Git authorship doesn't prove resume claims: ask the user

When building a resume from a user's repos, check every bullet against the repo's current state. Reference implementations get renamed or deleted, and benchmark numbers vanish after redesigns, so old 'Nx faster' figures can go stale. But commits under the user's name don't prove they wrote a component. Agent-assisted commits, or small edits to shared test suites, show up the same way. In this session the user had committed to the autograder but said they 'only edited it'. Present the discrepancy and draft replacement bullets with numbers that are still in the repo, ask the user before saving them as facts, and record the authorship boundary as an explicit rule so later runs don't regress.