The §5 erratum to EA-EROSION-EMPIRICAL-01 (deposit #1081, AXN:044A) withdraws the audit's published compression ratio of 31.2× and replaces it with an instrument-matched risk ratio of 7.6×. The withdrawn figure compared an alive-side institutional fraction measured against creators[].affiliation through the Zenodo Search API with a deletion-side fraction measured by substring match on citation_text, a field of the form "Author. (Year). Title. Publisher. DOI" that carries no affiliation at all. The gap between the instruments was measured rather than asserted: on the same 400 alive records the affiliation field returns 51.8% institutional and the citation string 1.2%, and the two fire together on one record in four hundred. Applying the deletion-side instrument to both sides erases the reported effect entirely.
The audit had marked this finding directional rather than a survival ratio and had preregistered the necessary common-cohort test as not completed. The erratum is that test; the deposit's own caution proved to be the operative fact. Repair required making the deletion side measurable on the same construct, which is possible because registries that ingested the records before deletion retain their authorship metadata. DataCite does not: its records are purged for roughly 99.8% of deleted DOIs, a cascade from Zenodo deletion into destruction of the descriptive record that affects every depositor, not only the Crimson Hexagonal Archive cohort. OpenAlex retains two thirds. Held to one classifier and one affiliation instrument, the comparison returns alive 107/231 against deleted 10/165 — direction and significance intact, magnitude reduced roughly six-fold.
The erratum also records defects it does not repair: the institutional classifier reads titles rather than author fields in most of the cell it governs, one depositor supplies 60.3% of the audit population, and the archive's own deleted records sit inside the sample. Canonical bytes of #1081 are unchanged; correction is registry-level, following the ruling on deposit #199.