What has changed, and what that does to old results

The overall score weights each category by the evidence weight actually verified in it, rather than counting every category equally.

Under the previous policy a category verified by a single medium-weight control moved the headline score as much as the entire security category. One real run scored 92.85 and fell to 69.44 the moment two thin categories gained their first partial evidence, which is a score reacting to the shape of the pack rather than to the product.

The original policy, weighting every control category equally.

Superseded. Runs scored under it keep their recorded policy version, so an older result stays interpretable as what it was rather than being silently restated.

Deterministic aggregation of structured council opinions into an agreement percentage, with the votes cast and the seats configured reported alongside it.

Current. The aggregation is arithmetic over opinions rather than an opinion of its own, which is what lets it be versioned at all.

The baseline pack: thirteen controls across security, reliability, operations and performance, mapped into six external vocabularies.

Current. Control keys are stable and a change in a control's meaning requires a new control version rather than an edit, so a result recorded against a control still means what it meant.