What problem does it solve? When developing machine-learning theory proofs, flawed proof sketches can waste significant effort if their structural defects, circular dependencies, or theorem-level obstructions are only discovered during detailed step-level proof work. This Skill acts as a strict early reviewer that validates a proof sketch against its formalized setting before any proof steps are attempted. ## Core Features & Use Cases - Goal Alignment Checking: Verifies the sketch's roadmap actually proves the formalized goal in exact-goal or target-spec mode, comparing quantifiers, domains, constants, and asymptotic scope. - Structure and Dependency Auditing: Ensures every step has a stable ID, exact claim, acyclic dependencies, and that high-risk obligations are localized into lemma-sized steps or explicit blockers. - Obstruction and Circularity Gates: Applies shared contract gates (noncircular closure, scope-accumulation compatibility, mechanism witness, entry-state trace, baseline invariance) to detect theorem-level obstructions early. - Use Case: After drafting a proof sketch for a new generalization bound, run this reviewer to receive an ACCEPTED, REVISE_SKETCH, or IDEA_FAIL verdict with a viability score and the smallest repair target before investing in full proofs. ## Quick Start Review the proof sketch in perspective_1/idea_2 against its setting.md and tell me whether it is ready for step-level proof work.
