Case files from the field. Each one is a real bug with a real fix: how we found it, the diff, whether the tests actually discriminate, and what tripped us up along the way.
An idempotency race with a merged fix still had two real gaps: a lost-race error surfacing as a raw 500, and responses outliving their claims. Reading acceptance criteria against shipped code pays.
PATCHED No.001A "$780 bounty program" with no way to get paid and README instructions written at AI agents. How we spotted it in thirty seconds, plus the one-line fix for the real bug inside.
The tune-up applies the same evidence standard to your endpoints. Flat fee, 72 hours.