What problem does it solve? When multiple specialized reviewers (structural, rigor, citation, adversarial) independently evaluate a machine-learning theory proof, their findings must be reconciled into a single authoritative accept/reject decision with a clear repair route. This Skill performs that final aggregation, validating each specialized review against machine-readable contracts and emitting one controller-facing verdict. ## Core Features & Use Cases - Contract Validation: Verifies that all four specialized review artifacts use exact raw-token verdicts (PASS/BLOCKING), aligned repair depths, and valid repair targets before aggregation. - Aggregate Contract Gates: Runs independent checks on the LaTeX bundle, including compile/support, theorem self-containedness, assumption provenance, explicit-rate declarations, appendix full-derivation coverage, citation integrity, and template-instantiation hygiene. - Failure-Localized Routing: Selects the deepest blocking issue (idea > sketch > step > assembly) and maps it to a Failure Type, Suggested Next Action, and Retry Mode for the controller. - Use Case: After four specialized reviewers finish evaluating a theorem candidate's LaTeX proof bundle, run this Skill to produce the single proof_review.md that decides acceptance or routes the smallest repair. ## Quick Start Aggregate the completed structural, rigor, citation, and adversarial reviews for perspective_M/idea_N into the final proof_review.md with score, failure type, and retry routing.
