What problem does it solve? AI-generated roof measurements can silently produce collapsed perimeters, missing ridge/valley topology, invalid pitch values, or customer reports that do not match the aerial imagery. This Skill enforces a QA contract that catches destructive perimeter edits, topology failures, and unsafe report gating before bad geometry reaches customers. ## Core Features & Use Cases - Perimeter Do-No-Harm Gates: Detects destructive refinement collapse (area ratio below 0.85), enforces region-based tree/patio/shadow exclusion, and bounds vertex snap distance with IoU regression reverts. - Topology & Structural Evidence Rules: Preserves deferred DSM ridge/valley/hip candidates, locks backbone seeds before face extraction, and requires a repair pass before rejecting undersegmented topology. - Customer Report Gating: Blocks customer-facing reports unless canonical route provenance, typed roof_lines, valid pitch (with Solar fallback), and all phase diagnostics pass. - Use Case: Before modifying perimeter-refinement.ts or the autonomous graph solver, run the Fonsica regression checklist to verify the eight hard-fail conditions and confirm safe fallback behavior on the canonical 4063 Fonsica Ave test case. ## Quick Start Review my changes to the perimeter refinement phase and verify they pass the Fonsica regression checklist and all geometry QA gates.