# Sanitized refresh fixtures

Tests construct the exact run-bound evidence in `TemporaryDirectory` because
`run_id`, wall-clock bounds, config hash, extractor hash, and intake paths are
deliberately run-specific. The two real observations are represented only as
fixed enums and counts:

- one reload, page-ready/DOM-read timeouts -> `REFRESH_FAILED_PAGE_UNREADABLE`;
- public discovery `code=-412` -> `BLOCKED_412`, never authoritative no-new.

No Cookie, token, DOM/HTML, response body, media, or formal-data bytes appear in
these fixtures.

HOLD/4 regression cases are generated with sanitized temporary bytes and cover:

- exact observation-card/item binding, missing/extra items, field drift, and
  overlapping non-identical identity components;
- mixed schema 1/2 catalog union, status precedence, identity conflicts, and
  schema 2 artifact path/bytes/SHA-256 drift;
- explicit no-formal-change receipts, immutable hourly slots, exact eight-key
  latest indexes, write-failure reopen, and latest rebuild;
- V004 CREATE_PLANNED recovery for absent and existing dead-holder claims,
  including same-volume quarantine ownership and cleanup.

HOLD/3 adds fail-closed cases for exact schema 2 UID types, schema 1 ASCII ID
length, pending-to-STARTED same-run reconstruction, third slot content,
TERMINAL_RECORDED plan/receipt identity, intent-before-quarantine crashes,
no-overwrite renames, directory flushes, and cleanup after the main lock has
already disappeared.