contract-review-eval-harness

A deterministic, offline harness that scores AI contract reviews against hand-authored answer sets. These pages are rendered from the evidence committed in the repository, which CI regenerates and verifies on every change.

• Scorecard — The deterministic stub reviewing the synthetic NDA, scored and gated.
• Release certificate — The release policy's decision across every bundled case, with its failure register.
• Live run, 2026-09-01 — One captured model run, re-scored offline with aliases off and on.
• Leaderboard — Every committed review output re-scored and decided by the release policy.
• Adversarial robustness report — How the scorer holds up against reviews built to game it.
• Playground — Score your own review JSON against a bundled case, in the browser, with the same scorer.

Source, tests and the release policy: https://github.com/sebastianfoerste/contract-review-eval-harness.