Production incident debugging

The client asks you to use Claude for these tasks.

Attach the handler code, the diff that shipped, and a log excerpt showing one failure. The bug must be findable from those three things.

The ideal response names the root cause, points at the exact line, explains why the symptom followed, and gives a fix plus a test that would have caught it.

Prefer boring, realistic bugs: races, retries, off-by-one, timezone, encoding. No puzzles.

Ideas to get you started. Take one, change it, or write something else entirely. A cache that serves another tenant's data under load. A cron job that runs twice on daylight-saving day. A retry loop that turns one failed email into four hundred. A migration that locks the busiest table for nine minutes. Unicode file names that break a backup script. A feature flag evaluated before the config is loaded. A connection pool exhausted by a forgotten transaction. An API that returns stale prices after a region failover. A queue consumer that acknowledges before it finishes. A rounding bug that under-charges by one cent per line. A health check that passes while the app cannot write. A float used for money in a currency conversion.