Summary

• adds the two-track EvidenceBench v4 evaluator and reproducible research protocol
• documents lessons from Harvey LAB and the validation/release gates
• publishes a safe aggregate-only four-model snapshot
• adds a public professional-review contribution path

Validation

• 37 Python tests pass
• sealed 240-item Doctrine and 60-task Matter corpus validated in the private control repository
• no item-level holdout fields in the public snapshot
• OpenRouter account delta: $90.84, below the $100 ceiling

This is explicitly an unreviewed research release, not an official or attorney-validated leaderboard.