Field note /
Routing change that passed its tests and was still wrong
A routing change in an open-source proxy passed tests written under the same assumption as the code. An independent review found the defect and the change was reverted.
How the finding changed
Belief
Narrowing a routing exception to a smaller pool of candidates was a safe tightening, and tests written around that behavior supported it.
Evidence
The code and its tests agreed on the narrowed pool.
Contradiction
An independent reviewer followed a caller path the tests did not cover and showed that removing the healthy fallback meant the pool no longer took a load spike. A separate monitoring investigation later found that internal warm-up traffic had inflated a request counter that was being read as client activity.
Correction
Revert the narrowing, add follow-up test cases, check delegated status claims against raw GitHub state, and define monitoring counters by where the traffic came from.
Current claim boundary
The correction shows the value of reviewers whose assumptions differ from the author's, and of careful monitoring. The change was never deployed, so no claim about production effectiveness is supported.