Skip to content

Releases

v0.2.0 — benchmark-verified frozen suite

Choose a tag to compare

@aidless aidless released this 10 Nov 10:36

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