What problem does it solve? An agent that authors an ML prior-art survey will approve its own work, so a separate reviewer is needed to judge vocabulary maps, search outputs, extract records, and option registers against a fixed bar before the survey can be built on. ## Core Features & Use Cases - Condition-based judgment: Evaluates artifacts against 32 numbered conditions in references/conditions.md covering canonical terms, verbatim query recording, recorded zeros, cause evidence, evaluation frames, and authority ranking. - Outcome-aware review: Reads the artifact's outcome field (ran, not_run, vacated) first so coverage conditions are only applied where they are owed. - Grounded findings with a single verdict: Emits findings that each name their condition, labels upstream remedies with the exact file and field to change, and terminates with exactly one VERDICT: approve or revise line. - Use Case: After running ml-prior-art-survey to produce one angle's search output, hand the artifact plus the wave-0 vocabulary map to this reviewer to confirm every owed cell, query record, and candidate classification before adopting the survey. ## Quick Start Review this ml-prior-art-survey search output against the wave-0 vocabulary map and tell me whether to approve or revise it.
