REPORT + CONTEXT: v0.276.0 released, floor 0.276.0

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0159rPz1ZhFKsS53msqPYxtS

REPORT.md

@@ -1,26 +1,23 @@
-# REPORT — v0.275.0: a backup's data and its version travel together (2026-09-26/27)
+# REPORT — v0.276.0: a restore and a drive move keep the app's records (2026-09-27)
 
-Brief: version-travel (Parts A, D2–D4). Architecture: `07-backup-architecture.md` §6 (+ new §6.6), `09` §3, §5.3, §6.4.
-Session record: `felhom.eu/documentation/audits/DRILL-version-travel-2026-09-26.md`.
+Follow-up to the version-travel brief (v0.275.0). Architecture: `07-backup-architecture.md` §6.6. Evidence:
+`felhom.eu/documentation/audits/records-carried-2026-09-27/`.
 
-**Baseline:** `main` `fb2bcdd5d619`, v0.274.0, floor 0.274.0. **Commits:** `b6810f1` (Part A, R-695, R-691 (1), R-694),
-`7fcda8f` (R-699, found live). **Released:** image `0.275.0` (rc `0.275.0-rc1` for the live proofs); floor 0.275.0 with
-MinAgent 0.131.0 — both demo boxes on 0.275.0 within 15 s (`audits/version-travel-2026-09-26/R/`); 9202 by hand (scratch).
+**Not done, or changed:** R-691 (2) (Load from the off-site copy) NOT built — a new restore path on household data with no
+box CC may prove it on; R-700 NOT proven live (no Tier-0 guest has two drives); R-697's carry not exercised by a real
+restore (it would spoil the night's release proof on 9202).
 
-**Tests:** full suite green (`go build/vet/test ./...` rc=0) after the last change; controller gates all OK. New tests:
-`internal/backup/a_version_travel_test.go` (15), `internal/stacks` `TestA5_TheReleaseRefusesAPreConversionDump`,
-`TestA4_AnOldRecordWithoutAServiceChecksEveryPostgresImage`, `TestA3_RestoredVersionPosition`, `TestR695_*`, `TestR694_*`,
-`internal/web` `TestA3_RestoredVersionSentence`, `TestR695_*`, `TestR691_*` (2), `TestR694_*`; new parity fixture
-`deploy_deployed_restored_login`. Two AST fixture names updated (`TestAdmission_IsWiredIntoEveryProductionWriteLeg`).
+**Baseline:** `main` `54bb4da`, v0.275.0, floor 0.275.0. **Commit:** `820e8ef`. **Released:** image `0.276.0`; floor 0.276.0
+(MinAgent 0.131.0); both demo boxes on 0.276.0 within 10 s; 9202 by hand.
 
-**Red-proofs (each seen failing, tree restored)** — `felhom.eu/documentation/audits/version-travel-2026-09-26/`:
-RP1 Tier 1 time = manifest ("the refresh 1s made a two-hour-old dump read as new"); RP2/RP2b the refresh rewrites
-`compose/` (the restore started 16 data with `postgres:18`); RP3 no version check (mismatch/mixed restored); RP4 the
-release on a pre-conversion dump ("released [nextcloud]; copies=0"); RP5 off-site uses the live definition; RP6 R-699 (a
-dump-less unit accepted); D2 R-695 ×2; D3 R-691 ×2; D4 R-694.
+- **R-697:** `PersistUnitRedeployConfig` now carries the life records from the `app.yaml` it replaces
+  (`carryLifeRecords`, `internal/stacks/life_records.go`); `earlier_conversion_copies` keeps a superseded copy's record and
+  `ReleaseConversionCopies` releases every kept copy by the same rule (`releaseOneConversionCopy`).
+- **R-700 (new):** `doFlipRedeploy` persisted through that same fresh write and dropped the pin; now `persistDriveFlip`
+  (load-then-save, `HDD_PATH` only) + `upFromAppConfig` (the tail shared with `RedeployFromEnv`).
 
-**Live (endpoint level, 9202):** `A5-live/README.md` — both shapes restore the old version whole and climb again; the box's
-own night showed the frozen definition and the data-time restore point.
+**Tests:** `internal/stacks/r700_records_carried_test.go` (4). Full suite `go build/vet/test ./...` rc=0; controller gates OK.
+**Red-proofs** RP1–RP4 (`redproofs/`), each seen failing with the wrong value, tree restored.
 
-**NOT yet live-validated:** the off-site restore's definition write; the conversion-copy release on a real 18 dump (due
-after 9202's next nightly backup); R-691 (1) with a real 0770 folder; the English restore sentence.
+**Live:** 9202 paperless-ngx keeps its `conversion_copy` record and volume across the upgrade (`R/R3`). The night check
+(the release on a real 18 dump, through the rewritten loop) is in `N/`.