Human Adjudication Results

Scoped outcomes, unresolved work, correction candidates, and publication consequences from validated human adjudication
Important

Adjudication resolves study disagreements; it does not silently rewrite accepted annotations. Correction candidates remain review-only proposals until a separate authorized promotion records and applies a canonical change.

This page is generated from validated machine-readable adjudication results and correction-candidate artifacts by scripts/human_reliability/generate_adjudication_results_page.py.

Cohort status

Case Cohort Language Layer State Decisions Accepted Rejected Deferred Unresolved Candidates
not started 0 0 0 0 0 0

Not started means no validated normalized adjudication result exists. Unresolved preserves deferred or unresolved decisions rather than inferring an answer. Results are never pooled across cases, languages, layers, or cohorts.

Rights and source-language scope

Cohort Language Storage Rights constraints Publication detail
No cohorts discovered.

Local-only cohorts expose aggregate state and counts only. Source text, adjudicated values, rationales, evidence, coder values, and reference values are not reproduced on this page.

Decision outcomes

No repository-safe adjudication decision metadata is available. This may mean adjudication has not started or the cohort is local-only.

The table reports decision metadata only. It intentionally omits source text, adjudicated values, rationales, evidence, coder values, and reference values.

Codebook implications

No codebook implications are available.

Codebook clarification, revision, training updates, or recoding needs remain recommendations for the separately governed codebook workflow.

Claim impacts

No claim impacts are available.

review_required and hold_pending_resolution keep affected claims under review; they do not automatically revise publication claims.

Correction candidates and promotion boundary

Candidate counts come from the dedicated correction-candidate layer. Every candidate remains pending_separate_authorization, has no promotion ID, and forbids direct writes. This page reports proposals, not promoted corrections.

Interpretation limits

  • Adjudication outcomes do not replace pre-adjudication human-human agreement.
  • A selected coder or reference basis is a human decision, not a vote.
  • Missing adjudication is an execution state, not evidence of agreement.
  • Source-language qualification and rights restrictions remain cohort-specific.
  • Canonical corpus, analysis, metadata, claim, and publication edits require a separate authorized promotion and regeneration workflow.