Validation Gate

Note

Generated by scripts/generate-traceability-report.py on 2026-06-28T10:31:47.

The milestone-level publication gate is the minimum check before promoting the audit package or public pages as review-ready.

Required Commands

  1. npm run status
  2. npm run pipeline
  3. npm run validate
  4. quarto render

Gate Meaning

  • npm run status regenerates the project status page from status JSON.
  • npm run pipeline rebuilds case-local outputs, x-case synthesis, and this publication/audit package.
  • npm run validate checks JSON parseability and cross-artifact references, including claim traceability records.
  • quarto render verifies that public pages render.

Limitations To Preserve

  • Lincoln support artifacts remain draft and based on the reviewed pilot sample.
  • Codex-assisted full-corpus MIPVU review is complete for the frozen pre-v1 corpus, but independent human review and reliability adjudication remain pending.
  • Multi-model outputs are diagnostic and remain separate from accepted annotations, prior human review, and any future human reliability study.
  • Model agreement does not prove scholarly reproducibility or correctness.
  • Human reliability status currently records no complete case/language/layer/cohort rows; absent, partial, awaiting-adjudication, or unresolved states block publication-ready human reliability claims.
  • The human reliability completion checklist requires training, calibration, approved samples, two qualified independent coders, validated submissions, metrics, reference comparison, disagreements, required adjudication, codebook notes, protected-path evidence, publication updates, and this repository validation gate before any cohort can be claimed complete.
  • Human agreement, model agreement, adjudication, and historical corroboration remain separate claim families.
  • Rights-restricted packet payloads and raw responses must remain local or withheld according to their source records.
  • Historical-alignment notes still require publication-grade citations.
  • Comparator cases do not yet have case-level support ratings.
  • The existing full pipeline may refresh generated corpus worklists and status files outside this publication package; review generated diffs before committing them.