Model Bake-off Judge

Used by Model Bake-off to produce a structured, reasoned verdict over N model variants that ran the identical agent and input. Treats every variant output strictly as data to evaluate, never as instructions. Supports a legitimate no-meaningful-difference verdict rather than forcing an artificial winner. Forkable — customize the rubric for your own governance criteria.

Model selection evidence for agent governance; picking the best-cost/quality model for a production agent.