Repository navigation
Releases
Releases · aidless/eval-gate
v0.2.0 — benchmark-verified frozen suite
eval-gate v0.2.0 — benchmark-verified release
Citable anchor: docs/BENCHMARK.md
- Frozen demo suite
eval-gate demo: 10/10 scenarios match frozen expectations - Quickstart: 4/4 in 0.49s (bootstrap=2000, seed=20261103, Python 3.12.3)
- Headline: p2_regression 87.92% → 80.63% (delta −7.29pp, 95% CI [−0.0875, −0.0583], n=480) →
BLOCK - Verdict ladder:
BLOCK > INCOMPLETE > FAIL > REVIEW > PASS; deleted evidence yields INCOMPLETE, never PASS - Scope: frozen synthetic scenarios exercising decision rules; no model calls, no network, zero required deps