Data-Center Incidents — Case Library

What this is

A root-gated, source-cited engineering reference on major data-center and cloud-infrastructure incidents worldwide — each a structured post-incident dossier (sequence of events, root-cause analysis, correction-of-errors, lessons learnt, engineering improvements). It exists for internal engineering education; every material fact is traced to a public post-incident report, and where an authoritative cause was never published the dossier says so rather than invent one.

How incidents are ranked

By a transparent magnitude composite — a weighted sum of four sourced sub-scores (each 0–10): blast radius 0.35 · users 0.25 · financial 0.20 · duration 0.20. Hover any magnitude bar or a dossier's radar for the exact breakdown. Ranking never uses a hidden score.

What “official RCA” means

The official-RCA filter and the count above mark incidents backed by a genuine vendor post-incident review, regulator/government report, or court record — not press reporting. The provenance gate refuses to set that flag unless such a source is actually cited.

Reading the visualizations

• Where these happened — a real map at each incident's origin site; gold = facility failures (power/cooling/fire), green = network/logical. Co-located incidents are fanned onto a small ring; hover a marker for the brief.
• Ranked bars — magnitude, coloured by severity (red = most severe).
• Risk map — blast radius (x) × outage duration (y); top-right is the worst quadrant. Dots sit at their true scores; hover any dot for its values.
• Semantic map — a projection of the research vector-index: incidents near each other share a failure signature.
• Per dossier — a failure-cascade block diagram, a magnitude radar, and a phased SOE timeline; hover elements for detail, and hover underlined terms for definitions.

Access

Root-only. The library is excluded from the public sitemap and search index.