Methodology changes are versioned in public.
Each entry names what changed, why it changed, and whether existing scores were affected. Public entries describe methodology concepts; implementation internals remain server-side.
v2.5 · 2026-05-26
Change: Public methodology language was aligned with the production banding behavior.
What was wrong: An older methodology draft described a prior banding scheme. The production score had already been running on the current banding model. The public methodology page now describes the band structure at the same disclosure level as the product: what each band means, how confidence affects interpretation, and where evidence is published.
Impact on existing scores: None. No live score, historical score, or forward-return record was recalculated for this changelog entry. The change corrected public documentation.
Why it matters: A rating system is only useful if the public explanation stays aligned with the operating system. When documentation drifts from production behavior, the correction belongs in the changelog rather than in a silent copy edit.
Evidence path: Forward-return evidence remains published at /backtest. Methodology concepts remain published at /methodology. Rows with incomplete forward-return windows are dropped from evidence tables rather than filled.
Disclosure standard: This changelog describes the public methodology. It does not publish implementation internals or private scoring code.
v2.4 · 2026-05-26
Change: The public methodology page was rewritten around six factor sections: definition, data source, public weight, evidence path, and known limitation.
Impact on existing scores: None. The page clarified how to read the score and confidence badge without changing the score engine.
Disclosure standard: Factor names, source families, and public weights are disclosed. Implementation details remain server-side.