What this tests

authretry-logic

Point your client at this specimen and find out — before prod does — whether it handles auth failure without panic-retrying into a second incident.

Anatomy

Azure AD rotated a signing key into an environment that didn't support it, and every token signed with it started failing validation planet-wide — users with perfectly healthy refresh flows met 401 invalid_signature, re-authenticated like good citizens, got a fresh token signed by the same bad key, and were bounced again. The door is locked. You get a new key. The new key is cut wrong at the factory. Repeat until morale improves. The fix, per the postmortem: rotate back, then forward again, properly this time — a procedure that works on paper and in production takes four hours because nothing involving keys ever happens at the speed you want at 1am. Your client sees an honest 401 with a WWW-Authenticate header that actually explains itself — invalid_token, signature invalid — which is more than most outages bother with. The trap is responding to it by hammering the token endpoint, which is how you turn an identity outage into two incidents.

Source: Microsoft preliminary RCA — Azure AD outage March 2021 · 2021-03-15

Replay it

curl -i https://faultydata.com/r/FD-2026-027

// response appears here

Signature

{Status:401 Headers:map[WWW-Authenticate:Bearer error="invalid_token", error_description="The signature is invalid"] Body:{"error":"invalid_client","error_description":"AADSTS90002: Invalid signature in the access token."} Corrupt: Delay:<nil> Flaky:<nil> Sequence:<nil> Drip:<nil> StallAfter:0 Hang:false Reset:false Raw: Phases:[]}

Keep digging

FD-2026-005 — Token expires mid-session (Entra ID pattern) FD-2026-026 — Global authentication outage (Google, Dec 2020) FD-2026-001 — Stripe-style rate limit FD-2026-004 — GitHub secondary rate limit FD-2026-009 — Provider overloaded (OpenAI-style 529) FD-2026-032 — Capacity-collapse 503s (Healthcare.gov launch, Oct 2013)