Methodology v5
Major36-factor scoring config (consumer-sentiment de-scoped; universal 12→11), and the first version to freeze the rubric text alongside its weights.
Impact on live scores
Published; superseded by v6 before any live score was computed under it.
Why we made this change
Closes the rubric-freeze gap: until now no MethodologyVersion referenced the rubric, so every rubric edit was retroactive across every published version — a reader could not see the standard a past score was judged against. v5 snapshots the rubric with the config, so the standard is frozen with the scores it produced.
The full rationale
Two things are true about the removal of consumer-sentiment in this version, and neither is offered as a reason for the other.
FIRST: consumer-sentiment was already slated for removal on independent grounds, recorded in the frozen factor catalog before the audit below existed — "We do not rate consumer sentiment. The only systematic records of it are consumer review platforms whose terms do not permit our access, and we do not publish a score whose basis we cannot show." That is a terms-access argument about what we are able to show, and it is why the factor is marked unrated.
SECOND: on 2026-08-25 a derived provenance audit found that the evidence behind that factor carries an impossible provenance assertion on ALL NINE brands — seventeen observations whose value text states "RENDERED FETCH, verified 2026-07-24" while their retrieval_method column is NULL and the renderer did not exist until 2026-08-23. Fourteen of those were scoring at the time.
The removal was NOT decided because of the finding, and the finding is NOT resolved by the removal. The observations remain on the ledger, uncorrected in themselves and unsuperseded, each with an appended ProvenanceCorrection recording what it asserts and what is actually established. Removing a factor does not unsay a claim.