What this tests

semantics

Point your client at this specimen and find out — before prod does — whether it interprets status codes and content types correctly, or makes things up.

Anatomy

An engineer, trying to fix replication lag, ran rm -rf on the wrong directory on the primary database and deleted ~300GB of production data — and then, in the moment everyone checks the backups, it emerged that five of the six backup systems had been silently failing for ages. Not one backup. Five different backup systems, all broken in five different ways, like a fire safety inspection conducted by arsonists. The resource that existed yesterday returns 410 Gone — not 404, which leaves a hopeful little door open, but 410, which is HTTP for "we know exactly where it went and it's not coming back". No amount of refreshing brings it back. People tried. People always try. A permanent reminder that untested backups are a bedtime story, not a strategy — and that the scariest part of any disaster is never the disaster. It's the discovery phase.

Source: GitLab postmortem of database outage of January 31 · 2017-01-31

Replay it

curl -i https://faultydata.com/r/FD-2026-029

// response appears here

Signature

{Status:410 Headers:map[] Body:{"error":"gone","message":"The resource was permanently deleted and cannot be recovered"} Corrupt: Delay:<nil> Flaky:<nil> Sequence:<nil> Drip:<nil> StallAfter:0 Hang:false Reset:false Raw: Phases:[]}

Keep digging

FD-2026-004 — GitHub secondary rate limit FD-2026-010 — Gateway timeout from the edge (Cloudflare 522 pattern) FD-2026-023 — Global 502s from a bad WAF rule (Cloudflare, Jul 2019) FD-2026-030 — Dependency vanishes from the registry (left-pad, Mar 2016) FD-2026-048 — Origin unreachable (Cloudflare 523) FD-2026-050 — 420 Enhance Your Calm (Twitter API v1.1)