The State of the Public Record, 2026-10

The gap index

weight(severity) × (1 + fragmentation bucket)

loaded_share * (0.5 + 0.25*joined + 0.25*rail)

publishable_share

Rulings file hash (sha256):a79a297cf159c9a90e2166108c64d91d9d6331dc462563b62f297f251fd24e66Copy the rulings file hash

The refresh log

A refresh row is one fetch of one register's published file. "Changed" means its sha256 differs from the immediately preceding fetch of the same register (the first fetch of a register is always counted as changed). "Retained" means the fetched bytes are kept in the verifiable source store. The 30-day window below counts every refresh fetched in the 30 days ending at this issue's own issuedAt, across every register this door has ever fetched.

The corrections log

Corrections counted, aggregate counts only, by month, never a subject or a person: flags opened and resolved on the review queue, claims published, and claim disputes opened and resolved. Parcel disputes are not counted: the property dispute door is guidance to the assessor and records no submissions of its own. The reported rail and suspensions are not counted either — both are records, not corrections. The window is grained by calendar month, so it covers every month touching the last 30 days, which can include a few days before the exact 30 day boundary.

The catalog

A register in the catalog carries one status (candidate, probed, spec_written, loaded_dark, public or refused) and, where its terms have been read, one terms verdict (open, non_commercial, unknown or refused). "Opened since" lists every register whose status entered loaded_dark or public within the lookback window, derived from the catalog file's own git history rather than a stored event. At mint time, the coverage door's own held count (11, per register) did not equal the public-corpora count above (12, per kind) — expected, since the two count at different granularities, not an error.