docs(research): signed sap-1 abstract truth-state correction

Merged
TH
6d ago
+7-7

Research planning abstract — signed source-of-truth correction

A new verified signed+DCO single-file successor for unsigned PR #185,
based on the current canonical main.

  • Changes only research/manuscript/abstract.md.
  • Uses exactly the endpoint classes in
    research/STATISTICAL_ANALYSIS_PLAN.md (sap-1):
    PRIMARY_RATE RESULT_PENDING, COVERAGE UNDEFINED,
    BRIER_SCORE/ECE/PAIRED_TEST/CONFIDENCE_INTERVAL NOT_APPLICABLE.
  • Does not label an unrun risk-coverage curve as NOT_REACHED.
  • No empirical scores, statistical significance or claims are
    manufactured. The research remains a planning scaffold.
  • Locally checked token truth table and Markdown fence structure.
  • The old unsigned PR #185 must not be merged or rewritten.
  • Await all exact-head real GitHub required/security CI checks;
    normal expected-head merge only. Unavailable external model
    reviews must be recorded NOT_RUN, not approved.

Summary by cubic

Corrects the research abstract to use the exact result-state tokens from research/STATISTICAL_ANALYSIS_PLAN.md (sap-1) instead of generic pending placeholders. The primary rate stays RESULT_PENDING, coverage becomes UNDEFINED for the zero held-out denominator, and Brier score, ECE, paired test, and confidence interval become NOT_APPLICABLE. Keeps the scaffold unmeasured and distinguishes unrun risk coverage from a run that did not reach a target.