What problem does it solve? Research phases produce derivations, numerical results, and plots whose correctness is easy to claim but hard to confirm. This Skill systematically validates phase outputs against the plan contract, independently recomputing results before asking the researcher to confirm, and records everything in a persistent VERIFICATION.md file. ## Core Features & Use Cases - Contract-backed verification: Extracts claims, deliverables, acceptance tests, and forbidden proxies from the PLAN contract and turns each into a check with qualitative expectations plus a concrete computational test. - Independent pre-computation: Runs numerical spot-checks (dimensional analysis, limiting cases, convergence tests) via shell-executed Python/SymPy before presenting results, so the researcher compares against concrete numbers. - Persistent session state: Writes .gpd/phases/XX-name/{phase}-VERIFICATION.md with frontmatter ledgers (contract_results, comparison_verdicts, gaps) that survive context resets and feed diagnosed gaps into plan-phase. - Use Case: After executing a turbulence modeling phase, run targeted checks with --dimensional or --convergence flags to confirm equations are dimensionally consistent and numerical results converge at the expected order. ## Quick Start Ask the AI to verify phase 4 of the research roadmap by running the gpd-verify-work workflow with the phase number as the argument.
