[gregagi]

What and why

Qualify rustic’s existing shared-restic-format copy with the official mixed-client pruning/write boundary. The official FAQ says rustic does not honor restic locks; restic prune overlapping rustic writes can leave a repository inconsistent. Surface this boundary at the tool-evaluation point, with an official source and a disposable restore-trial recommendation.

• Guarded reversible description migration; update content timestamp only when copy changes.
• Preserve existing URL/title/license/stars/listing date and unrelated rows/fields. No direct production write.
• Cached native Maud source guidance; no new JS, SQL query shapes, request network calls or dependencies.
• Claim guidance, mechanical truth rule, sanitized public-copy apply record and changelog.

Verification

• 30 Rust tests; 17 Python tests; 5 JavaScript tests passed.
• SQLx metadata regenerated against migrated disposable SQLite; no metadata diff. fmt/clippy/build and deterministic monitoring rebuild passed.
• Transactional startup replay verified on a disposable old-copy catalog fixture: only description/updated_at changed.
• Four actual preview states (390/1280, light/dark), visually inspected; eight keyboard journeys to source and category passed. Analytics blocked during QA.
• Focused HTTP, canonical, indexability, cache gzip/304, sitemap membership, category inbound and unrelated listing isolation checks passed.
• Catalog listings emit no JSON-LD before or after this change; no schema-validation or field-CWV claim.
• Initial local build lacked the default cc; using the existing clang-14 completed checks. No checks skipped or waivers applied. PR CI passed for the exact head; deployment40555662293 succeeded for merge aeaefff. Exact production revision, description/date-only catalog change, four live viewport/theme states, eight keyboard journeys and focused HTTP/cache/route checks verified.

Test value: three tiny in-memory SQLite cases cover this migration’s scoped transaction/reversal, rejection of intervening copy edits, and absent/already-applied no-ops. Existing migration tests do not execute this new migration; no live network, sleeps or production databases in these tests.

Scope / follow-up

No new article or duplicate qsv discovery placement. Search/intent checkpoint, exact source observations and experiment windows remain private in the existing Rowset history; no adoption/uplift claim. Preserve earlier outcome windows and October 29 indexing watch. No outreach, extra IndexNow submission or social/newsletter publication. Paid research: no calls, USD0 cap/spend; model/platform billing unknown and separate.