The known-bad corpus offered to GoogleCloudPlatform/knowledge-catalog issue #207 and taken up for conformance fixtures: 111 cases across 16 classes drawn from the 2026-06-19 termination event and its aftermath, each expressed in the thread's ratified axes (validity / presence / edges; presence sparse with exactly removed and never_landed as named states; reason objects with not_disclosed first-class; the removal-fact/content-destruction split embodied by two disclosed producer withdrawals). Verification at build, 2026-07-28: 109 identifiers probed — verified_tombstone 62/62 at HTTP 410; verified_registered 18/18 at 200; verified_erased_registration retained as the untrustworthy class (7×404, 5×410); the new registry_resolution_divergence class 8-for-8 with the DataCite API returning 200-findable while the DOI resolves 410-gone in the same minute. never_landed carried in both directions (record-without-content, content-without-record), commit-hashed. Drift ledger v1.0→v2.0 same-day: population 1,938→1,937 by disclosed withdrawal; recorded registered 75→73; observation drift on the v1 base zero. Deterministic builder and re-probe apparatus included. Canonical home: https://www.alexanarch.org/datasets/deletion-conformance-fixture/
Wiki Article
Deletion Semantics Conformance Fixture v2.0 is a versioned dataset by Lee Sharks, deposited 28 July 2026 — the ratified-axes release of a known-bad corpus built from the archive's own destruction.
It holds 111 cases across 16 classes, drawn from the 19 June 2026 termination event and its aftermath, each expressed in the axes ratified in the upstream discussion: validity, provenance, and the others the thread settled. The corpus was offered to `GoogleCloudPlatform/knowledge-catalog` issue #207 and taken up there for conformance fixtures.
Its significance is in the conversion it performs. A repository deletion that erased 871 works becomes, in this dataset, a test suite — the cases that broke are specified precisely enough that other systems can check whether they break the same way. That is the archive's characteristic move on its own injuries: the loss is not merely documented but rendered into an instrument that outlives the grievance.
Deletion Semantics Conformance Fixture v2.0 — The Ratified-Axes Release: A Known-Bad Identifier Corpus (EA-OKF-FIXTURE-01 v2.0)
Falsification Conditions
Every number re-derivable: verify.py re-probes all cases; build_v2.py rebuilds deterministically from the archived v1.0 base and current sources. A class whose expectation fails on re-probe is a drift datum, recorded, not hidden.
Canonical content: datasets/deletion-conformance-fixture/ in the alexanarch repository — cases.json (111 cases, 16 classes, probed), cases-v1.0.json (archived base), build_v2.py (deterministic builder), verify.py (re-probe apparatus), README.md (the numbers and what they mean), manifest.json. This record is the fixture's address and abstract; the README carries the full verification tables, the drift ledger, the axes expression, and the provenance appendix pinning every figure to its derivation or its source. Offered to knowledge-catalog#207 per the standing exchange (the measurement apparatus behind a conformance fixture's known-bad corpus).
External metadata recovered post-severance (non-authoritative). The sidecar maps each DOI to its locator in the bulk data stores.
Record modifications
2026-07-30 — substrate: Empty substrate declaration repaired during the 1419–1423 mint batch; strict validation (REQ-005) had flagged it fleet-wide.
2026-07-31 — body_status: Body-substance audit: record declared class=full while its body measured as a metadata stub. Reclassified from the bytes.
2026-07-31 — body_status: Body classified by reading its own opening declaration rather than by counting words. Word counting misread works in Greek, emoji, and natively short forms as stubs.
2026-08-04 — publisher: PUB-POPULATE: dc:publisher from venues.json v1.1 press mapping (CP-R3 RULED-EXTENDED 2026-08-01); Alexanarch = publisher of record where no imprint applies
The deposited text is immutable; these are changes to the record's metadata and declared state.