Latest aggregate-cohort evidence
UnknownAug 17, 2026, 8:17 AM UTC
94distinct sites make up the measurement cohort this page's aggregates describe, and the date above is the newest eligible evidence inside that same cohort; 99 sites have at least one successful load.
Today the aggregate cohort also holds the newest eligible evidence in the committed corpus. This is not the cohort every report page uses. Each report ranks against its own exact cohort, and most committed pages carry a different one than this: a page compares only against scans that share its schema revision, methodology, tracker catalog, role taxonomy, metric contract, producer and requested-GPC state. Where no cohort reaches fifty sites, that page falls back to fixed thresholds instead. A scan run today records the current methodology, and no committed cohort matching its exact tuple, under either requested-GPC state, has reached the 50-site floor, so it is ranked against fixed thresholds and not against this number. Each percentile card names its own denominator, and the methodology states the rule. Category medians are published one cohort per category, and today that is this same cohort. Current means no more than eight days old. It does not mean every site was refreshed in that window. A site that loaded but whose evidence was cut short, by a request cap or a censored family, is counted as covered and never as measured, so the two numbers differ by more than failed loads. Nearly all committed corpus reports are frozen schema v1 from the disclosed fallback collection lane; the methodology states what v1 evidence can and cannot support.