What problem does it solve? A numerically correct computational result is still unusable if it cannot be regenerated, was silently overwritten, or its parameters and seeds were never recorded. This Skill provides a structured 18-item checklist to verify that TDL result files — caches, seeds, output paths, date suffixes, and vault entries — are reproducible and correctly recorded before they are trusted as downstream inputs or paper-facing sources. ## Core Features & Use Cases - Provenance Checklist: Walks 18 checks covering date-suffixed filenames, no-overwrite behavior, two-path output rules, seed recording, cache parameter matching, and vault [RESULT] entries. - PASS/GAP Reporting: Reports PASS or GAP per item with the specific file, path, and corrective action (regenerate, re-date, record the seed, file the vault entry). - Escalation Rules: Defines stop conditions such as unrecoverable overwritten results, mismatched reused caches, or missing committed JSONs consumed downstream. - Use Case: Before citing a permutation-test p-value in a paper draft, run this review to confirm the cache was built with the current seed, B, and null model, the deliverable JSON is committed, and the vault entry is filed — catching a stale-cache mismatch like the T1.36 incident. ## Quick Start Review the result file power_analysis_2026-06-03.json against the result provenance checklist and report PASS or GAP for each item.
