Evidence methodology · changelog
What changed, and what did not.
Calculated-record update · Luma v0.4 · August 4, 2026
A new synthetic row-level dataset snapshot enabled the locked v0.3 paired-bootstrap calculation. The 95% interval was 33.3%–38.2%, producing a final Claim did not hold up decision. Methodology remained v0.3. View immutable calculated revision · Download reproducibility bundle.
v0.3 · August 4, 2026
Published interval-governed threshold certainty and immutable calculated-record revisions. Luma and Harbor were recalculated under v0.3 using unchanged demonstration snapshots; their v0.2 decisions remain in visible correction history. Luma’s required paired bootstrap interval was unavailable at this historical v0.3 calculated revision, so its threshold certainty was not evaluable and final decision was inconclusive.
v0.2 · July 21, 2026
Archived: added endpoint-specific strength, a predefined primary-endpoint rule, raw quantitative-measurement requirements, paired-cohort display limits, eligibility states, verification and missing-data requirements, and documented limitations.
v0.1 · July 20, 2026
Established the founding-pilot framework for Claim Cards, Decision Receipts, verification, timed follow-ups, and evidence labels.
This change log does not retroactively alter a historical record. Each record retains its earlier decision and shows the later revision separately.
Return to methodology overview