Abstract

A method is disclosed for detecting drift between a corpus of structured documents held in a version-controlled repository and the database table from which a serving surface presents them, without reading or adding any provenance, digest or synchronisation column. Because an ingester that loads the table stores each entry's whole file text, with carriage-return/line-feed pairs normalised to line feeds, as the entry's body, a digest computed on the repository side over the identically normalised file equals a digest the database computes server-side over the stored body after explicit conversion to a fixed character encoding. One read-only query returns only each row's key, publication status and server-computed digest, so no body crosses the network and the result does not depend on the database's native encoding. The repository side replays the ingester's own validator once over the whole corpus at a configured admission phase, so an entry absent from the store is classified as missing-eligible (the replayed gate admits it, so its absence is rot) or missing-refused (the gate refuses it, so its absence is correct), and only the former enters the alarm count. Present entries are classified as current or stale by digest and, orthogonally, as unpublished when their status hides them from the serving surface; store rows without a repository entry are orphans. A comparison run that cannot reach the store, or cannot read or parse a saved snapshot of it, ends in a distinct unmeasured status, never a clean result.

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.

Share

COinS