overview / benchmarks / correctness
Until fbd6f6e, native/postuse_test.sh piped identical hook JSON into both the Node gate and the native gate and asserted the same verdict (exit code), state.json and spool; 13 of its cases were explicit node-vs-native differentials (9 differential() cases, 3 BR2 asserts, 1 BR11 twin-dedupe), 13/13 identical. When the Node gate was retired, those answers were pinned in native/fixtures/postuse_golden.jsonl, recorded while parity held for every case: the suite now compares the native gate, event for event, against them, and corrupting one pinned answer turns its case red.
BENCHMARK.md §2 · marked proven. Machine: Apple M2, arm64, macOS (Darwin 24.2.0). Toolchain: Apple clang 17.0.0, node v26.5.0, python 3.14.6.
Suite totals, measured 2026-09-25:
| suite | result | what it compares |
|---|---|---|
postuse_test.sh | 75 ok, 0 fail | native against the pinned Node answers |
pushgate_test.sh | 11 ok, 0 fail | native-only push-gate proof (invokes node 0 times) |
pushgate_test.sh is a native-only push-gate proof, not a parity comparison — grep -cE 'node ' native/pushgate_test.sh = 0. Its passing cases are native-behavior proofs (a hanging suite is killed and the push blocked, etc.).
Commands: RABADON_NOTIFY=0 ./native/postuse_test.sh → gate postuse: 75 ok, 0 fail RABADON_NOTIFY=0 ./native/pushgate_test.sh → gate push: 11 ok, 0 fail