Repository navigation
Fixed code-review scoring so every same-file candidate reaches the semantic judge before one-to-one assignment. Confirmed expected matches take priority over ignored matches, with line distance used only to break ties. Result construction now requires explicit matches instead of falling back to location-only matching. The judge also accepts identical repeated verdict arrays while rejecting conflicting or malformed output.
Updated the default BC PR Review engine and its inherited Microsoft-layer BCQuality content. Filtering now removes excluded articles' sibling AL samples as well as the articles, and Windows child processes retain command-resolution support. Updated article-coverage annotations and removed associations with retired articles without changing the dataset's required findings.
Fixed AL development-tool compatibility by selecting .NET 8 for BC 24-28 and .NET 10 for BC 29+. Historical Scope = OnPrem table declarations are removed and committed as a common prebuild baseline, preventing compiler incompatibilities from contaminating agent changes.
Results now record harness versions independently of experiment configuration and keep different harness revisions in separate aggregates. Typed metrics payloads retain PR Review usage diagnostics, and bceval exports preserve unavailable AI credits as null rather than measured zero. Result consumers must support the metrics kind discriminator; scores from different benchmark versions must not be combined.
Added modified-entry and explicit engine-SHA options for controlled PR Review evaluation without publishing those runs to the leaderboard or external result stores. Evaluation workflow actions are pinned to immutable commit SHAs for reproducibility.
Versions
• GitHub Copilot CLI 1.0.82
• Claude Code 2.1.221
• Microsoft.Dynamics.BusinessCentral.Development.Tools 18.0.37.11445-beta
• BC-ALAgents 1.39.6 — ecf8e31759d6ddd6d78e3a0b7836b40134368009
• BCQuality content 1.6 — 1a5afdc0ebb6eb3fde1c3c69a13f235275cf4549
• Code-review judge gpt-5.3-codex
• LMChecklist judge gpt-41-2025-04-14