BENCHMARK.md §5. Machine: Apple M2, arm64, macOS (Darwin 24.2.0). Toolchain: Apple clang 17.0.0, node v26.5.0, python 3.14.6.
This section read 0 for two weeks after it stopped being 0, which is the exact failure the rest of this page exists to guard against: a number kept by hand drifts away from the ledger it claims to summarise, and nobody goes back for the one that flatters nobody.
The two are on expressjs/express @ a3714473, its own 1,260-test mocha suite in the judge's seat, all 91 of its test files hash-locked, the working tree never edited. Raw events and both patches: bench/evidence/2026-08-01-g3-first-held-repair/. In the same run a proposal that reached for describe.skip was refused as test-tamper and the counter did not move.
Two things this number is not. It is not breakage found in the wild — both bugs were planted to drive the loop end to end, and the number that matters next is the one on breakage nobody planted. And it is not the 46 a raw ledger scan shows: 44 of those came from a single scripted pipe inside four minutes, which is what native/drill.h rule 5 was written to catch and now does.