What was wrong, and how I knew
Short debugging stories. Each one is a time the obvious explanation was wrong.
mcaPhysicsChainFour theories, one identical number
Four different fixes to the bake all returned the same error. That sameness was the clue: the error wasn't coming from the bake. Case study →
mcaPoseSkinnerThe warning that was there all along
After two wrong crash fixes, I shipped debug symbols and reproduced outside Maya in 20 seconds. A compiler warning on every build said the call I'd credited did nothing. Case study →
mcaMeshTilerThe test that certified the bug
Output that couldn't be animated, and a test asserting exactly that, written backwards. The bug passed its own test. Case study →
mcaCageWrapEvery synthetic test passed
The mesh moved half as far down the centre line. A 0/0 term in the published integral, only visible in real float32 numbers out of Maya. Case study →
Mirror calibrationThe eighth wrong answer wasn't the mirror
Seven wrong sign-flip answers, each killed by a geometric check. The eighth was a blend slider left at 61%. Case study →
If this were broken in the most likely way, would this check still pass?