro-poster-portrait.webp with new naming; renamed to canonical resistancezero-intro-poster.webp + resistancezero-intro-portrait-poster.webp so index.html modal works without further edits.
v1.8.0MINOR
Plan v15 Track A — mobile responsive sweep, partial
User: "Perbaiki responsiveness semua page ini contoh saat mobile, imagenya kekiri nggak auto adjust agar center page atau fill. Begitu juga card di bawah atau navbar footer itu. Dan navbar atas jadi tidak ada hilang semua... Audit semua page literally semua page. Deploy more agent to paralel audit total."
Mobile responsive patches applied across 60 pages in this commit (3 of 7 parallel agents have landed; remaining 4 ship in v1.8.1+):
Agent 1 — Calc pages (7)
pue/capex/opex/roi/tco/cx/carbon-footprint — patched with /* v1.8.0 — mobile responsive patch */. Each gains: body overflow-x guard, image responsive default, navbar mobile collapse, footer 3-col → single-col, KPI grid 2-col phone / 1-col tiny phone, breakdown-table horizontal scroll, mode-bar wrap, button stacking, tap targets ≥44px.
Lost traffic now lands on a beautiful branded page with clear navigation back to popular content (Engineering Journal, DC Solutions, CAPEX Calculator, etc.).
v1.7.2PATCH
Nav link: added Tools & Calculators to index.html Insights dropdown with mint accent + NEW badge. Changelog NEW badge moved to Tools (more recent ship).
IndexNow ping for v1.7.x: 7 URLs submitted (HTTP 200).
v1.7.1PATCH
public /tools.html hub page
NEW: /tools.html (591 lines, 38 KB) — public hub page listing all 18 calculators + tools across 4 categories:
Operator-Grade Simulations (2): Datahall AI BMS, DC Conventional Sim
Design: aurora mesh hero, gradient-shift "Tools & Calculators" H1, per-card accent color via --tool-accent CSS variable + shine-sweep on hover + 3-layer glow shadow.
SEO: full meta + Open Graph + Twitter Cards + CollectionPage JSON-LD with 18-item ItemList + BreadcrumbList.
Navigation: linked from articles.html Insights dropdown (between Changelog and All Insights).
Sitemap regen: 102 → 103 URLs (added tools.html).
llms.txt regen: 98 pages now indexable to AI search engines.
v1.7.0MINOR
Remotion v3 — landscape + portrait + auto-detect, plus title polish
Remotion video v3 — orientation-aware
NEW: assets/resistancezero-intro-portrait.mp4 — 60s 1080×1920 portrait composition (ResistanceZeroIntroPortrait). For mobile users where landscape would letterbox awkwardly.
UPDATED: assets/resistancezero-intro.mp4 — landscape (1920×1080) re-rendered with deeper VFX (higher glow strength, vignette, color grading, 12→16 frame transitions, more electricity callouts in Scene 3 SLD: ANSI relays 50/51 + 87T + 25 + 27/59 + 32 + 67, transformer Z=8% impedance, ΔT=5°C cooling annotation).
NEW posters: resistancezero-intro-poster.webp + resistancezero-intro-portrait-poster.webp.
JS auto-detect: openIntroVideo() now reads window.matchMedia('(max-width: 768px) and (orientation: portrait)') and swaps <video src> accordingly. Modal aspect-ratio also flips between 16:9 and 9:16.
Source elements: <source media="..."> tags as a CSS-only fallback if JS fails.
File sizes: 10.6 MB landscape + 10.3 MB portrait — both within hard cap.
SEO title polish
TIA-942 checklist: 69 → 47 chars (was the persistent SEO title-length WARN).
TCO calculator: 64 → 53 chars (in SEO sweet spot 30-60 now).
Bump 1.6.4 → 1.7.0 (MINOR — adds responsive video tier).
v1.6.4PATCH
small polish: humans.txt + TIA-942 title + author links
NEW: /humans.txt — web-tradition file at site root listing owner / certifications / tech stack / tooling / inspirations. Linked from index, articles, datacenter-solutions, changelog via <link rel="author" href="/humans.txt"> on those 4 pages.
Fix: tia-942-checklist.html title shortened from 66 → 56 chars (now in SEO sweet spot 30-60). Was the last audit-seo title-length WARN.
User: "saat video remotionnya kasi tombol x close button" (give the video an X close button).
Root cause: same class as the v1.4.1 share-button bug — the .video-modal-close CSS was in styles.css but index.html loads styles-index.min.css. The X close button rendered as a default browser button, easy to miss against the dark video.
Fix:
Copied the video-modal + overlay + close button rules into styles-index.css.
Enhanced the close button: 44×44 mint-bordered floating button positioned ABOVE the video frame (not overlapping native video controls), with backdrop blur, glow on hover, 90° rotate animation on hover.
Tap target: 48×48 on mobile (≤560 px width).
Portrait orientation modal: when device is portrait + ≤768 px wide, modal flips to 9:16 aspect ratio (420 px max width) — sets up for the upcoming portrait Remotion video.
Cache-bust: ?v=20260509-modal-fix.
v1.6.2PATCH
articles.html hub Awwwards uplift
Aurora mesh hero on .articles-hero (blue/mint/violet/gold/pink radial gradients drifting)
Gradient-shift H1 on "Operations Engineering Journal" (slate→blue→mint→slate sweep, 12s)
Article-card dark-mode override: was background: #fff (hardcoded white) — now rgba(30,41,59,0.6) + 1px white-mix border + 8px backdrop blur. Cards finally render properly in dark mode.
Article-card shine sweep on hover + 3-layer mint-glow shadow (matches index + datacenter-solutions pattern).
Current-version badge on the latest entry (mint pill in top-right).
GitHub commit hashes auto-linked to GitHub commit URLs (e.g., 5a0235c → live link).
Nav links added: index.html + articles.html Insights dropdown gain a Changelog item.
SEO meta complete: title, description, canonical, OG card (uses assets/og/index.webp), Twitter, JSON-LD WebPage + BreadcrumbList, ai-content-declaration.
Generator preserved at tools/build-changelog-html.py — re-run on every CHANGELOG.md update.
ai-content-declaration sweep on tool pages
Patched 6 more pages that audit-seo flagged: tia-942-checklist.html, tier-advisor.html, water-system.html, fire-system.html, fuel-system.html, ict.html. chiller-plant.html already had it (idempotent skip). Total tagged pages: 39 → 45.
Added: View Transitions API opt-in (@view-transition { navigation: auto; }) — supported browsers (Chrome 126+, Safari 18+, Edge) get smooth fade+slide transitions when navigating between pages on the site. Older browsers no-op gracefully.
Continuity: declared view-transition-name: rz-brand-mark on .nav-logo, .nav-avatar, .footer-logo, #rzVersionStamp img so the brand mark visually persists across navigation (one of the signature 2026 web feels — Apple, Vercel, Linear all use this).
Honours prefers-reduced-motion.
v1.5.2PATCH
FAQ + HowTo schema for AI search ranking
Added FAQPage schema (@type: FAQPage) to 5 calculator pages: pue / capex / opex / roi / tco. Each block has 3-4 Q&A pairs covering: how the metric is calculated, typical industry ranges, country/climate sensitivity, biggest input drivers. Surfaces in Google rich-results, Google AI Overview, ChatGPT Search, Perplexity.
Added HowTo schema (@type: HowTo) to tia-942-checklist.html (5-step audit workflow). tier-advisor.html + cx-calculator.html already had HowTo blocks (idempotent skip).
Each calc page now signals 4 schema types: WebApplication + HowTo + BreadcrumbList + FAQPage — a rich signal stack for AI search engine ranking.
29 JSON-LD blocks across 8 files validated cleanly (no syntax errors).
New tool: tools/inject-schema-faq-howto.py (idempotent, marker-gated).
Patched 12 HTML pages: replaced og:image + twitter:image to point at the new per-page WebP. Added og:image:width=1200 + og:image:height=630 where missing. dc-market-tracker.html gained its first-ever twitter:image.
RSS feed alternate link (sitemap.xml as feed source)
Bump 1.4.2 → 1.5.0 (MINOR — feature-class uplift across many pages + global polish).
v1.4.2PATCH
Proactive sweep: ran a comprehensive regex audit across all 7 calc pages for any class with hardcoded white/light backgrounds lacking a [data-theme="dark"] override. ONE remaining gap surfaced: .scenario-card on opex-calculator.html (line 947, background: white).
Fix: added 5 dark-mode rules covering .scenario-card base + .current active state + scenario-name / scenario-total / scenario-diff text colours. Active scenario card now shows a soft mint gradient instead of solid white.
Audit clean: all 7 calc pages now report CLEAN on the regex audit (every class with light bg has a corresponding dark override).
Inline style="background:#fffbeb" PDF-template callouts (10 in capex, 1-2 each in other pages) are intentional cream-accent info boxes used inside print-window templates — not user-visible in dark mode and correctly left alone.
The capex legacy #loginModal (hidden display:none, replaced by auth.js widget) intentionally untouched.
v1.4.1PATCH
Fix: .input-field selects + inputs were rendering with white backgrounds in dark mode on opex/capex/roi/pue/carbon-footprint. Root cause: class-mismatch — HTML uses <select class="input-field"> but the dark-mode CSS targeted page-prefixed classes (.opex-input / .capex-input etc.) that don't exist in the markup. Effectively the entire input dark-mode coverage was a no-op on 5 calc pages.
Pages affected: opex / capex / roi / pue / carbon-footprint. tco + cx were already correct (they use prefixed .tco-input-field + .cx-input-field consistently in HTML + CSS).
Fix scope: added [data-theme="dark"] .input-field + .country-select + option overrides + focus state to all 5 affected pages. Fields now render with slate (#1e293b) background, light text (#f1f5f9), and emerald focus glow.
Film grain noise overlay: body::before (dark mode) carries an SVG fractal-noise texture at 3.5% opacity with mix-blend-mode: overlay. Adds analog/cinematic depth.
Gradient-shift H1: .bento-name ("Bagus Dwi Permana") now uses background-clip:text with a 4-stop linear-gradient (slate→mint→gold→slate) and 12s sweep animation.
Card shine sweep: .bento-card::after carries a diagonal light streak that translates across on hover (0.9s cubic-bezier).
Card hover glow: replaces solid border with a 3-layer shadow (mint outline + dark depth + emerald aura).
Engineering keyword marquee: new <div class="rz-marquee"> strip below the identity row, scrolls 12 keywords (Hyperscale Operations, PUE 1.25, Tier III, N+1, SAP HV/LV, SCADA·BMS, CDFOM, Ahli K3 Listrik, ISO 50001, TIA-942, 99.999%, Mission-Critical) at 60s linear loop with edge fade-out gradient masks.
Scroll-reveal helper: .rz-reveal class + IntersectionObserver in inline <script> — fade-up on 10% viewport entry. Available for retroactive application on any element.
Result: index.html now feels like a 2026 dev portfolio (linear/vercel/resend territory) instead of a generic dark theme.
v1.3.1PATCH
Fix: chiller-plant.html — was missing canonical, all OG tags, all Twitter cards (audit-seo flagged as REQUIRED-tag errors). Added full meta-tag block + ai-content-declaration. Title bumped from 24 to 60 chars to fit SEO range.
Tooling: tools/audit-seo.py now correctly skips <meta name="robots" content="noindex..."> pages (LTC labs, redirects). Strict mode no longer false-positives on intentionally-internal pages.
IndexNow: synced .indexnow-key store to use the existing 2026-03 verification key (768683436ffdfcc2bb9140345660b139.txt) — Bing already verified this key, no need to register a new one.
Fix: dark-mode regression on opex-calculator.html — staffing-model cards (.model-card for In-House / Hybrid Mix / 100% Outsource) had hardcoded background: white (line 592) with no dark override. Unselected cards rendered as bright white blocks against the dark page. Added 8 [data-theme="dark"] .model-card* rules covering base, hover, active, name, desc, icon states. Audited other calc pages — only opex uses the .model-card pattern.
v1.2.2PATCH
Fix: dark-mode regression on opex-calculator.html + capex-calculator.html — the .brief-card hero intro block (the "OPEX is what actually kills the margin..." paragraph + stats row) was rendered with a transparent gradient rgba(16,185,129,0.04) over a dark page, making the entire intro card invisible on dark mode. The Plan v13 dark-mode agent missed the .brief-* class family because tco uses prefixed .tco-brief-* while opex/capex use unprefixed .brief-*. Added 9 dark-mode rules per page covering .brief-card, .brief-lead, .brief-body, .brief-stats, .brief-stat, .brief-stat-icon, .brief-disclaimer, .brief-hero-img. The card now has a visible accent-coloured gradient + border in dark mode.
v1.2.1PATCH
Fix: gridline pattern (linear-gradient 1px @ 50×50 px) was still present on datacenter-solutions.html — same noise that was killed on index.html in v1.1.1 had a sibling instance on the second-most-prominent landing page. Both [data-theme="dark"] .page-background (line 141) and base .page-background (line 256) now have only the soft radial washes, no grid.
Cross-page audit confirms 5 major landing pages are gridline-free: index.html, datacenter-solutions.html, articles.html, dc-market-tracker.html, future-forward.html.
v1.2.0MINOR
Plan v13 — Calc dark-mode audit
Fixed: opex-calculator.html — "Detailed Cost Breakdown" card (.breakdown-table) and "Category Comparison" chart card (.chart-card) showed WHITE backgrounds in dark mode. Added 35+ [data-theme="dark"] rules covering .breakdown-table th/td/hover, .chart-card, .results-card, .results-panel, .input-section, .breakdown-card, .kpi-card, .narrative-card, .calc-disclaimer, and mode-bar elements.
Added: cx-calculator.html — had ZERO dark-mode rules (was dark-only, no toggle). Added 45+ [data-theme="dark"] reinforcement rules + theme-init script + nav toggle button + toggleCalcTheme() function, making it consistent with other calc pages.
Standard: standarization/UI_FEATURES_STANDARD.md — appended Plan v13 dark-mode coverage mandate with pre-merge checklist.
Version: js/rz-version.js bumped 1.1.0 → 1.2.0.
v1.1.1PATCH
Fix: hero gridline pattern was still visible after Plan v12 ship — agent had patched only .hero-background::before but the base .hero-background rule (and dark-mode override) carried the actual grid via crossed linear-gradients @ 60×60 px. Now both light + dark hero backgrounds are fully transparent; only the ::before soft radial wash remains.
v1.1.0MINOR
Plan v12 shipped, commits 22548ba + c1667a4
Landing: removed rotated side tabs, replaced "↓ SCROLL TO EXPLORE" with Pixel Rise soft animation, added floating 5-icon share column (LinkedIn/X/WhatsApp/Instagram/Facebook), Get Started + Contact Us CTA pair in hero, navbar Contact link scroll-aware (hidden at top, fades in past hero), navbar transparent → frosted-glass on scroll.
Visual: removed dot-grid pattern from hero (clean ambient gradient now), pastel mint user pill replacing default purple, calm pastel bento card palette (mint/lavender/peach/pink/cream), GitHub label/URL removed from Contact and footer (kept in schema.org metadata).
Video: new Remotion intro composition ResistanceZeroIntro (30 s, 1920×1080), rendered to assets/resistancezero-intro.mp4. Plays in inline modal triggered by Get Started.
Site-wide: introduced js/rz-version.js as single-source-of-truth for version, RZ.injectVersionStamp() injects "Latest version: vX.Y.Z" stamp at every page footer.
Tooling: new tools/insert-version-script.py + tools/audit-version-stamp.py. New standarization/VERSIONING_STANDARD.md.
v1.0.0MAJOR
semver baseline
First semver-tagged release. This entry consolidates prior shipped work and establishes the versioning regime. From this point forward, every meaningful change MUST bump js/rz-version.js and append a CHANGELOG entry per standarization/VERSIONING_STANDARD.md.
Major shipped milestones (pre-baseline, abridged):
pln-java-grid.html — new dedicated detail page for the PLN Java-Bali (Jamali) transmission system. Geographic Map view (Leaflet/CARTO dark, Java + Bali fitBounds) and Single-Line Diagram view (inline SVG, IEC 60617 symbols, ~100 nodes target with "Show all 150 kV" toggle for the long tail). Province tabs (Jakarta+Banten / Jabar / Jateng+DIY / Jatim) with deep-link support (#prov=jabar). Substation slide-in side panel on click.
js/rz-map.js — new shared Leaflet wrapper engine. Public API window.RZMap.init(containerId, opts) returning { map, addMarker, addLine, setMarkerVisible, setLineVisible, fitBounds, setView, refresh, destroy }. Stations as circleMarker (color by voltage 500/275/150, radius √(MVA)*0.35). Plants as divIcon with FontAwesome glyph per fuel type. Polylines per voltage tier with rzm-line-{500|275|150} className for CSS dash-flow. Optional layer control on voltage/fuel toggles. prefers-reduced-motion guard. Resilient: no-ops if Leaflet isn't loaded.
js/pln-java-grid-data.js — data module for window.PLN_JAVA_GRID exposing { version, nodes[], edges[], national }. Topology source: PLN P2B 2016 single-line diagram. Coordinate confidence flag per node (high from Wikipedia infobox / OSM Nominatim, low from province-centroid fallback — none invented).
Changed
datacenter-solutions.html #pln-monitor section reverted to a 6-card grid (.ds-strat-card). Java-Bali card is active and links to pln-java-grid.html. Sumatera, Kalimantan, Sulawesi, Maluku-Papua, Nusa Tenggara cards render as dimmed is-soon placeholders (<div> not <a>, pointer-events:none, "Coming soon" pill instead of CTA — not crawlable as dead links).
dc-market-tracker.html refactored to consume RZMap.init() instead of its inline initLeafletMap() IIFE. Visual output identical.
standarization/UI_FEATURES_STANDARD.md: replaced the earlier "SLD Inline-SVG Animation Pattern" section with the broader "Card → detail-page hub + shared js/rz-map.js engine" pattern.
Removed
All .pln-* CSS rules from datacenter-solutions.html (~280 lines of SLD-only styling). Verified by grep -rln 'pln-grid-card\|pln-mini-stat\|pln-list-title' /home/baguspermana7/rz-work/ returning only the post-revert file itself.
Rationale
User feedback: SLD did not belong on the landing page; the hand-drawn SVG was inaccurate; the existing Leaflet/CARTO map from dc-market-tracker.html was the correct base; SLD detail target was "very detailed" (~100 nodes, not the prior ~25).
confidence distribution per voltage tier (flags >50% low)
province coverage (≥10 nodes per province)
Bali isolation (must have ≥1 edge crossing the strait)
cross-tier jumps (500↔20 without 150 kV intermediate)
Output as human-readable report or --json. --strict exits 1 on CRITICAL findings (CI-gate ready).
Current state: 0 CRITICAL, 38 HIGH (32 remote orphans, 1 statistical confidence skew, 5 cross-tier jumps from OSM lazy line tagging — all candidates for future YAML-overlay corrections).
Fixed the datahallAI.html "Root Access Required" modal that blocked logged-in PRO/root users. Root cause: race condition — gate IIFE ran before window._rzAuth was defined by auth.js. Patched the gate to fall back to a direct localStorage.rz_premium_session read with the same email-allowlist (admin@, bagus@), so the page works whether or not auth.js has loaded yet. Also added a storage event listener for cross-tab logout sync.
Fixed Java-Bali submarine topology in tools/pln-java-grid-overlay.yaml:
prov_override: bali on Cable Head Gilimanuk (osm_way_339796954) and GI Gilimanuk (osm_way_192989828) — both were OSM-tagged jatim despite being on the Bali side of the strait.
Replaced the wrong paiton → banyuwangi @ 275 kV curated edge with the actual physical reality: 4×150 kV submarine cables (~340 MW total, commissioned 1989-1996). The 275 kV submarine is planned but not commissioned.
Added curated Bali internal 150 kV ring (Gilimanuk → Negara → Antosari → Pemecutan → Pesanggaran → Pecatu, plus Sanur → Gianyar → Amlapura → Kubu → Celukan Bawang → back to Gilimanuk). 14 new edges fully connect the 40 Bali nodes (up from 38 — two were correctly retagged from jatim to bali).
UpdatedApps/second brain/index.html knowledge graph: added 5 new nodes (pjg, pjg-jkb, pjg-jb, pjg-jt, pjg-jm) and 11 edges connecting them to existing reports / DC Solutions / DC Markets hubs. Second-brain visualization now reflects the full Java-Bali grid family.
Edge total stable at 495 (52×500 / 0×275 / 418×150 / 25×70). 275 kV edge correctly dropped to reflect physical reality of the submarine link.
2026-04-30-v5 — Full province coverage + datahallAI cleanup + scheduled OSM refresh
Addedpln-java-grid-jateng.html (Jawa Tengah + DIY) and pln-java-grid-jatim.html (Jawa Timur). Pages mirror the v4-fixed Jakarta+Banten / Jabar template: default labels OFF, tier-graded thin lines, animation only ≥150 kV, hover tooltips, 5-tier voltage toggles. Java-Bali sub-page family is now 4/4 complete.
Addedjs/pln-java-grid-data-jateng.js and js/pln-java-grid-data-jatim.js — curated 20 kV DC + industrial overlays for each province.
Promoted Jawa Tengah + DIY and Jawa Timur cards on the overview page from is-soon placeholders to active links. All 4 province cards on pln-java-grid.html now click through to working sub-pages.
Removed the <section> with 10 academic-style references (NVIDIA, Uptime, Equinix, ASHRAE, OCP, Schneider, SemiAnalysis, IEA, Berkeley Lab, Lawrence Berkeley) from datahallAI.html. The page is a DC simulation tool, not a research article — citations were a category mismatch. datahall.html (DC conventional sibling) was already clean.
Sitemap: 2 new entries for the province pages, priority 0.85, monthly changefreq.
Scheduled quarterly OSM dataset refresh routine — python3 tools/build-osm-dataset.py --force runs on the 1st of each quarter; opens a PR if the dataset diff is non-trivial.
infer_edges_by_proximity in tools/build-osm-dataset.py connects any 500/275/150/70 kV station not already in an OSM or curated edge to its nearest same-voltage neighbour within 30 km (20 km for 70 kV). Source: inferred-nn.
infer_plant_evacuation connects each unattached plant to its nearest 500/275/150 kV substation within 5 km. Source: inferred-evacuation. Solves "plants float as isolated dots" issue.
Visual confidence: inferred edges render with opacity:0.35 + tighter dash + no animation (CSS [data-source^="inferred"] rule on all 3 pages). Curated/OSM edges remain bright with full laser-flow. Users can see at a glance which edges are factual vs. heuristic.
Labels default OFF on the SLD across all 3 pln-java-grid pages. With 744 nodes, drawing every name produced massive overlap. Names now appear only via hover tooltip. Labels toggle is preserved for users who want them.
Laser-flow animation locked to ≥150 kV only. 70 kV and 20 kV lines are static thin dashes (no animation property). Confirmed via CSS rule audit.
OSM line-endpoint matching threshold relaxed in tools/build-osm-dataset.py from 0.5 km to 1.5 km (bbox prefilter 0.01° → 0.03°).
Curated edges block added to tools/pln-java-grid-overlay.yaml — 28×500 kV backbone (Suralaya → Cilegon → Balaraja → Gandul → Bekasi → Cibatu → Cirata → … → Paiton plus radials + 275 kV Java-Bali submarine + key 150 kV corridors). Merged into the JS data file by the crawler with dedup against OSM. Edge total: 51 → 80 (28×500 / 1×275 / 47×150 / 4×70).
Crawler enhancement: load_overlay_edges(nodes) reads edges: block from YAML, fuzzy-matches from/to slugs against node names. Logs unresolved-endpoint warnings.
First-paint flicker prevented: SLD root group renders with class="*-svg-root no-labels" baked into the HTML (no JS race).
Why: user feedback after v3 deployment — "tulisan nama gardu sudah saya bilang jangan disini, tapi di tooltip" + "garis koneknnya kurang lengkap dan perlu yang tipis" + "arah flow laser itu hanya >=150kv saja" + "enhance banyak collision".
2026-04-29-v3 — Data accuracy expansion (OSM crawl + tooltip system + multi-tier toggles)
Addedtools/build-osm-dataset.py — Python OSM Overpass crawler for Java+Bali. Queries power=substation and power=plant/generator features, parses voltage tags, writes js/pln-java-grid-data.js with provenance fields per node (source, osm_id, wikidata, confidence).
Addedtools/pln-java-grid-overlay.yaml — hand-curated overlay (~60 entries) carrying mva, year, served_areas, notes for known substations and plants. Merged into the JS data file at build time.
Modifiedjs/rz-map.js (303 → 317 LOC) — now accepts per-marker tooltipData opt; auto-wires PLNTooltip.attach if module is loaded. Backward-compatible (existing dc-market-tracker.html consumer unaffected).
Modifiedpln-java-grid.html, pln-java-grid-jakarta-banten.html, pln-java-grid-jabar.html — added 5-tier voltage layer toggles (500/275/150 default ON; 70/20 default OFF on overview, 20 default ON on province pages). Per-fuel plant toggles. Display master toggles (Labels / Capacity / kV badges). Wired tooltips on every node + edge midpoint. SLD viewBox bumped to 1800×900 (overview) and 1400×900 (province) to absorb the larger dataset. Collision-nudge increased from 6 to 10 iterations with ±20 px search radius.
Schema additions per node: source, confidence (high/medium/low), osm_id, osm_type, wikidata, served_areas[], notes, secondary_voltages[], last_verified. Visible in tooltip header (kV + confidence badges) and footer (OSM/Wikidata/Map links).
Dataset growth: from 118 nodes hand-curated → 744 nodes OSM-sourced (563 stations + 181 plants), 6.3× expansion. Voltage breakdown: 33×500 kV / 1×275 kV / 442×150 kV / 55×70 kV / 213×20 kV. Province breakdown: jakarta-banten 213, jabar 196, jatim 185, jateng 112, bali 38. Confidence: 503 high / 224 medium / 17 low. User's specific concern resolved: GIS Summarecon now in dataset (osm_way_966209499, 150 kV, jakarta-banten, confidence:high) — alongside GIS Bekasi II, GISTET Tambun II, GI Tambun, GI Cikarang, GI Cikarang Lippo, KCIC Karawang, etc.
Why: user feedback on accuracy ("very accurate, very precise") and request that all voltage tiers be selectable. The user's specific complaint about GI Bekasi vs GI Summarecon is addressed via the served_areas annotation (Summarecon Bekasi, Harapan Indah, Logos Bekasi listed as served areas of GI Cibitung 150/20 kV).
Cards-on-landing → detail-page-on-click model matches the existing .ds-strat-card pattern used elsewhere in the section (TCO, ROI, DMT cards).
## [Unreleased]
Planned
Extract calc-auth.js shared engine (Phase 1 of calculator consolidation roadmap, see standarization/CALC_ENGINE_PLAN.md).
Phase S2.5 — expand RZEngine.models.{opex,capex,tco} API to support utilization-aware power, climate/cooling adjustments, multi-factor CAPEX build-up, and multi-stream TCO. Required before tco-/capex-/opex-calculator math can migrate to engine.
Hero images for articles 1–19 (currently missing assets/article-N-hero.webp).
References sections for articles 2, 4, 5, 6, 9, 10, 11, 12, 14, 15, 16, 17, 18, 19, 20 — older articles still missing canonical references-section markup; some have legacy <ol class="references"> and could be migrated to canonical pattern in a separate sweep (articles 21, 22 done 2026-04-30).
Tighten Independence Disclaimer placement in articles 19–27 (currently inserted before </main>; older convention is before References — cosmetic only).
Reconcile auth.js vs rz-engine.jsVALID_USERS role strings (auth.js: demo='pro', bagus/admin='root'; rz-engine.js: demo='demo', bagus/admin='admin'). Email-based gate makes drift safe but harmonization remains hygiene work.
auth.js — added isRootEmail(email), isRootAccess(session), isRootSession() helpers exposed on window._rzAuth.*. Email-based check uses pre-existing ROOT_EMAILS = ['admin@resistancezero.com', 'bagus@resistancezero.com'] and is robust to the role-string drift between auth.js and rz-engine.jsVALID_USERS lists.
auth.jsROOT_ONLY_PATHS — extended from ['/dcmoc'] to ['/dcmoc', '/dc-market', '/datahallai.html', '/dc-conventional.html', '/dc-market-tracker.html']. Auto-applies the navbar 🔒 lock icon (fas fa-lock rz-lock-icon) to all matching links across the 60+ pages with the dropdown — no per-page HTML edits needed for the lock visualization. Click handler enforces root-account gate via existing handleRootOnlyLinkClick.
dc-conventional.html — full root-only gate added (CSS body.locked blur + .rz-restricted-overlay modal + IIFE that subscribes to rz-auth-change and toggles body.locked based on _rzAuth.isRootSession()). Page was previously unguarded; demo and anonymous now blocked.
dc-market-tracker.html — same gate pattern (CSS + overlay + IIFE). Pre-existing hub card linking to dc-market/ retained (PLN session added it on 2026-04-29).
/home/baguspermana7/.claude/projects/-home-baguspermana7/memory/feedback_simulation_pages_no_refs.md — new memory feedback rule: never add <section> References blocks to simulation/dashboard pages (datahallAI.html, dc-conventional.html, future BMS/SCADA-style mimics). Trigger: 2026-04-29's discoverability sweep mistakenly added one to datahallAI.html; reverted on 2026-04-30 commit df0fbd7.
Changed
datahallAI.html gate — replaced minified ia(s){return!!(s&&(s.role==='root'||s.role==='pro'));} IIFE (lines 9768-9779) with _rzAuth.isRootSession()-based check. Previous version allowed role==='pro' to pass — under auth.js's VALID_USERS, the demo account had role:'pro', so demo bypassed the gate. New version uses email-based ROOT_EMAILS check and rejects demo while admitting only bagus@ / admin@.
roi-calculator.htmlcalcNPV and calcIRR — both now delegate to RZEngine.models.roi.npv / RZEngine.models.roi.irr when the engine is available, falling back to inline math otherwise. Pattern matches pue-calculator.html S2 pilot. Engine smoke verified: npv([-100, 30×5], 0.10) = 13.7236 matches inline; IRR via engine bisection = 0.1524 for the same series.
DC Market dropdown consolidation — across 66 HTML pages (articles.html, glossary.html, dashboard.html, insights.html, index.html, all article-N.html 1-27, all compare-*.html, all geopolitics-*.html, all pillar-*.html, all ltc-*.html, all infographic-*.html, all FF-*.html, future-forward.html, achievements.html, asean-dc-report-2026.html, tco-calculator.html), the navbar dropdown's "Market Tracker" label was renamed to "DC Market" via tools/dc-market-consolidator.py. index.html additionally had its sibling "DC Markets (10 cities)" line consolidated into the single "DC Market" item — that secondary link is now reachable via the in-page hub card on dc-market-tracker.html instead. Locked icon auto-renders because dc-market-tracker.html is in ROOT_ONLY_PATHS.
Fixed
roi-calculator.html JavaScript SyntaxError (lines 1780-1782) — the printPDF function had a single-quoted string literal that spanned three lines without \ continuations or template-literal backticks, causing the entire IIFE containing calculate(), calcNPV(), calcIRR(), attemptLogin(), handlePremiumTab() to fail to parse. Every JS-dependent feature on the calculator was silently broken in browsers (curl returned HTTP 200 because HTML still served). Fixed by splitting the broken multi-line string into three concatenated html += '...'; statements with <\/script> escape sequences.
capex-calculator.html and opex-calculator.html Login button no-response — <script src="auth.js"> and <script src="rz-engine.js"> tags were trapped INSIDE the printHTML template literal (lines 4028-4029 and 4613-4614 respectively), so they only loaded inside the PDF print window, never on the calculator page itself. Result: _rzAuth.* and RZEngine.auth.* were undefined on the calculator page → login modal flow silently failed. Fixed by adding real top-level <script> tags before </body>. The script tags inside printHTML stay (they're correct for the PDF output).
roi-calculator.html script tags — same issue (top-level tags missing); added before </body>.
Reason: 2026-04-29 commits 72b81ce feat(capex,opex,cx-calculator): migrate to RZEngine.auth and af8875c feat(roi+tco-calculator): migrate to RZEngine.auth mistakenly placed the migration's script tags inside the PDF print template literals on capex/opex/roi calculators. tco-calculator.html, cx-calculator.html, and pue-calculator.html were correctly wired (top-level tags before </body>) and weren't affected.
Status: Super Engine consumers (delta vs 2026-04-28j)
0 remaining "Market Tracker" labels in nav dropdowns (66 → "DC Market"); 1 remaining standalone reference is the <h1> page title on dc-market-tracker.html itself, which is intentional (page is still the global Market Tracker dashboard).
Forged-session DevTools resistance: setting rz_premium_session with {email:'demo@…', role:'root', tier:'pro'} keeps the gate locked — email-based check rejects forged role strings.
Rationale
Email-based root gate chosen over role-based to neutralize the role-string drift between auth.js (role:'pro' for demo) and rz-engine.js (role:'demo' for demo). Whichever file writes the session wins; email is stable. ROOT_EMAILS already exists at auth.js:20, matching the working dcmoc gate convention.
DC Market consolidation keeps dc-market-tracker.html as the global Leaflet/Chart parent ("DC Market") with the 10-city deep-dive hub reached via in-page card linking to dc-market/. Single navbar item replaces the previous two-line "Market Tracker" + "DC Markets (10 cities)" pattern. User intent: "DC Market itu parentnya, tambahkan menu di page itu atau card untuk menuju /dc-market/".
No References on simulation pages — operational dashboards (datahallAI's 4-tab BMS mimic, dc-conventional's facility infographic) take a "Legal Notice" disclaimer instead of academic citations. New memory rule prevents future discoverability sweeps from re-adding them.
article-26.html PFAS calculator IIFE migrated from inline VALID_USERS array + bespoke session check to RZEngine.auth.validateLogin, RZEngine.auth.getSession, RZEngine.auth.setSession, RZEngine.auth.dispatchAuthChange. Inline VALID_USERS declaration removed entirely. Legacy fallback retained for safety if engine fails to load.
<script src="rz-engine.js?v=2026-04-28"> wired into 30 additional pages (articles 1–22 + articles.html + 5 standalone calcs + dashboard adjacents). Total rz-engine.js consumers across the site now: 35 pages. Most don't yet consume the engine API but are now set up for future migration without another script-tag pass.
Status: Super Engine consumers
Article
Loads engine
Uses auth.*
Uses models.*
Uses data.*
article-23
✅
—
—
—
article-24
✅
✅
—
—
article-25
✅
—
—
—
article-26
✅
✅
—
—
article-27
✅
✅ (S2 pilot)
✅ workforce.*
✅ regions, salaryBenchmarks, attritionFactors
article-1 through article-22, articles.html, +standalone calcs
✅ (script tag only)
—
—
—
## [2026-04-28i] — Standalone calc nav glossary link
Added
Glossary link (#14b8a6 teal) inserted into the .nav-links custom navbars on 12 standalone calc/tool pages:
capex-calculator, opex-calculator, roi-calculator, tco-calculator, pue-calculator (5 main calcs)
ltc-system-modelling-lab, standards-ltc-lab (2 LTC labs — used .nav-back style for these)
datacenter-solutions (1 solutions hub)
This closes the standalone-calculator nav backlog from [Unreleased] (2026-04-28g). Glossary is now reachable from every page on the site that has any kind of navbar — main-pattern (.nav-menu), custom (.nav-links), or LTC-lab (.nav-back).
Status
The discoverability audit is now functionally complete:
✅ Glossary linked from every page with a navbar (~77 pages total).
✅ Glossary linked from footer NAVIGATION across 60 pages.
✅ All Tier-1 and Tier-2 report pages have References sections.
✅ insights.html surfaces the Reports cluster.
✅ Second Brain graph reflects current site truth.
Remaining backlog (small)
Article-26 PFAS IIFE migration to RZEngine.auth.* (currently kept as A/B control).
dashboard.html and datacenter-solutions.html References — optional, these are tool pages.
References sections on all 10 dc-market/*.html city pages (~6 region-specific citations each, 60 citations total). Each uses authoritative regional sources:
Singapore: IMDA, EMA, NEA, CBRE APAC, JLL Asia, IEA.
Jakarta: Kominfo, PLN, BPS, JLL Indonesia, CBRE Indonesia, Asia Cloud Computing Association.
Kuala Lumpur: MyDigital, MCMC, TNB, JLL/Cushman/EPU Malaysia.
Frankfurt: Bundesnetzagentur, BMWK, DENA, JLL/CBRE Germany, eco Association.
Dubai: TDRA, DEWA, RTA, JLL/Cushman MENA, UAE Ministry of Energy.
Mumbai: TRAI, CEA, MAHADISCOM, JLL/CBRE India, NIXI.
Northern Virginia: Dominion Energy IRP, FERC, NERC, PJM, Loudoun County EDA, JLL Mid-Atlantic.
References sections on all 3 infographic pages (~6 citations each, 18 citations total):
PUE Global: IRENA, Uptime, IEA, LBNL, ASHRAE, Green Grid.
DC Sustainability: IEA, AWS, Google, Microsoft, Greenpeace, CDP.
DC Cost Breakdown: CBRE, JLL, Uptime, NVIDIA, OCP, Schneider.
<script src="rz-engine.js"> wired into article-23.html, article-25.html (joining article-24, article-26, article-27 as Super Engine consumers — 5 of 27 articles now load the engine).
Status of discoverability audit
✅ All Tier-1 (high-traffic report pages) have References.
✅ All Tier-2 (10 city pages + 3 infographics) have References.
✅ Glossary navigation in navbar + footer across 65 pages.
✅ Reports & Trackers cluster surfaces all reports from insights.html.
✅ Second Brain graph: 0-edge nodes (CX, Glossary) connected; stale labels fixed; RZEngine + 3 plan docs added.
Remaining
dashboard.html and datacenter-solutions.html References — these are tool pages, references optional.
~29 standalone calculator pages with .nav-links (custom navbar pattern) still need glossary link addition. Separate audit.
IIFE migration of article-26's PFAS calculator to RZEngine.auth.* (kept as A/B control through the v1.2.0 ship; can migrate now since the engine is stable).
Glossary navigation surfaces: glossary link in navbar Insights dropdown across 65 HTML pages (color #14b8a6) and in the footer NAVIGATION column across 60 HTML pages.
References sections for the three highest-traffic report pages:
dc-market-tracker.html — 10 citations (CBRE 2025 Global DC Trends, JLL 2025, Cushman & Wakefield 2025, Synergy Research 2024, Uptime 2024, IEA 2024, McKinsey, BloombergNEF, Data Center Frontier, government / utility filings).
asean-dc-report-2026.html — 10 citations (CBRE APAC, JLL Asia Outlook, Synergy, IMDA Singapore, Kominfo Indonesia, MyDigital Malaysia, DEPA Thailand, Cushman, IEA, Uptime APAC). This page was previously orphaned with zero inbound visible links — now linked from insights.html.
Reports & Trackers cluster on insights.html — 6 cards surfacing dc-market-tracker, asean-dc-report-2026, datahallAI, and the 3 infographics. Closes the inbound-link gap.
Second Brain graph new nodes: a27 (Article 27 Workforce Crisis), rzeng (RZEngine v1.2.0), sse (SUPER_ENGINE.md), scep (CALC_ENGINE_PLAN.md), scmp (CALC_MODELS_PLAN.md).
Fixed
Second Brain graph CX Calculator (ccx) was 0-connection — now linked to dash, sdcv, copx, croi, rzeng (5 edges).
Second Brain graph Glossary (glos) was 0-connection — now linked to idx, arts, ins, articles 23-27, calculators with terms (cpue, cpp, cpa), rzeng (12 edges).
Second Brain graph stale labels:a24 was "FF-1: The Web Didn't Die" → now "Art-24: Manpower Shortage". a25 was "FF-2: Engineer Shortage" → now "Art-25: PJM 6 GW Short". Both moved out of Future Forward tagging into their actual content categories.
Unreleased follow-ups (logged for next session)
References sections for the 10 dc-market/*.html city pages (~5 region-specific refs each).
References sections for infographic-pue-global.html, infographic-dc-sustainability.html, infographic-dc-cost-breakdown.html.
References sections for dashboard.html and datacenter-solutions.html.
Glossary link insertion for the ~29 standalone calculator pages with .nav-links (custom navbar pattern, separate audit).
## [2026-04-28f] — Super Engine S4 + S5 + S6 (capex/opex/tco/pue math + UI primitives)
Added
RZEngine.data.capexPerMw — per-MW build cost baselines for airCooledTier2/3/4, liquidCooledTier3, immersionTier3 (sources: 451 Research 2024, JLL DC OpCost 2024, Cushman & Wakefield 2024).
RZEngine.data.mepPctOfCapex — MEP percentage by tier (36/42/48% for T2/T3/T4).
RZEngine.data.modularPremiumPct — modular vs stick-built premium by tier.
ui.badge, ui.kpiCard, ui.gateOverlay, ui.glossaryAnchor all return well-formed HTML strings.
Status
All 4 math domains (workforce / capex / opex / tco / pue / roi / forecast) and core UI primitives now live in the engine. Phases S0–S2, S4, S5, S6 of SUPER_ENGINE.md are SHIPPED (S3 PDF consolidation deferred to remote agent on 2026-05-05).
## [2026-04-28e] — Super Engine S2 (workforce + ROI + forecast models) + modal helper
Added
RZEngine.models.workforce — annualHiresRequired, attritionCost, strategyFitScore, cumulativeHires, yearsToCloseGap. Closed-form math, defaults pulled from RZEngine.data.attritionFactors so a single benchmark refresh propagates to every workforce calculator.
RZEngine.modal.create({id, title, accentColor, subtitle, bodyHTML, submitLabel, onSubmit}) — singleton modal helper. Auto-injects backdrop with rgba(0,0,0,0.85) + backdrop-filter:blur(8px) per PRO_MODE standard. Returns {show, hide, destroy} controls. Reuses existing element on repeat calls (idempotent).
Engine bumped to v1.1.0.
Changed
article-27 IIFE now calls RZEngine.models.workforce.attritionCost(...) and RZEngine.models.workforce.annualHiresRequired(...) for the corresponding KPIs (with hardcoded fallbacks if engine missing). This is the first calculator on the site to share math via the engine, not just constants.
RZEngine.events.{dispatch, on, off} — generic CustomEvent bus.
Stubs for RZEngine.{models, modal, pdf, charts, ui} filled in S2–S6.
Script tag added to article-27.html (after auth.js, before script.min.js) and article-26.html (after auth.js).
Changed
article-27 pilot (S0 first consumer):
wsCheckSession now delegates to RZEngine.auth.getSession() with legacy fallback.
REGION_MULT and REGION_LABEL derived from RZEngine.data.regions at IIFE init (with hardcoded fallback if engine missing).
avgSalary baseline pulled from RZEngine.data.salaryBenchmarks.dcTechMid.US ($75,100, was hardcoded $72,000 — refresh to 2024 BLS / Uptime number).
replacementFactor pulled from RZEngine.data.attritionFactors.replacementCostMult (213%).
Constants are now editable in ONE place (rz-engine.js) and propagate to article-27. Future migrations move article-26 + standalone calculators to the same engine in subsequent phases.
article-27 + article-26 modal backdrop now rgba(0,0,0,0.85) + backdrop-filter:blur(8px) (was rgba(0,0,0,0.7) no blur — caused article body to bleed through behind the Pro Analysis modal).
article-27 IIFE now listens for rz-auth-change event so navbar login propagates to the embedded calculator without a page reload. Also fixed a session-format mismatch: IIFE was reading sess.exp (numeric timestamp) while auth.js writes sess.expires (ISOString) — IIFE now accepts both formats. Local IIFE login now writes the auth.js-compatible format and emits rz-auth-change so the navbar reflects the login state immediately. (Article-26 already had this listener; only the modal fix applied there.)
Added
standarization/SUPER_ENGINE.md — master architectural design unifying CALC_ENGINE_PLAN.md (plumbing) and CALC_MODELS_PLAN.md (math) under a single window.RZEngine.* API. Documents the "Shared Anchor Parameters" rule: even when a new calculator is custom-built, parameters like Target Year, Region, Currency, Inflation, salary benchmarks, attrition factors, PUE defaults, and power costs MUST be sourced from RZEngine.data rather than inlined. Includes 6-phase rollout (~10–11 weeks), versioning discipline, consumer template, DCMOC relationship, failure modes, and 5 open questions for review before S0 starts.
Cross-references: CALC_ENGINE_PLAN.md and CALC_MODELS_PLAN.md now declare SUPER_ENGINE.md as their parent vision.
article-27 dark-mode group-header badges (CREATE/SUBSTITUTE/EXTEND) now have [data-theme="dark"] overrides; they were the empty-rectangle badges visible at the top of each strategy group in earlier dark-mode screenshots.
Changed
article-27 calculator expanded from 8 → 12 inputs and 7 → 10 KPIs.
New inputs: Target Year (2025–2030), Region (US/EU/APAC/LATAM), Workforce Mix (Physical-heavy/Balanced/NOC-heavy), Risk Tolerance (Conservative/Balanced/Aggressive).
New KPIs: Annual Hires Required, Cumulative Hires by [Target Year], Years to Close Gap.
Cost-related KPIs now scale by region multiplier (US 1.00 / EU 0.85 / APAC 0.45 / LATAM 0.55).
5-Year Investment renamed to N-Year Investment, length driven by Target Year.
Narrative auto-references Target Year, Region, Workforce Mix, and Risk Tolerance.
article-27 added a 5th Pro panel: Year-by-Year Hiring Trajectory chart (multi-line: Remaining Staff Gap, Cumulative Hires, Strategy Capacity with maturity ramp).
article-27 PDF export now includes the new KPIs and Target Year/Region in the header.
article-27 in-prose first occurrences of AIOps, NOCaaS, and apprenticeship now link to glossary.html#term-[slug] per the new glossary workflow.
Added
standarization/CALC_MODELS_PLAN.md — sibling roadmap to CALC_ENGINE_PLAN.md covering the calculation math layer (CalcModels.{workforce, capex, opex, roi, tco, pue, forecast} plus CalcModels.data for shared constants like salary benchmarks, region multipliers, attrition factors). 4-phase rollout. Closes user concern about scattered math without a "big engine" for shared parameters.
Cross-reference between CALC_ENGINE_PLAN.md and CALC_MODELS_PLAN.md.