Human Adjudication Results Page

scripts/human_reliability/generate_adjudication_results_page.py generates human-adjudication-results.qmd from approved cohort manifests, normalized adjudication results, and dedicated correction-candidate artifacts.

Run:

npm run human-reliability:adjudication-results

The page reports:

• cohort, source-language, task-layer, and execution state;
• accepted, rejected, deferred, and unresolved outcomes;
• codebook and recoding implications;
• affected-claim dispositions and follow-up requirements; and
• correction-candidate counts and their unpromoted authority state.

The page never reproduces source text, adjudicated values, rationales, evidence, coder values, or reference values. For local_only cohorts it also suppresses all item-level decision metadata and exposes aggregate counts only.

Correction candidates remain proposals beneath the dedicated review layer. Rendering a candidate cannot assign a promotion ID, authorize a direct write, or alter a canonical corpus, analysis, metadata, claim, or publication artifact.