clear_hold (D64 = B) + the box command through the running controller (R-1343), R-1340, R-1341 — tests red first; commit only

- clear_hold: a fifth verb in the closed operator-action list; lifts a hold of any reason in the RUNNING
  controller (memory + settings.json together), forgets the update's held flag, drops a waiting put-back
  from the app-stop marker (copy kept). Refused on a bad name or no hold. No new event type.
- --clear-restore-hold asks the running controller over <data dir>/ctl/hold.sock (unix, 0600, peer uid
  checked, no TCP); edits settings.json only when no controller answers, and says so (in-guest restart).
- R-1340: the operator's HELD STOPPED mail is built from the hold record (restore kind, mixed cause,
  failed put-back step, database replay or not, kept copy) — never the option-C database story by default.
- R-1341: the OOM WARN line once per new kill (container, start time, kill count); event unchanged.

CHANGELOG.md

@@ -1,3 +1,49 @@
+## Unreleased (2026-10-29, afternoon) — clear_hold (D64 = B), the box command through the running controller (R-1343), R-1340, R-1341; commit only, rides the next release
+
+**MinAgent: 0.158.1** (unchanged). The next release also carries R-1342 (`652a690`, already on main, the entry below).
+**No new event type** — the hub records the operator action's result as its own `operator_action` event.
+
+- **`clear_hold` — a fifth verb in the CLOSED operator-action list (operator ruling D64 = B, 2026-10-29 12:10)**
+  (`internal/report/opactions.go`, wired in `cmd/controller/opactions_wiring.go`; arg = the app's stack name, letters,
+  digits, `-`, `_`, ≤64 — checked before any handler). It lifts the app's hold of ANY reason in the RUNNING controller
+  (`backup.AppStopGuard.ClearHoldByOperator` → `settings.ClearRestoreHold`: memory and `settings.json` together), so no
+  later save can write it back. It starts nothing. Refused, nothing changed: a bad name, an app with no hold. One log
+  line names who: „hold CLEARED by the operator via the hub, action #<id>". Every reader of the hold (page, start gate,
+  API, `RestoreHoldFor`, hold reporter) reads the same settings record; the ONE held flag cached elsewhere —
+  `stacks.Manager`'s „the last update ended held" (it made the crash-loop observer and the nightly update leg skip the
+  app) — is forgotten too (`ForgetUpdateHeld`). A `putback_pending` hold is also dropped from the app-stop marker, or the
+  next start with the drive bound would pour the old copy over what the app wrote after the clear; the copy stays on
+  disk. The hold reporter needs no change: its next tick (≤1 min) sends `app_hold_cleared`.
+  Tests: `TestOpActions_ClosedList` (updated deliberately), `TestOpActions_ClearHoldBadArgumentIsRefusedAndCallsNothing`,
+  `TestOpActions_ClearHoldDoneRefusedFailed`, `TestD64_ClearHoldViaTheHub`, `TestD64_ClearHoldRefusals`,
+  `TestD64_ClearHoldStaysGoneAfterALaterSave`, `TestD64_ClearPutBackPendingDropsTheMarkerEntry`,
+  `TestD64_TheHoldReporterTellsTheHubAfterTheClear`, `TestD64_ForgetUpdateHeld`, `TestD64_MainWiresTheNewPieces`.
+- **R-1343 — the box command `--clear-restore-hold <app>` goes through the running controller** (`cmd/controller/holdclear.go`):
+  a unix socket `<data dir>/ctl/hold.sock`, 0600 in a 0700 dir, peer uid checked (SO_PEERCRED), nothing on TCP — not
+  reachable through traefik, the LAN or an app container. Output: „Cleared by the RUNNING controller … no restart is
+  needed". Only when no controller answers (no socket, or a stale one) does it edit `settings.json`, and it says so,
+  naming the restart INSIDE the guest (`pct exec <vmid> -- systemctl restart felhom-controller-bootstrap.service`). The
+  flag's help names the exact guest command. `--restore-holds` (the list) is unchanged.
+  Tests: `TestR1343_TheBoxCommandClearsThroughTheRunningController` (red with the socket skipped: „the hold came back
+  after a later save of the running controller" — R-1343 reproduced), `TestR1343_TheBoxCommandFallsBackWhenNoControllerRuns`,
+  `TestHoldCtl_NotReachableFromOutside`, `TestHoldCtl_UnknownRequestIsRefused`, `TestHoldCtl_ANonSocketFileIsNotReplaced`.
+- **R-1340 — the operator's „HELD STOPPED" mail names the real cause** (`cmd/controller/holdmail.go`
+  `heldStoppedOperatorText`, built from the hold record). Hold writers now record `RestoreKind` (off-site / local),
+  `MixedCause` (`definition_write` / `put_back` / `interrupted_put_back`), `FailedStep` (the put-back step that failed,
+  with its error), `DBReplay`, and the kept copy's folder (`settings.RestoreHold`, all `omitempty`). The mail never says
+  „database" unless the failure was a database replay, never „off-site" for a local restore; a pre-R-1340 mixed hold
+  names only what its record says. Legs unchanged (`restore-hold`, `restore-hold-mixed`); a local restore's RunKind is
+  `restore`. Tests (red with the old sentence/record): `TestR1340_UnhealthyStop`, `_UpdateFailed`,
+  `_RestoreFailedWithADatabase`, `_PutBackWithoutADatabase` (the measured Radicale case), `_PutBackAfterADatabaseReplay`,
+  `_DefinitionWrite`, `_LocalRestore`, `_PutBackPending`, `_InterruptedPutBack`, `_LegacyMixedHoldAssumesNothing`,
+  `_NotifyReadsTheHoldRecord`; writer side `TestR1340_AVolumeFailureWhosePutBackStopFailsRecordsTheStepAndNoDatabase`,
+  `_ADatabaseReplayFailureRecordsTheDatabase`, `_ADefinitionWriteFailureRecordsItsCause`, `_ALocalRestoreRecordsItsKind`,
+  `_AnInterruptedPutBackRecordsItsCause`. `TestEventMessageWireTextIsFrozen` is untouched (this text is passed through).
+- **R-1341 — the OOM line once per NEW kill** (`cmd/controller/oomlog.go` `oomLogGate`, key = container + start time +
+  kill count): a second scan of the same kill logs nothing; a new kill or a restarted container logs again. The
+  `app_oom` event call is unchanged. Tests: `TestR1341_ASecondScanOfTheSameKillLogsNothing`, `_ANewKillLogsAgain`,
+  `_ARestartedContainerLogsAgain`, `_ContainersAreJudgedApart`.
+
 ## Unreleased (2026-10-29, after v0.347.0) — R-1342: the restore copy is safe during the stable-start watch; a removed app does not alarm; commit only, rides the next release
 
 **MinAgent: 0.158.1** (unchanged). Found live on Tester 1 at 10:39 CET: a false `backup_run_failures` mail („Gokapi did

REUSE.md

@@ -82,12 +82,13 @@
 | `restoreOpBlocked` | controller/internal/web/restore_wizard.go | `() (msg string, blocked bool)` | THE refusal gate before starting ANY restore | **Use this, never a bare `IsRunning()`.** It reads BOTH flags: `RestoreStatus().Running` (set synchronously by `BeginRestoreOp`, true for the whole off-box restore) and `IsRunning()` (the concurrency flag, the only one the nightly backup holds). **R-351: all seven handlers read only `IsRunning()`, which the goroutine acquires AFTER the handler returns — a second press started a second run and was told „…elindult".** Returns the Hungarian refusal, which names the running app and a route | **R-360 (v0.226.0): `offboxVerifyCopyDeleteHandler` was the LAST holdout and its doc comment claimed it already did this — the sentence is why nobody looked. `IsRunning()` is FALSE for the whole of a verification restore, so the copy a restore was writing into could be deleted from the UI (observed live 2026-08-21 22:35). No app-name comparison: refusing during ANY restore is stronger and uniform.**
 | `Server.beginRestoreOp` → `backup.TryBeginRestoreOp` / `RestoreBusy` (R-1256 H6) | controller/internal/web/restore_wizard.go; controller/internal/backup/opstatus.go | `(op, stack) (msg string, ok bool)` / `(op, stack) (busy, mirrorApp string)` | THE way a restore op begins: the check for every other operation (a restore op, the backup run, a second-drive copy) and the record in ONE step under `m.mu` — the mirror asks under the same lock (`beginTier2Run`) | **Never call `BeginRestoreOp` from a handler** (`TestR1256_H6_NoHandlerBeginsARestoreUnchecked`): `restoreOpBlocked` asked early can be stale by the time the form is read, and a mirror that began in between ran beside the restore. `restoreOpBlocked` stays as the EARLY refusal; the kept loads begin inside `LoadKeptApp` / `LoadKeptOffsite` and return `ErrRestoreBusy` |
 | `Manager.RestoreFromRecoveryUnitAt` / `RestoreFromRecoveryUnit` + `UnitRestoreResult` (R-353 v0.226.0; R-102 v0.229.0) | controller/internal/backup/restore_unit.go | `(stack, unitDir string) (UnitRestoreResult, error)` / `(stack string) (…)` | THE local recovery-unit restore, and the facts its surface must state | **Returns a RESULT, not just an error** — volumes replayed, DBs replayed, and what the manifest LISTED. The Manifest* counts are load-bearing: zero-replayed has two causes (the backup held no data / the backup listed data that did not come back) and they are opposite news. Pair it with `unitRestoreOutcomeMsg`; do NOT write a new sentence. **A claim about the APP is forbidden on this path** — it has no `SafetyDump` discriminator, unlike the off-site twin (CONTEXT.md ruling, 07-backup-architecture §6.3). `restoreDockerVolumes` now returns `(int, error)`; `restoreDockerVolumesFrom` is unchanged and still the shared implementation. **R-102: `…At` holds the whole body and takes the unit DIRECTORY; the one-argument form is the thin caller naming the primary unit.** THE SOURCE MOVES; THE DESTINATION DOES NOT — `GetAppDrivePath` still resolves where the data lands. The R-47 mutation order, the unit-over-guest secret precedence, the fail-closed data-key gate and the no-unit `RestoreApp` fallback are all pinned and unchanged. The volume leg goes through the R-354 `volumeReplayFrom` seam so the source directory is assertable without Docker |
-| `Manager.putBackAfterFailedRestore` + `liveVolumeCopy` (R-893 option A, v0.336.0) | controller/internal/backup/offbox_restore_undo.go | `(stack, cause, live, copy, volumesMoved, filesPlaced, restart, holdKey) error` | THE restore undo — off-site (v0.336.0) and, since D22 (unreleased), the LOCAL unit restore (`RestoreFromRecoveryUnitAtWith`, so also the second drive's and a kept load): the live volumes it will replace, measured before the stop (`planLiveVolumeCopy`), copied right after it (`copyLiveVolumes`), poured back on any failure | One copy per restore under `<ns>/backups/restore-undo/<app>/` (no capture set reads `backups/`); the put-back uses the restore's own `volumeReplayFrom`/`restoreDockerVolumesFrom`, so there is one volume-replay implementation. A put-back that fails sets `keep` and HOLDS (option C) — never start on volumes that are not known to be back. Seams for tests without Docker: `undoVolumeProbeFn`, `undoVolumeTarFn`, `undoVolumeRemoveFn`; operator notice `SetRestorePutBackNotify` (leg `restore-put-back`). Pinned by `r893_putback_test.go` and `r1177_local_putback_test.go` |
+| `Manager.putBackAfterFailedRestore` + `liveVolumeCopy` (R-893 option A, v0.336.0) | controller/internal/backup/offbox_restore_undo.go | `(stack, cause, live, copy, volumesMoved, filesPlaced, restart, holdKey, restoreKind) error` (R-1340: `restoreKind` = `settings.RestoreKindOffsite`/`Local`; a failed put-back records its step in the hold's `MixedCause`/`FailedStep`/`DBReplay`) | THE restore undo — off-site (v0.336.0) and, since D22 (unreleased), the LOCAL unit restore (`RestoreFromRecoveryUnitAtWith`, so also the second drive's and a kept load): the live volumes it will replace, measured before the stop (`planLiveVolumeCopy`), copied right after it (`copyLiveVolumes`), poured back on any failure | One copy per restore under `<ns>/backups/restore-undo/<app>/` (no capture set reads `backups/`); the put-back uses the restore's own `volumeReplayFrom`/`restoreDockerVolumesFrom`, so there is one volume-replay implementation. A put-back that fails sets `keep` and HOLDS (option C) — never start on volumes that are not known to be back. Seams for tests without Docker: `undoVolumeProbeFn`, `undoVolumeTarFn`, `undoVolumeRemoveFn`; operator notice `SetRestorePutBackNotify` (leg `restore-put-back`). Pinned by `r893_putback_test.go` and `r1177_local_putback_test.go` |
 | `Manager.SweepUnusedRestoreCopies` (R-1256 H8) | controller/internal/backup/restore_copy_sweep.go | `(now) []string` | THE only AUTOMATIC remover of a restore copy nothing names any more (`restore-copy-sweep`, every 6 h); the household's „Törlés most" is `DeleteRestoreCopy` (D62, restore_copies.go — same „needed" rule via `restoreCopyNeededFor`, check + removal under `m.mu`) | Removes a copy only when ALL hold: older than 30 days (D62), the app has no restore hold of ANY kind, the app-stop marker is readable and does not name the app, no restore op runs. It answers „no" when it cannot answer — a held or waiting copy is support's way back |
 | `CrashPutBack` + `Manager.persistUndoForCrash` + `AppStopGuard.SetUndo` (D22 / R-1195, v0.338.0) | controller/internal/backup/restore_crash_putback.go | `NewCrashPutBack(stop, writeDef, logger)`; `PutBackInterruptedRestore(stack, undoDir) error` | A restore cut by a power loss is put back at start: the copy (volumes + `definition/` + `undo.json`) is recorded in the app-stop marker once whole; `Recover` pours it back BEFORE the restart | Call `persistUndoForCrash` right after `copyLiveVolumes` in any new restore path. `Recover` runs before the backup manager exists, so the put-back is wired in main.go with `stackMgr.StopStack` and the stack adapter's definition writer. A failed put-back = a FAILED recovery (app not started, marker kept). The household reads `note.restore.interrupted_put_back` (`LoadRestoreRecordAfterRecovery`). Pinned by `TestR1195_*` |
 | `tier2CaptureSetFull(...) (legs, missing, warns, classified)` + `keptCopyRels` + `dropKeptCopiesUnder` (R-1252, unreleased) | controller/internal/backup/tier2_capture.go, tier2_kept.go | `classified` = the legs come from a VALID backup block | **The second-drive reconcile deletes only under a valid classification**; a rejected or absent block keeps every dest dir; kept copies (D25) are always in the keep set; a mirror deleted by Remove drops its kept records | A new reconcile caller must take `classified` and the kept rels too — never reconcile against the legacy legs. A rejected block is told by `Manager.BackupBlockRejected` (stacks; via the adapter's optional `BackupBlockRejected` method) and signalled once a day (`SetTier2SettingsUnreadableNotify`). Pinned by `r1252_rejected_block_keeps_copies_test.go` |
 | `holdreport.Reporter` + `settings.RestoreHold.Why` / `WireReason` (D53, unreleased) | controller/internal/holdreport/holdreport.go · controller/internal/settings/settings.go | `New(store, notifier.PushEvent, logger)`; `Tick(now)` from the scheduler | Tells the hub every hold in place (`app_held`), every hold gone (`app_hold_cleared`), each again daily | **One observer of the hold store — never a send inside a hold writer.** A new hold writer needs nothing here. The hold sentence is `RestoreHold.Why()` (one per reason) — never a second copy. Pinned by `internal/holdreport/holdreport_test.go` |
 | `AppStopGuard.ResumePutBack` / `ResumeWaitingPutBacks` + `HoldReasonPutBackPending` (R-1251, unreleased) | controller/internal/backup/appstop_putback_wait.go | `ResumePutBack(stack) bool` (true = nothing owed, the caller may start); `ResumeWaitingPutBacks() []string` (puts back, then starts through the gated starter) | A cut restore's copy on a drive that is not bound is WAITING: Recover writes the restore hold `putback_pending` (copy in `SafetyDump`) and starts nothing; the put-back runs when the drive binds, before any start | **Any new path that starts an app when its drive arrives calls `ResumePutBack` first** (the boot recreate and `restartStacks` do, in web via `Server.resumePutBack`). Every other start path is covered by the hold store it already honours — do not add a second check. A failed put-back (at start or on resume) writes the D20 hold (`restore_mixed`) and drops the app from the marker. Needs `SetHolds` + `SetUndoReachable` before `Recover` (main.go). Pinned by `r1251_putback_waits_test.go`, `TestR1251_DriveStartPathsPutBackBeforeTheyStart`, `TestR1251_MainWiresThePutBackWait` |
+| `AppStopGuard.ClearHoldByOperator` + `clearHoldNow` / `listenHoldCtl` / `clearRestoreHoldCLI` (D64 = B, R-1343, unreleased) | controller/internal/backup/hold_clear.go · controller/cmd/controller/holdclear.go | `ClearHoldByOperator(stack) (RestoreHold, error)`; `clearHoldNow(deps, stack, who) (msg, error)` | THE operator's hold clear, any reason — the hub's `clear_hold` and the box command `--clear-restore-hold` both go through `clearHoldNow` in the RUNNING controller | **Never clear a hold from a second process** — the running controller's next save writes it back (R-1343, measured). The box command reaches the running controller over `<data dir>/ctl/hold.sock` (unix, 0600, peer uid checked; never TCP/HTTP). A `putback_pending` clear also drops the app from the app-stop marker (the copy stays). Refusals (`ErrHoldBadStack`, `ErrNotHeld`) change nothing. The update's in-memory held flag goes via `stacks.Manager.ForgetUpdateHeld`. Pinned by `cmd/controller/holdclear_test.go` |
 | `Manager.RestoreTier2Unit` + `Tier2Coverage.CanRestoreUnit` / `Tier2CopyDate` (R-102/R-103, v0.229.0) | controller/internal/backup/tier2_restore.go | `(stack string) (UnitRestoreResult, error)` / `() bool` / `() (string, bool)` | THE full restore from the SECOND DRIVE's mirror, and the predicate that gates it | **Do NOT widen `CanRestore()`** — it answers only "can the additive file restore run?" and one predicate answering two questions is R-356. Fail-closed: `tier2UnitIsOpenable` requires a parseable manifest, because a directory is not a package. The single-writer flag is taken INSIDE `RestoreFromRecoveryUnitAt` — a second `acquireRunning()` here would refuse the restore it guards. `Tier2CopyDate` prefers the SUCCESS anchor over the attempt clock (R-101) and reports which it returned |
 | `unitRestoreOutcomeMsg` + its three message constants (R-353, v0.226.0) | controller/internal/web/handlers.go | `(app string, res backup.UnitRestoreResult) string` | THE customer sentence for a completed LOCAL restore | Twin of `reconstituteOutcomeMsg`; copy its SHAPE (clauses earned by having done the thing, no filesystem path, base names only), never its text. The constants are named because `r353_unit_outcome_test.go` asserts them verbatim — a silent edit is how an honest message drifts back into a comforting one, which is the documented history of the sentence it replaces |
 | `offsiteNoSpaceMsgFmt` + `offsiteSizeUnknownMsg` (R-357, v0.226.0) | controller/internal/backup/offbox_restore.go | two consts | EVERY headroom refusal on the off-site restore surface | **All four gates share these** (prepare, scratch, place, and the destructive reconstitute). A customer meeting one wording on one path and a different one on another has to work out whether it is the same problem. **The reconstitute gate uses NO ×1.1 margin** — it copies a measured tree; `OffboxRestorePrepareFull`'s ×1.1 predicts a download. **Fail closed when either probe reads ≤ 0**: `free < need` with `need == 0` is FALSE, so an unmeasurable input sails through — a gate present and inert |
@@ -456,7 +457,7 @@ Cross-repo edges:
 | timeAgo switch body ×2 | controller/internal/web/funcmap.go `timeAgo` vs `timeAgoStr` (identical formatting logic) |
 | Budapest timezone loader ×4 | controller/internal/scheduler/scheduler.go `getBudapestLocation` vs controller/internal/web/funcmap.go `getTimezone` vs controller/internal/quiesce/quiesce.go `budapestLocation` vs controller/internal/api/kept_install.go `budapest` (v0.277.0, the kept choice's dates) (v0.168.0 window gate — Budapest wall-clock, kept local to avoid a scheduler↔quiesce import edge) |
 | JSON writers ×5, 3 envelope shapes | api `writeJSON`; web `writeDiskJSON`, `jsonResponse`/`jsonError`, `writeDebugJSON` |
-| Safe-name validators ×4 | controller/internal/web/validate.go `validStackName`; controller/internal/api/router.go `validStackParam` (same body — api↔web import cycle); controller/internal/backup/offbox.go `isSafeStackName`; controller/internal/appexport/validate.go `ValidateSegment` (strictest) |
+| Safe-name validators ×5 | controller/internal/web/validate.go `validStackName`; controller/internal/api/router.go `validStackParam` (same body — api↔web import cycle); controller/internal/backup/offbox.go `isSafeStackName` (exported as `IsValidHoldStack` for the hold commands); controller/internal/report/opactions.go `validHoldStack` (same alphabet as `isSafeStackName` — report cannot import backup; clear_hold's argument check); controller/internal/appexport/validate.go `ValidateSegment` (strictest) |
 | DB wait/import ×2 | controller/internal/appbackup/dbdump.go `waitDBReady`/`ImportDump` vs controller/internal/appexport/restore.go `waitForDB`/`importDBDump` |
 | qBittorrent Web API login ×2 | controller/internal/mediawire/mediawire.go `qbLogin` (writes the drive wiring) vs controller/internal/upwatch/explain.go `QBittorrent.login` (reads every minute; keeps its session cookie, which mediawire's per-run login does not need) |
 | compose exec ×2 | controller/internal/stacks/manager.go `composeExecCustomEnv` vs controller/internal/appexport/restore.go `composeExecEnv` (the latter has ctx+timeout; the former has the userdata belt) |

controller/README.md

@@ -1017,7 +1017,8 @@ Each app can define rich metadata in `.felhom.yml`:
     the hold is persisted in `Settings.RestoreHolds`, every start path refuses it (customer button,
     app-stop `Recover()`, boot sweep, via `driveStartGate` **above** its driveless early return), the
     app-stop marker is ended so nothing auto-restarts it, and the row goes red. Cleared with
-    `--clear-restore-hold <app>`, **which requires a controller restart**. `--single-transaction` on
+    `--clear-restore-hold <app>` (since 2026-10-29, R-1343: through the running controller, no restart — see
+    „Operator actions") or the hub's „Clear hold" (`clear_hold`). `--single-transaction` on
     the Postgres import is a belt only; MariaDB DDL is not transactional, which is why the rollback
     is the fix.
   - **Off-site restore of one app: a failed restore puts the app back EXACTLY as it was and starts it (v0.336.0,
@@ -3655,6 +3656,19 @@ Anything else is answered `refused` and calls nothing. Each id runs once per pro
 (`operator_action_results: [{id, outcome, message}]`, outcome `done` / `refused` / `failed`). No action deletes data,
 starts a countdown or shortens one (`TestOpActions_ClosedList`). The box's log records every action it receives.
 
+- **`clear_hold` (2026-10-29, operator ruling D64 = B; arg = the app's stack name, letters/digits/`-`/`_`, ≤64):** lifts
+  the app's hold of ANY reason, in the RUNNING controller (`backup.AppStopGuard.ClearHoldByOperator`, the door
+  `clearHoldNow` in `cmd/controller/holdclear.go`): memory and `settings.json` together, so no later save writes it back
+  (R-1343). It starts nothing — the household's Start (or a restore) does. A waiting put-back (`putback_pending`) is
+  dropped from the app-stop marker too; its copy stays on disk. The update's in-memory „held" flag
+  (`stacks.Manager.ForgetUpdateHeld`) goes with it; the hold reporter's next tick (≤1 min) sends `app_hold_cleared`.
+  Refused, nothing changed: a bad name, or an app with no hold. No new event type: the hub records the result.
+- **The box command `--clear-restore-hold <app>`** (`pct exec <vmid> -- docker exec felhom-controller felhom-controller
+  --clear-restore-hold <app>`) asks the running controller over a unix socket, `<data dir>/ctl/hold.sock` (0600 in a
+  0700 dir, peer uid checked, nothing on TCP — not reachable through traefik, the LAN or an app), and says no restart is
+  needed. Only when no controller answers does it edit `settings.json` itself, and it says so, naming the in-guest
+  restart. `--restore-holds` lists the holds (read-only).
+
 #### Report Push (`internal/report/`)
 
 Periodic JSON push (default every 15 min) to the central felhom-hub service:

controller/cmd/controller/holdclear.go

@@ -0,0 +1,260 @@
+package main
+
+import (
+	"bufio"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"io"
+	"log"
+	"net"
+	"os"
+	"path/filepath"
+	"strings"
+	"time"
+
+	"gitea.dooplex.hu/admin/felhom-controller/internal/backup"
+	"gitea.dooplex.hu/admin/felhom-controller/internal/report"
+	"gitea.dooplex.hu/admin/felhom-controller/internal/settings"
+)
+
+// ── D64 = B + R-1343: the operator clears a hold in the RUNNING controller ─────────────────────────────────────────
+//
+// Two routes reach the same door (clearHoldNow):
+//   - the hub's „Clear hold" → the operator action clear_hold (internal/report/opactions.go), run in this process;
+//   - the box command `felhom-controller --clear-restore-hold <app>` (run with `docker exec` inside the guest) → the
+//     local socket below → the running controller.
+// The running controller clears the hold in memory AND saves settings.json at once (backup.ClearHoldByOperator), so
+// no later save — the hold reporter's, a hub report's — can write it back (R-1343). The household's „held" sentence
+// goes with it: every reader (the page, the start gate, the API, the hold reporter) reads the same settings record.
+// The ONE held flag kept elsewhere — stacks.Manager's „the last update ended held" — is forgotten here too.
+
+// holdClearDeps are clearHoldNow's seams (production: holdClearDepsFor).
+type holdClearDeps struct {
+	clear            func(stack string) (settings.RestoreHold, error) // backup.AppStopGuard.ClearHoldByOperator
+	forgetUpdateHeld func(stack string) bool                          // stacks.Manager.ForgetUpdateHeld (nil = none)
+	logger           *log.Logger
+}
+
+// clearHoldNow lifts stack's hold and returns the result sentence (English, operator-facing). who names the person
+// and the route for the box's own log, e.g. „the operator via the hub, action #12". A refusal (bad name, no hold) is
+// wrapped in report.ErrActionRefused, and nothing was changed.
+func clearHoldNow(d holdClearDeps, stack, who string) (string, error) {
+	if d.clear == nil {
+		return "", fmt.Errorf("%w: clearing a hold is not available on this box", report.ErrActionRefused)
+	}
+	h, err := d.clear(stack)
+	switch {
+	case errors.Is(err, backup.ErrHoldBadStack):
+		return "", fmt.Errorf("%w: %q is not a valid app name — nothing was changed", report.ErrActionRefused, stack)
+	case errors.Is(err, backup.ErrNotHeld):
+		return "", fmt.Errorf("%w: %s has no hold — nothing was changed", report.ErrActionRefused, stack)
+	case err != nil:
+		return "", fmt.Errorf("clearing the hold on %s failed: %v", stack, err)
+	}
+	if d.forgetUpdateHeld != nil && d.forgetUpdateHeld(stack) {
+		if d.logger != nil {
+			d.logger.Printf("[INFO] [hold] %s: the update's held state is forgotten with the hold", stack)
+		}
+	}
+	if d.logger != nil {
+		d.logger.Printf("[WARN] [hold] %s: hold CLEARED by %s — it was %s (set %s); the app is not started by this", stack, who, h.WireReason(), h.At)
+	}
+	msg := fmt.Sprintf("the hold on %s is cleared (it was: %s). The app is NOT started by this — the household's Start, or a restore, starts it", stack, h.Why())
+	if h.Reason == settings.HoldReasonPutBackPending {
+		dir := h.SafetyDump
+		if dir == "" {
+			dir = "(not named in the hold)"
+		}
+		msg += ". Its waiting put-back is dropped: the pre-restore copy stays on disk at " + dir + " and is never poured back by itself"
+	}
+	return msg, nil
+}
+
+// ── the local channel (R-1343) ────────────────────────────────────────────────────────────────────────────────────
+//
+// WHY A UNIX SOCKET IN THE CONTROLLER'S DATA DIR, mode 0600 in a 0700 directory, and nothing on TCP:
+//   - traefik, the LAN and every app reach the controller only over TCP (:8080, :8081). A unix socket has no port,
+//     so none of them can reach it, whatever a proxy header or a forwarded request says;
+//   - the data dir is the controller's own volume: no app container mounts it, and only root in the guest
+//     (`pct exec <vmid> -- docker exec felhom-controller …`, the operator gate this project already uses) can enter
+//     the controller's container to use it;
+//   - the directory is created 0700 BEFORE the socket is bound, so the moment between bind and chmod is not a window;
+//   - the peer's uid must equal this process's (SO_PEERCRED, holdctl_linux.go) — a second lock on the same door.
+// The customer's HTTP surface is deliberately NOT used: it authenticates as the household, and a household clearing
+// its own hold is what the hold exists to prevent. Pinned by TestHoldCtl_NotReachableFromOutside.
+
+const (
+	holdCtlDir  = "ctl"
+	holdCtlSock = "hold.sock"
+	// holdCtlOpClear is the one request this channel answers.
+	holdCtlOpClear = "clear_hold"
+	// holdCtlMaxRequest bounds what one request may send.
+	holdCtlMaxRequest = 4096
+)
+
+// holdCtlPath is the socket's path under the controller's data dir.
+func holdCtlPath(dataDir string) string { return filepath.Join(dataDir, holdCtlDir, holdCtlSock) }
+
+type holdCtlRequest struct {
+	Op    string `json:"op"`
+	Stack string `json:"stack"`
+}
+
+type holdCtlResponse struct {
+	Outcome string `json:"outcome"` // done | refused | failed (report.Outcome*)
+	Message string `json:"message"`
+}
+
+// listenHoldCtl binds the socket and serves it until the listener is closed (closing it removes the socket file).
+// clear is clearHoldNow bound to the running controller's deps.
+func listenHoldCtl(path string, clear func(stack, who string) (string, error), logger *log.Logger) (net.Listener, error) {
+	dir := filepath.Dir(path)
+	if err := os.MkdirAll(dir, 0o700); err != nil {
+		return nil, err
+	}
+	if err := os.Chmod(dir, 0o700); err != nil {
+		return nil, err
+	}
+	// A socket left by a controller that died is replaced; anything else at that path is refused, never removed.
+	if fi, err := os.Lstat(path); err == nil {
+		if fi.Mode()&os.ModeSocket == 0 {
+			return nil, fmt.Errorf("%s exists and is not a socket — refusing to replace it", path)
+		}
+		if err := os.Remove(path); err != nil {
+			return nil, err
+		}
+	}
+	ln, err := net.Listen("unix", path)
+	if err != nil {
+		return nil, err
+	}
+	if err := os.Chmod(path, 0o600); err != nil {
+		ln.Close()
+		return nil, err
+	}
+	go func() {
+		for {
+			conn, err := ln.Accept()
+			if err != nil {
+				if !errors.Is(err, net.ErrClosed) && logger != nil {
+					logger.Printf("[WARN] [hold] local channel stopped: %v", err)
+				}
+				return
+			}
+			go serveHoldCtl(conn, clear, logger)
+		}
+	}()
+	return ln, nil
+}
+
+func serveHoldCtl(conn net.Conn, clear func(stack, who string) (string, error), logger *log.Logger) {
+	defer conn.Close()
+	_ = conn.SetDeadline(time.Now().Add(30 * time.Second))
+	answer := func(r holdCtlResponse) { _ = json.NewEncoder(conn).Encode(r) }
+	if ok, why := holdCtlPeerAllowed(conn); !ok {
+		if logger != nil {
+			logger.Printf("[WARN] [hold] local channel: a request was refused — %s", why)
+		}
+		answer(holdCtlResponse{Outcome: report.OutcomeRefused, Message: "refused: " + why})
+		return
+	}
+	line, err := bufio.NewReader(io.LimitReader(conn, holdCtlMaxRequest)).ReadBytes('\n')
+	if err != nil && len(line) == 0 {
+		answer(holdCtlResponse{Outcome: report.OutcomeRefused, Message: "no request was read"})
+		return
+	}
+	var req holdCtlRequest
+	if err := json.Unmarshal(line, &req); err != nil || req.Op != holdCtlOpClear {
+		answer(holdCtlResponse{Outcome: report.OutcomeRefused, Message: "unknown request — this channel only clears a hold"})
+		return
+	}
+	msg, err := clear(req.Stack, "the operator on the box (--clear-restore-hold)")
+	switch {
+	case err == nil:
+		answer(holdCtlResponse{Outcome: report.OutcomeDone, Message: msg})
+	case errors.Is(err, report.ErrActionRefused):
+		answer(holdCtlResponse{Outcome: report.OutcomeRefused, Message: strings.TrimPrefix(err.Error(), report.ErrActionRefused.Error()+": ")})
+	default:
+		answer(holdCtlResponse{Outcome: report.OutcomeFailed, Message: err.Error()})
+	}
+}
+
+// askRunningControllerToClear sends one clear request. reached=false: no controller answered at path (no socket, or
+// a socket nobody listens on) — nothing was asked, so the caller may edit settings.json itself. reached=true with an
+// error: the request may have been applied; the caller must not edit the file.
+func askRunningControllerToClear(path, stack string) (resp holdCtlResponse, reached bool, err error) {
+	conn, err := net.DialTimeout("unix", path, 3*time.Second)
+	if err != nil {
+		return holdCtlResponse{}, false, err
+	}
+	defer conn.Close()
+	_ = conn.SetDeadline(time.Now().Add(60 * time.Second))
+	if err := json.NewEncoder(conn).Encode(holdCtlRequest{Op: holdCtlOpClear, Stack: stack}); err != nil {
+		return holdCtlResponse{}, true, err
+	}
+	if err := json.NewDecoder(io.LimitReader(conn, 64*1024)).Decode(&resp); err != nil {
+		return holdCtlResponse{}, true, err
+	}
+	return resp, true, nil
+}
+
+// clearRestoreHoldCLI is the box command `--clear-restore-hold <app>` (R-379, R-1343). It asks the running
+// controller first; only when none answers does it edit settings.json itself, and it says which one happened.
+// Returns the process exit code: 0 cleared, 2 refused (nothing changed), 1 failed or unknown.
+func clearRestoreHoldCLI(dataDir, stack string, stdout, stderr io.Writer) int {
+	path := holdCtlPath(dataDir)
+	resp, reached, err := askRunningControllerToClear(path, stack)
+	if reached {
+		if err != nil {
+			fmt.Fprintf(stderr, "the running controller was reached, but its answer was lost: %v\n", err)
+			fmt.Fprintf(stderr, "  The hold may or may not be cleared. Check with: felhom-controller --restore-holds\n")
+			fmt.Fprintf(stderr, "  settings.json was NOT edited by this command.\n")
+			return 1
+		}
+		switch resp.Outcome {
+		case report.OutcomeDone:
+			fmt.Fprintf(stdout, "%s.\n", resp.Message)
+			fmt.Fprintf(stdout, "  Cleared by the RUNNING controller (memory and settings.json together) — no restart is needed.\n")
+			return 0
+		case report.OutcomeRefused:
+			fmt.Fprintf(stderr, "%s\n", resp.Message)
+			return 2
+		default:
+			fmt.Fprintf(stderr, "the running controller could not clear the hold: %s\n", resp.Message)
+			return 1
+		}
+	}
+	// No controller answered: it is not running (or is older than this channel). The file is the only copy.
+	fmt.Fprintf(stdout, "No running controller answered on %s (%v).\n", path, err)
+	fmt.Fprintf(stdout, "  Editing settings.json directly instead.\n")
+	if !backup.IsValidHoldStack(stack) {
+		fmt.Fprintf(stderr, "%q is not a valid app name — nothing was changed\n", stack)
+		return 2
+	}
+	lg := log.New(stderr, "", 0)
+	sett, lerr := settings.Load(filepath.Join(dataDir, "settings.json"), lg)
+	if lerr != nil {
+		fmt.Fprintf(stderr, "restore-hold: loading settings: %v\n", lerr)
+		return 1
+	}
+	cleared, cerr := sett.ClearRestoreHold(stack)
+	if cerr != nil {
+		fmt.Fprintf(stderr, "restore-hold: clearing %s: %v\n", stack, cerr)
+		return 1
+	}
+	if !cleared {
+		// "there was nothing to clear" is not success — a silent 0 here would let an operator believe they had
+		// unblocked an app whose name they mistyped.
+		fmt.Fprintf(stderr, "no restore hold in force for %q — nothing was changed\n", stack)
+		return 2
+	}
+	fmt.Fprintf(stdout, "restore hold cleared for %s in settings.json (no controller was running to ask).\n", stack)
+	fmt.Fprintf(stdout, "  The controller reads it when it starts. If a controller IS running in this guest after all, restart it NOW,\n")
+	fmt.Fprintf(stdout, "  INSIDE THE GUEST — from the Proxmox host: pct exec <vmid> -- systemctl restart felhom-controller-bootstrap.service\n")
+	fmt.Fprintf(stdout, "  — or its next settings save writes the hold back (R-1343).\n")
+	return 0
+}
+
+// holdClearCommandHint is the exact guest command, for the help text and the operator's mail.
+const holdClearCommandHint = "pct exec <vmid> -- docker exec felhom-controller felhom-controller --clear-restore-hold <app>"

controller/cmd/controller/holdclear_test.go

@@ -0,0 +1,516 @@
+package main
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"go/ast"
+	"go/parser"
+	"go/token"
+	"io"
+	"log"
+	"net"
+	"os"
+	"path/filepath"
+	"strconv"
+	"strings"
+	"sync"
+	"testing"
+	"time"
+
+	"gitea.dooplex.hu/admin/felhom-controller/internal/backup"
+	"gitea.dooplex.hu/admin/felhom-controller/internal/config"
+	"gitea.dooplex.hu/admin/felhom-controller/internal/holdreport"
+	"gitea.dooplex.hu/admin/felhom-controller/internal/report"
+	"gitea.dooplex.hu/admin/felhom-controller/internal/settings"
+)
+
+// D64 = B (operator ruling 2026-10-29 12:10) + R-1343: the operator clears a hold — from the hub (clear_hold) or with
+// the box command — IN THE RUNNING CONTROLLER, so the hold is gone from memory and from settings.json together and no
+// later save brings it back.
+//
+// RED-PROOFS (2026-10-29):
+//   - TestR1343_TheBoxCommandClearsThroughTheRunningController: with clearRestoreHoldCLI's socket call skipped (the
+//     old second-process edit of settings.json), it fails on „the hold came back after a later save of the running
+//     controller" — the R-1343 measurement, in a test;
+//   - TestD64_ClearPutBackPendingDropsTheMarkerEntry: with the dropFromMarker call removed from
+//     ClearHoldByOperator, it fails on „a later controller start poured the old copy back";
+//   - TestD64_ClearHoldViaTheHub: with the forgetUpdateHeld call removed from clearHoldNow, it fails on „the
+//     update's held flag was not forgotten".
+//   - TestD64_ClearHoldViaTheHub + TestD64_ClearHoldStaysGoneAfterALaterSave: with ClearHoldByOperator clearing a
+//     SECOND Settings loaded from disk (the old CLI's shape) instead of the running one, they fail on „the running
+//     controller still holds the app" / „the hold came back after a later save of the running controller".
+
+// d64Rig is a running controller's hold side: its Settings, the app-stop guard wired to them, and a log.
+type d64Rig struct {
+	dir      string
+	sett     *settings.Settings
+	guard    *backup.AppStopGuard
+	logBuf   *syncBuffer
+	logger   *log.Logger
+	forgot   []string
+	holdDeps holdClearDeps
+}
+
+type syncBuffer struct {
+	mu sync.Mutex
+	b  bytes.Buffer
+}
+
+func (s *syncBuffer) Write(p []byte) (int, error) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	return s.b.Write(p)
+}
+func (s *syncBuffer) String() string { s.mu.Lock(); defer s.mu.Unlock(); return s.b.String() }
+
+func newD64Rig(t *testing.T) *d64Rig {
+	t.Helper()
+	r := &d64Rig{dir: t.TempDir(), logBuf: &syncBuffer{}}
+	r.logger = log.New(r.logBuf, "", 0)
+	sett, err := settings.Load(filepath.Join(r.dir, "settings.json"), r.logger)
+	if err != nil {
+		t.Fatal(err)
+	}
+	r.sett = sett
+	r.guard = backup.NewAppStopGuard(filepath.Join(r.dir, "appstop-state.json"), r.logger)
+	r.guard.SetHolds(sett)
+	r.holdDeps = holdClearDeps{
+		clear:            r.guard.ClearHoldByOperator,
+		forgetUpdateHeld: func(s string) bool { r.forgot = append(r.forgot, s); return true },
+		logger:           r.logger,
+	}
+	return r
+}
+
+func (r *d64Rig) hold(t *testing.T, h settings.RestoreHold) {
+	t.Helper()
+	if h.At == "" {
+		h.At = "2026-10-29T08:18:00Z"
+	}
+	if err := r.sett.SetRestoreHold(h); err != nil {
+		t.Fatal(err)
+	}
+}
+
+func (r *d64Rig) onDisk(t *testing.T, stack string) bool {
+	t.Helper()
+	s, err := settings.Load(filepath.Join(r.dir, "settings.json"), log.New(io.Discard, "", 0))
+	if err != nil {
+		t.Fatal(err)
+	}
+	_, held := s.GetRestoreHold(stack)
+	return held
+}
+
+func (r *d64Rig) actions() *report.OperatorActions {
+	return report.NewOperatorActions(context.Background(), operatorActionHandlers(nil, nil, r.holdDeps, nil, r.logger), r.logger)
+}
+
+func onlyResult(t *testing.T, o *report.OperatorActions, id int64) report.OperatorActionResult {
+	t.Helper()
+	for _, x := range o.Results() {
+		if x.ID == id {
+			return x
+		}
+	}
+	t.Fatalf("no result for #%d: %+v", id, o.Results())
+	return report.OperatorActionResult{}
+}
+
+// The hub's clear_hold: the hold is gone (memory and disk), the household's held sentence is gone, the update's
+// held flag is forgotten, the box logs who did it, and the result names what was cleared.
+func TestD64_ClearHoldViaTheHub(t *testing.T) {
+	r := newD64Rig(t)
+	r.hold(t, settings.RestoreHold{Stack: "paperless", ReplayError: "ERROR: relation exists", RollbackErr: "psql: refused"})
+	cfg := &config.Config{}
+	cfg.Paths.DataDir = r.dir
+	mgr := backup.NewManager(cfg, r.sett, r.logger)
+	if held, why := mgr.RestoreHoldForLang("paperless", "hu"); !held || why == "" {
+		t.Fatal("fixture: the household must read a held sentence before the clear")
+	}
+
+	o := r.actions()
+	o.Reconcile([]report.OperatorAction{{ID: 12, Action: report.OpClearHold, Arg: "paperless"}})
+	res := onlyResult(t, o, 12)
+	if res.Outcome != report.OutcomeDone {
+		t.Fatalf("result %+v", res)
+	}
+	for _, want := range []string{"the hold on paperless is cleared", "held after a failed restore whose rollback also failed", "NOT started by this"} {
+		if !strings.Contains(res.Message, want) {
+			t.Errorf("result message %q lacks %q", res.Message, want)
+		}
+	}
+	if _, held := r.sett.GetRestoreHold("paperless"); held {
+		t.Fatal("the running controller still holds the app")
+	}
+	if r.onDisk(t, "paperless") {
+		t.Fatal("settings.json still holds the app")
+	}
+	if held, why := mgr.RestoreHoldForLang("paperless", "hu"); held || why != "" {
+		t.Fatalf("the household still reads a held sentence: %q", why)
+	}
+	if len(r.forgot) != 1 || r.forgot[0] != "paperless" {
+		t.Fatalf("the update's held flag was not forgotten: %v", r.forgot)
+	}
+	if lg := r.logBuf.String(); !strings.Contains(lg, "paperless: hold CLEARED by the operator via the hub, action #12") {
+		t.Fatalf("the box's log must say who cleared it:\n%s", lg)
+	}
+}
+
+// Refusals change nothing: an app nobody ever held, a second press on an app already cleared, a bad name.
+func TestD64_ClearHoldRefusals(t *testing.T) {
+	r := newD64Rig(t)
+	r.hold(t, settings.RestoreHold{Stack: "kimai", Reason: settings.HoldReasonUpdateFailed})
+	o := r.actions()
+	o.Reconcile([]report.OperatorAction{{ID: 1, Action: report.OpClearHold, Arg: "nosuchapp"}})
+	if res := onlyResult(t, o, 1); res.Outcome != report.OutcomeRefused || !strings.Contains(res.Message, "nosuchapp has no hold") {
+		t.Fatalf("unknown app: %+v", res)
+	}
+	if _, held := r.sett.GetRestoreHold("kimai"); !held || !r.onDisk(t, "kimai") {
+		t.Fatal("a refusal changed another app's hold")
+	}
+	if len(r.forgot) != 0 {
+		t.Fatalf("a refusal touched the update state: %v", r.forgot)
+	}
+	o.Reconcile([]report.OperatorAction{{ID: 2, Action: report.OpClearHold, Arg: "kimai"}})
+	o.Reconcile([]report.OperatorAction{{ID: 3, Action: report.OpClearHold, Arg: "kimai"}})
+	if res := onlyResult(t, o, 3); res.Outcome != report.OutcomeRefused || !strings.Contains(res.Message, "kimai has no hold") {
+		t.Fatalf("not held (second press): %+v", res)
+	}
+	if _, err := clearHoldNow(r.holdDeps, "../kimai", "test"); err == nil || !strings.Contains(err.Error(), "not a valid app name") {
+		t.Fatalf("a bad name must be refused by the door itself too: %v", err)
+	}
+}
+
+// R-1343's order: clear, then ANOTHER component saves settings (the hold reporter, a hub report), then the file is
+// read back — the hold must stay gone, in memory and on disk.
+func TestD64_ClearHoldStaysGoneAfterALaterSave(t *testing.T) {
+	r := newD64Rig(t)
+	r.hold(t, settings.RestoreHold{Stack: "radicale", Reason: settings.HoldReasonRestoreMixed})
+	o := r.actions()
+	o.Reconcile([]report.OperatorAction{{ID: 5, Action: report.OpClearHold, Arg: "radicale"}})
+	if res := onlyResult(t, o, 5); res.Outcome != report.OutcomeDone {
+		t.Fatalf("result %+v", res)
+	}
+	if err := r.sett.SetReportedHolds(map[string]string{"other": "x|y"}); err != nil { // another component's save
+		t.Fatal(err)
+	}
+	if r.onDisk(t, "radicale") {
+		t.Fatal("the hold came back after a later save of the running controller")
+	}
+	if _, held := r.sett.GetRestoreHold("radicale"); held {
+		t.Fatal("the running controller holds the app again")
+	}
+}
+
+// The same order through the BOX COMMAND: it reaches the running controller over the socket; no restart.
+func TestR1343_TheBoxCommandClearsThroughTheRunningController(t *testing.T) {
+	r := newD64Rig(t)
+	r.hold(t, settings.RestoreHold{Stack: "radicale", Reason: settings.HoldReasonRestoreMixed})
+	ln, err := listenHoldCtl(holdCtlPath(r.dir), func(stack, who string) (string, error) { return clearHoldNow(r.holdDeps, stack, who) }, r.logger)
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer ln.Close()
+
+	var out, errb bytes.Buffer
+	if code := clearRestoreHoldCLI(r.dir, "radicale", &out, &errb); code != 0 {
+		t.Fatalf("exit %d, stdout %q, stderr %q", code, out.String(), errb.String())
+	}
+	if err := r.sett.SetReportedHolds(map[string]string{"other": "x|y"}); err != nil { // the running controller's next save
+		t.Fatal(err)
+	}
+	if r.onDisk(t, "radicale") {
+		t.Fatal("the hold came back after a later save of the running controller")
+	}
+	if !strings.Contains(out.String(), "Cleared by the RUNNING controller") || !strings.Contains(out.String(), "no restart is needed") {
+		t.Fatalf("the command must say it went through the running controller and needs no restart:\n%s", out.String())
+	}
+	if !strings.Contains(r.logBuf.String(), "hold CLEARED by the operator on the box (--clear-restore-hold)") {
+		t.Fatalf("the box's log must say who cleared it:\n%s", r.logBuf.String())
+	}
+
+	// A second run: refused through the socket, exit 2, nothing edited.
+	out.Reset()
+	errb.Reset()
+	if code := clearRestoreHoldCLI(r.dir, "radicale", &out, &errb); code != 2 || !strings.Contains(errb.String(), "radicale has no hold") {
+		t.Fatalf("not held: exit %d, stderr %q", code, errb.String())
+	}
+}
+
+// No controller runs (no socket, or a socket nobody listens on): the command edits settings.json and says so plainly,
+// naming the guest restart for the case a controller runs after all.
+func TestR1343_TheBoxCommandFallsBackWhenNoControllerRuns(t *testing.T) {
+	for _, stale := range []bool{false, true} {
+		r := newD64Rig(t)
+		r.hold(t, settings.RestoreHold{Stack: "kimai", Reason: settings.HoldReasonUpdateFailed})
+		if stale { // a socket left by a controller that died
+			ln, err := listenHoldCtl(holdCtlPath(r.dir), func(string, string) (string, error) { return "", nil }, r.logger)
+			if err != nil {
+				t.Fatal(err)
+			}
+			ln.(*net.UnixListener).SetUnlinkOnClose(false)
+			ln.Close()
+		}
+		var out, errb bytes.Buffer
+		if code := clearRestoreHoldCLI(r.dir, "kimai", &out, &errb); code != 0 {
+			t.Fatalf("stale=%v: exit %d, stderr %q", stale, code, errb.String())
+		}
+		for _, want := range []string{"No running controller answered", "Editing settings.json directly", "INSIDE THE GUEST", "pct exec <vmid> -- systemctl restart felhom-controller-bootstrap.service"} {
+			if !strings.Contains(out.String(), want) {
+				t.Errorf("stale=%v: output lacks %q:\n%s", stale, want, out.String())
+			}
+		}
+		if r.onDisk(t, "kimai") {
+			t.Fatalf("stale=%v: settings.json still holds the app", stale)
+		}
+		out.Reset()
+		errb.Reset()
+		if code := clearRestoreHoldCLI(r.dir, "kimai", &out, &errb); code != 2 {
+			t.Fatalf("stale=%v: not held must exit 2, got %d", stale, code)
+		}
+		if code := clearRestoreHoldCLI(r.dir, "../x", &out, &errb); code != 2 {
+			t.Fatalf("stale=%v: a bad name must exit 2, got %d", stale, code)
+		}
+	}
+}
+
+// A put-back WAITING for its drive is also in the app-stop marker: clearing the hold must drop it there too, or the
+// next controller start with the drive bound would pour the old copy over what the app wrote after the clear. The
+// copy itself stays on disk (nothing is deleted).
+func TestD64_ClearPutBackPendingDropsTheMarkerEntry(t *testing.T) {
+	r := newD64Rig(t)
+	undo := filepath.Join(r.dir, "drive", "felhom-data", "backups", "restore-undo", "immich", "20261029T080000Z")
+	if err := os.MkdirAll(undo, 0o700); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.WriteFile(filepath.Join(undo, "undo.json"), []byte(`{"stack":"immich"}`), 0o600); err != nil {
+		t.Fatal(err)
+	}
+	if err := r.guard.Begin("unit-restore:immich", backup.ReasonUnitRestore, []string{"immich"}); err != nil {
+		t.Fatal(err)
+	}
+	if err := r.guard.SetUndo("immich", undo); err != nil {
+		t.Fatal(err)
+	}
+	r.hold(t, settings.RestoreHold{Stack: "immich", Reason: settings.HoldReasonPutBackPending, SafetyDump: undo})
+
+	o := r.actions()
+	o.Reconcile([]report.OperatorAction{{ID: 3, Action: report.OpClearHold, Arg: "immich"}})
+	res := onlyResult(t, o, 3)
+	if res.Outcome != report.OutcomeDone || !strings.Contains(res.Message, "waiting put-back is dropped") || !strings.Contains(res.Message, undo) {
+		t.Fatalf("result %+v", res)
+	}
+	if _, err := os.Stat(filepath.Join(undo, "undo.json")); err != nil {
+		t.Fatal("the copy must stay on disk")
+	}
+	// The consequence: a later controller start with the drive bound pours nothing back.
+	poured := &recordingPutBacker{}
+	g2 := backup.NewAppStopGuard(filepath.Join(r.dir, "appstop-state.json"), r.logger)
+	g2.SetHolds(r.sett)
+	g2.SetPutBacker(poured)
+	g2.SetStarter(nopStarter{})
+	g2.SetUndoReachable(func(string, string) bool { return true })
+	g2.Recover()
+	if len(poured.stacks) != 0 {
+		t.Fatalf("a later controller start poured the old copy back over the app: %v", poured.stacks)
+	}
+}
+
+type recordingPutBacker struct{ stacks []string }
+
+func (p *recordingPutBacker) PutBackInterruptedRestore(stack, _ string) error {
+	p.stacks = append(p.stacks, stack)
+	return nil
+}
+
+type nopStarter struct{}
+
+func (nopStarter) StartStack(string) error { return nil }
+
+// The hold reporter's next tick tells the hub the hold is gone (it reads the same settings record; nothing cached).
+func TestD64_TheHoldReporterTellsTheHubAfterTheClear(t *testing.T) {
+	r := newD64Rig(t)
+	r.hold(t, settings.RestoreHold{Stack: "kimai", Reason: settings.HoldReasonUpdateFailed})
+	var events []string
+	rep := holdreport.New(r.sett, func(et, _, _ string, _ interface{}) { events = append(events, et) }, r.logger)
+	rep.Tick(timeAt(0))
+	o := r.actions()
+	o.Reconcile([]report.OperatorAction{{ID: 1, Action: report.OpClearHold, Arg: "kimai"}})
+	rep.Tick(timeAt(60))
+	if strings.Join(events, ",") != holdreport.EventHeld+","+holdreport.EventCleared {
+		t.Fatalf("events %v — want app_held then app_hold_cleared", events)
+	}
+}
+
+// „Not reachable from outside": the channel is a unix socket (no TCP anywhere in the file), 0600 in a 0700 directory
+// — even when the directory existed with a looser mode — and main opens it once, under the data dir.
+func TestHoldCtl_NotReachableFromOutside(t *testing.T) {
+	dir := t.TempDir()
+	if err := os.MkdirAll(filepath.Join(dir, holdCtlDir), 0o755); err != nil {
+		t.Fatal(err)
+	}
+	ln, err := listenHoldCtl(holdCtlPath(dir), func(string, string) (string, error) { return "", nil }, nil)
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer ln.Close()
+	if n := ln.Addr().Network(); n != "unix" {
+		t.Fatalf("the channel listens on %q, want unix", n)
+	}
+	fi, err := os.Lstat(holdCtlPath(dir))
+	if err != nil || fi.Mode()&os.ModeSocket == 0 || fi.Mode().Perm() != 0o600 {
+		t.Fatalf("socket mode %v (err %v), want a socket with 0600", fi.Mode(), err)
+	}
+	di, _ := os.Stat(filepath.Join(dir, holdCtlDir))
+	if di.Mode().Perm() != 0o700 {
+		t.Fatalf("socket dir mode %v, want 0700", di.Mode().Perm())
+	}
+
+	// No TCP and no HTTP in the channel's file: every net.Listen/Dial names "unix".
+	f, err := parser.ParseFile(token.NewFileSet(), "holdclear.go", nil, parser.ImportsOnly)
+	if err != nil {
+		t.Fatal(err)
+	}
+	for _, im := range f.Imports {
+		if p, _ := strconv.Unquote(im.Path.Value); strings.HasPrefix(p, "net/http") {
+			t.Fatalf("holdclear.go imports %s — the channel must not be HTTP", p)
+		}
+	}
+	full, err := parser.ParseFile(token.NewFileSet(), "holdclear.go", nil, 0)
+	if err != nil {
+		t.Fatal(err)
+	}
+	nets := 0
+	ast.Inspect(full, func(n ast.Node) bool {
+		call, ok := n.(*ast.CallExpr)
+		if !ok {
+			return true
+		}
+		sel, ok := call.Fun.(*ast.SelectorExpr)
+		if !ok {
+			return true
+		}
+		if x, ok := sel.X.(*ast.Ident); ok && x.Name == "net" && len(call.Args) > 0 {
+			nets++
+			if lit, ok := call.Args[0].(*ast.BasicLit); !ok || lit.Value != `"unix"` {
+				t.Errorf("net.%s is called with a network other than \"unix\"", sel.Sel.Name)
+			}
+		}
+		return true
+	})
+	if nets < 2 {
+		t.Fatalf("expected the listen and the dial in holdclear.go, found %d net calls", nets)
+	}
+
+	// main opens it exactly once, at holdCtlPath(cfg.Paths.DataDir).
+	mf, err := parser.ParseFile(token.NewFileSet(), "main.go", nil, 0)
+	if err != nil {
+		t.Fatal(err)
+	}
+	opened := 0
+	ast.Inspect(mf, func(n ast.Node) bool {
+		call, ok := n.(*ast.CallExpr)
+		if !ok {
+			return true
+		}
+		if id, ok := call.Fun.(*ast.Ident); ok && id.Name == "listenHoldCtl" {
+			opened++
+			arg, ok := call.Args[0].(*ast.CallExpr)
+			if fn, ok2 := arg.Fun.(*ast.Ident); !ok || !ok2 || fn.Name != "holdCtlPath" {
+				t.Error("main must open the channel at holdCtlPath(<data dir>)")
+			}
+		}
+		return true
+	})
+	if opened != 1 {
+		t.Fatalf("main.go opens the channel %d time(s), want 1", opened)
+	}
+}
+
+// The socket answers only its one request; anything else is refused and calls nothing.
+func TestHoldCtl_UnknownRequestIsRefused(t *testing.T) {
+	dir := t.TempDir()
+	called := 0
+	ln, err := listenHoldCtl(holdCtlPath(dir), func(string, string) (string, error) { called++; return "x", nil }, nil)
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer ln.Close()
+	conn, err := net.Dial("unix", holdCtlPath(dir))
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer conn.Close()
+	if _, err := conn.Write([]byte(`{"op":"set_hold","stack":"x"}` + "\n")); err != nil {
+		t.Fatal(err)
+	}
+	var resp holdCtlResponse
+	if err := json.NewDecoder(conn).Decode(&resp); err != nil {
+		t.Fatal(err)
+	}
+	if resp.Outcome != report.OutcomeRefused || called != 0 {
+		t.Fatalf("response %+v, calls %d — want refused and nothing called", resp, called)
+	}
+}
+
+// Something at the socket's path that is NOT a socket is refused, never removed.
+func TestHoldCtl_ANonSocketFileIsNotReplaced(t *testing.T) {
+	dir := t.TempDir()
+	p := holdCtlPath(dir)
+	if err := os.MkdirAll(filepath.Dir(p), 0o700); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.WriteFile(p, []byte("keep me"), 0o600); err != nil {
+		t.Fatal(err)
+	}
+	if _, err := listenHoldCtl(p, func(string, string) (string, error) { return "", nil }, nil); err == nil {
+		t.Fatal("a regular file at the socket path must be refused")
+	}
+	if b, _ := os.ReadFile(p); string(b) != "keep me" {
+		t.Fatal("the file at the socket path was changed")
+	}
+}
+
+// main wires every piece once (the seam-built-but-never-wired trap): the clear door into the operator actions, the
+// R-1340 text into the hold notification, the R-1341 gate into the OOM loop, and the box command.
+func TestD64_MainWiresTheNewPieces(t *testing.T) {
+	f, err := parser.ParseFile(token.NewFileSet(), "main.go", nil, 0)
+	if err != nil {
+		t.Fatal(err)
+	}
+	found := map[string]int{}
+	var opArgs []ast.Expr
+	ast.Inspect(f, func(n ast.Node) bool {
+		call, ok := n.(*ast.CallExpr)
+		if !ok {
+			return true
+		}
+		switch fn := call.Fun.(type) {
+		case *ast.SelectorExpr:
+			if x, ok := fn.X.(*ast.Ident); ok {
+				found[x.Name+"."+fn.Sel.Name]++
+			}
+		case *ast.Ident:
+			found[fn.Name]++
+			if fn.Name == "operatorActionHandlers" {
+				opArgs = call.Args
+			}
+		}
+		return true
+	})
+	for _, want := range []string{"heldStoppedOperatorText", "heldHoldFromNotify", "oomLog.fresh", "clearRestoreHoldCLI", "listenHoldCtl", "clearHoldNow"} {
+		if found[want] != 1 {
+			t.Errorf("main.go calls %s %d time(s), want exactly 1", want, found[want])
+		}
+	}
+	if len(opArgs) < 3 {
+		t.Fatal("operatorActionHandlers call not found")
+	}
+	if id, ok := opArgs[2].(*ast.Ident); !ok || id.Name != "holdDeps" {
+		t.Error("main must hand holdDeps (the running controller's clear door) to operatorActionHandlers")
+	}
+}
+
+func timeAt(sec int) time.Time { return time.Date(2026, 10, 29, 12, 0, sec, 0, time.UTC) }

controller/cmd/controller/holdctl_linux.go

@@ -0,0 +1,34 @@
+//go:build linux
+
+package main
+
+import (
+	"fmt"
+	"net"
+	"os"
+	"syscall"
+)
+
+// holdCtlPeerAllowed (R-1343) admits a peer on the hold socket only when its uid equals this process's (SO_PEERCRED).
+// The socket's 0600 mode in a 0700 directory already says so; this is the second lock on the same door.
+func holdCtlPeerAllowed(conn net.Conn) (bool, string) {
+	uc, ok := conn.(*net.UnixConn)
+	if !ok {
+		return false, "not a unix-socket peer"
+	}
+	raw, err := uc.SyscallConn()
+	if err != nil {
+		return false, "the peer's identity could not be read"
+	}
+	var cred *syscall.Ucred
+	var cerr error
+	if err := raw.Control(func(fd uintptr) {
+		cred, cerr = syscall.GetsockoptUcred(int(fd), syscall.SOL_SOCKET, syscall.SO_PEERCRED)
+	}); err != nil || cerr != nil || cred == nil {
+		return false, "the peer's identity could not be read"
+	}
+	if int(cred.Uid) != os.Geteuid() {
+		return false, fmt.Sprintf("the peer's uid %d is not this controller's", cred.Uid)
+	}
+	return true, ""
+}

controller/cmd/controller/holdctl_other.go

@@ -0,0 +1,14 @@
+//go:build !linux
+
+package main
+
+import "net"
+
+// holdCtlPeerAllowed: the peer check needs SO_PEERCRED (Linux, where the controller runs). Elsewhere the socket's
+// 0600 mode in a 0700 directory is the only lock.
+func holdCtlPeerAllowed(conn net.Conn) (bool, string) {
+	if _, ok := conn.(*net.UnixConn); !ok {
+		return false, "not a unix-socket peer"
+	}
+	return true, ""
+}

controller/cmd/controller/holdmail.go

@@ -0,0 +1,155 @@
+package main
+
+import (
+	"fmt"
+	"strings"
+
+	"gitea.dooplex.hu/admin/felhom-controller/internal/settings"
+)
+
+// ── R-1340: the operator's „HELD STOPPED" mail names the hold's REAL cause ────────────────────────────────────────
+//
+// Until 2026-10-29 the restore-hold mail told one story for every restore_mixed hold — R-893 option C's „its off-site
+// database restore failed. The database was rolled back …, but its files and named volumes are the snapshot's" —
+// written when option C was the only mixed hold. Measured on Tester 1 (D63's H5 proof): Radicale, which has no
+// database, whose put-back stopped at „the app could not be stopped" with nothing moved, got exactly that text.
+// Support read a wrong account of the box.
+//
+// The text is now built from the HOLD RECORD the writer persisted just before the mail (its reason and the R-1340
+// fields: which restore, which branch, which step failed, whether the failure was a database replay, where the copy
+// is kept). It never says „database" unless the restore's failure WAS a database replay, and never „off-site" for a
+// restore from the app's own backup. One sentence per reason; pinned by TestR1340_* (holdmail_test.go).
+
+// heldStoppedOperatorText is the operator mail for a held app: the message, its leg (the hub reads the leg, so the
+// two existing ones are kept: restore-hold for a restore hold, restore-hold-mixed for a mixed one) and a short reason.
+func heldStoppedOperatorText(h settings.RestoreHold) (msg, leg, reason string) {
+	stack := h.Stack
+	const tail = " The app will not start from any path until the hold is cleared (the hub's Clear hold button, or on the box: " + holdClearCommandHint + ")."
+	kind := restoreKindPhrase(h.RestoreKind)
+	switch h.Reason {
+	case settings.HoldReasonUnhealthyStop:
+		what := "a crash loop"
+		if h.UnhealthyKind == "oom_storm" {
+			what = "an out-of-memory storm"
+		}
+		trip := ""
+		if h.Trip >= 2 {
+			trip = fmt.Sprintf(" This is stop %d within 24 hours.", h.Trip)
+		}
+		msg = fmt.Sprintf("App %q is HELD STOPPED: the box stopped it for %s at %s.%s No restore or update was involved and no data was touched. "+
+			"The household's Start gives it one more try; clearing the hold lifts it too.", stack, what, h.At, trip)
+		return msg, "unhealthy-stop", "stopped for " + what
+	case settings.HoldReasonUpdateFailed:
+		undo := ""
+		switch h.UndoState {
+		case "untouched":
+			undo = " The automatic undo failed; the data was not touched."
+		case "half":
+			undo = " The automatic undo failed half-way; the data may be partly the new version's."
+		case "not_started":
+			undo = " The automatic undo did not start."
+		}
+		copyLine := " Restoring it from its backup starts it."
+		if h.NoWholeCopy {
+			copyLine = " No copy on this box brings it back whole — it needs support."
+		} else if h.CopyDate != "" {
+			copyLine = fmt.Sprintf(" Restoring it from its backup (tier %d, %s) starts it.", h.CopyTier, h.CopyDate)
+		}
+		msg = fmt.Sprintf("App %q is HELD STOPPED: its update at %s did not come up healthy.%s%s", stack, h.At, undo, copyLine) + tail
+		return msg, "update-hold", "update failed"
+	case settings.HoldReasonPutBackPending:
+		msg = fmt.Sprintf("App %q is HELD STOPPED: a restore was cut after its copy of the live data was taken, and that copy (%s) is on a drive that is not bound. "+
+			"The copy is poured back and the app started when the drive is bound. Clearing the hold DROPS that put-back; the copy stays on disk.", stack, orNone(h.SafetyDump))
+		return msg, "putback-pending", "put-back waits for its drive"
+	case settings.HoldReasonRestoreMixed:
+		return mixedHoldText(h, kind, tail)
+	default: // HoldReasonRestoreFailed: a database replay failed AND its rollback failed (holdAppAfterFailedRollback)
+		what := "its " + kind + " failed"
+		if h.DBReplay || h.RestoreKind == "" { // every restore hold before R-1340 was written by the database branch
+			what = "its " + kind + " failed at the database replay AND the rollback to the pre-restore database copy also failed — the database may be half-written"
+		}
+		msg = fmt.Sprintf("App %q is HELD STOPPED: %s.", stack, what) + tail +
+			fmt.Sprintf(" Replay error: %s. Rollback error: %s", orNone(h.ReplayError), orNone(h.RollbackErr))
+		reason = h.RollbackErr
+		if reason == "" {
+			reason = "rollback failed"
+		}
+		return msg, "restore-hold", reason
+	}
+}
+
+func mixedHoldText(h settings.RestoreHold, kind, tail string) (msg, leg, reason string) {
+	stack := h.Stack
+	leg = "restore-hold-mixed"
+	failedAt := ""
+	if h.DBReplay {
+		failedAt = " at the database replay"
+	}
+	copyLine := " There is no copy of the live volumes."
+	if h.SafetyDump != "" {
+		copyLine = " The copy of the live data is kept at " + h.SafetyDump + "."
+	}
+	switch h.MixedCause {
+	case settings.MixedCauseDefinitionWrite:
+		msg = fmt.Sprintf("App %q is HELD STOPPED for support: its %s failed while writing the backup's app definition, before any data was touched, "+
+			"and %s. The app's data is the live data; the definition on disk may be the backup's.", stack, kind, orNone(h.FailedStep)) + tail +
+			" Error: " + orNone(h.ReplayError)
+		return msg, leg, "definition write failed; writing the live definition back failed too"
+	case settings.MixedCausePutBack:
+		msg = fmt.Sprintf("App %q is HELD STOPPED for support: its %s failed%s, and putting the live data back FAILED — %s. "+
+			"The app may be on a mix of the backup's and the live data.%s", stack, kind, failedAt, orNone(h.FailedStep), copyLine) + tail +
+			" Restore error: " + orNone(h.ReplayError)
+		return msg, leg, "put-back failed: " + firstClause(h.FailedStep)
+	case settings.MixedCauseInterruptedPutBack:
+		msg = fmt.Sprintf("App %q is HELD STOPPED for support: a restore cut by a stop could not be put back at the next start — %s. "+
+			"The app may be on a mix of the backup's and the live data.%s", stack, orNone(h.FailedStep), copyLine) + tail
+		return msg, leg, "interrupted restore: put-back failed"
+	}
+	// A mixed hold written before R-1340 carries no cause: name only what the record says, and assume nothing.
+	msg = fmt.Sprintf("App %q is HELD STOPPED for support: a restore stopped half-way (set %s); the app may be on a mix of the backup's and the live data. "+
+		"The hold record says: %s.", stack, h.At, orNone(h.ReplayError)) + tail
+	return msg, leg, "restore stopped half-way"
+}
+
+func restoreKindPhrase(k string) string {
+	switch k {
+	case settings.RestoreKindLocal:
+		return "restore from its own backup"
+	case settings.RestoreKindOffsite:
+		return "off-site restore"
+	}
+	return "restore"
+}
+
+func orNone(s string) string {
+	if strings.TrimSpace(s) == "" {
+		return "(none recorded)"
+	}
+	return s
+}
+
+// firstClause is a step's text up to its error (the short reason field).
+func firstClause(s string) string {
+	if i := strings.Index(s, ":"); i > 0 {
+		return s[:i]
+	}
+	return s
+}
+
+// heldHoldFromNotify is the hold the notification describes: the record the writer persisted just before it, or —
+// when that write failed — one rebuilt from the errors handed over (the pre-R-1340 rule: no rollback error = mixed).
+func heldHoldFromNotify(get func(string) (settings.RestoreHold, bool), stack string, replayErr, rollbackErr error) settings.RestoreHold {
+	if get != nil {
+		if h, ok := get(stack); ok && (h.Reason == settings.HoldReasonRestoreFailed || h.Reason == settings.HoldReasonRestoreMixed) {
+			return h
+		}
+	}
+	h := settings.RestoreHold{Stack: stack, Reason: settings.HoldReasonRestoreMixed}
+	if rollbackErr != nil {
+		h.Reason, h.RollbackErr = settings.HoldReasonRestoreFailed, rollbackErr.Error()
+	}
+	if replayErr != nil {
+		h.ReplayError = replayErr.Error()
+	}
+	return h
+}

controller/cmd/controller/holdmail_test.go

@@ -0,0 +1,140 @@
+package main
+
+import (
+	"errors"
+	"strings"
+	"testing"
+
+	"gitea.dooplex.hu/admin/felhom-controller/internal/settings"
+)
+
+// R-1340: the operator's „HELD STOPPED" mail names the hold's real cause — one test per reason. Each asserts the words
+// that MUST be there and the words that must NOT (the option-C database story was the bug).
+//
+// RED-PROOF (2026-10-29): with mixedHoldText's body replaced by the pre-R-1340 sentence („its off-site database restore
+// failed. The database was rolled back to the pre-restore copy and the live definition written back, but its files
+// and named volumes are the snapshot's"), TestR1340_PutBackWithoutADatabase, _DefinitionWrite, _LocalRestore and
+// _InterruptedPutBack fail on „database" / „off-site" being present and their own step missing.
+
+const optionCStory = "files and named volumes are the snapshot's"
+
+func mustHave(t *testing.T, msg string, want ...string) {
+	t.Helper()
+	for _, w := range want {
+		if !strings.Contains(msg, w) {
+			t.Errorf("the mail must say %q:\n%s", w, msg)
+		}
+	}
+}
+
+func mustNotHave(t *testing.T, msg string, bad ...string) {
+	t.Helper()
+	for _, b := range bad {
+		if strings.Contains(strings.ToLower(msg), strings.ToLower(b)) {
+			t.Errorf("the mail must NOT say %q:\n%s", b, msg)
+		}
+	}
+}
+
+func TestR1340_UnhealthyStop(t *testing.T) {
+	msg, leg, _ := heldStoppedOperatorText(settings.RestoreHold{Stack: "romm", At: "2026-10-29T08:00:00Z", Reason: settings.HoldReasonUnhealthyStop, UnhealthyKind: "crash_loop"})
+	mustHave(t, msg, `"romm"`, "crash loop", "no data was touched", "Start gives it one more try")
+	mustNotHave(t, msg, "database", "restore failed", optionCStory)
+	if leg != "unhealthy-stop" {
+		t.Errorf("leg %q", leg)
+	}
+	oom, _, _ := heldStoppedOperatorText(settings.RestoreHold{Stack: "romm", Reason: settings.HoldReasonUnhealthyStop, UnhealthyKind: "oom_storm", Trip: 2})
+	mustHave(t, oom, "out-of-memory storm", "stop 2 within 24 hours")
+}
+
+func TestR1340_UpdateFailed(t *testing.T) {
+	msg, leg, _ := heldStoppedOperatorText(settings.RestoreHold{Stack: "kimai", At: "2026-10-29T03:10:00Z", Reason: settings.HoldReasonUpdateFailed,
+		UndoState: "half", CopyTier: 1, CopyDate: "2026-10-29T02:00:00Z"})
+	mustHave(t, msg, "update at 2026-10-29T03:10:00Z did not come up healthy", "undo failed half-way", "tier 1, 2026-10-29T02:00:00Z", "--clear-restore-hold")
+	mustNotHave(t, msg, "database", optionCStory)
+	if leg != "update-hold" {
+		t.Errorf("leg %q", leg)
+	}
+}
+
+func TestR1340_RestoreFailedWithADatabase(t *testing.T) {
+	msg, leg, reason := heldStoppedOperatorText(settings.RestoreHold{Stack: "paperless", Reason: settings.HoldReasonRestoreFailed, RestoreKind: settings.RestoreKindOffsite,
+		DBReplay: true, ReplayError: "ERROR: relation exists", RollbackErr: "psql: connection refused"})
+	mustHave(t, msg, "off-site restore failed at the database replay", "rollback to the pre-restore database copy also failed", "ERROR: relation exists", "psql: connection refused")
+	mustNotHave(t, msg, optionCStory)
+	if leg != "restore-hold" || reason != "psql: connection refused" {
+		t.Errorf("leg %q reason %q", leg, reason)
+	}
+}
+
+// The measured case: Radicale has no database; the volume could not be removed and the put-back's stop failed.
+func TestR1340_PutBackWithoutADatabase(t *testing.T) {
+	msg, leg, reason := heldStoppedOperatorText(settings.RestoreHold{Stack: "radicale", Reason: settings.HoldReasonRestoreMixed, RestoreKind: settings.RestoreKindOffsite,
+		MixedCause: settings.MixedCausePutBack, FailedStep: "stopping the app before the put-back failed, so nothing was put back: compose down: timeout",
+		ReplayError: "volume rm -f radicale_data: volume is in use", SafetyDump: "/mnt/hdd/felhom-data/backups/restore-undo/radicale/20261029T081800Z"})
+	mustHave(t, msg, `"radicale"`, "off-site restore failed, and putting the live data back FAILED", "stopping the app before the put-back failed",
+		"compose down: timeout", "kept at /mnt/hdd/felhom-data/backups/restore-undo/radicale/20261029T081800Z", "volume is in use")
+	mustNotHave(t, msg, "database", "rolled back", optionCStory)
+	if leg != "restore-hold-mixed" || reason != "put-back failed: stopping the app before the put-back failed, so nothing was put back" {
+		t.Errorf("leg %q reason %q", leg, reason)
+	}
+}
+
+func TestR1340_PutBackAfterADatabaseReplay(t *testing.T) {
+	msg, _, _ := heldStoppedOperatorText(settings.RestoreHold{Stack: "immich", Reason: settings.HoldReasonRestoreMixed, RestoreKind: settings.RestoreKindOffsite,
+		MixedCause: settings.MixedCausePutBack, DBReplay: true, FailedStep: "putting the live volumes back failed: tar xf failed"})
+	mustHave(t, msg, "failed at the database replay", "putting the live volumes back failed: tar xf failed", "There is no copy of the live volumes")
+}
+
+func TestR1340_DefinitionWrite(t *testing.T) {
+	msg, _, _ := heldStoppedOperatorText(settings.RestoreHold{Stack: "gokapi", Reason: settings.HoldReasonRestoreMixed, RestoreKind: settings.RestoreKindOffsite,
+		MixedCause: settings.MixedCauseDefinitionWrite, FailedStep: "writing the live app definition back failed: app.yaml write failed", ReplayError: "app.yaml write failed"})
+	mustHave(t, msg, "while writing the backup's app definition, before any data was touched", "writing the live app definition back failed", "data is the live data")
+	mustNotHave(t, msg, "database", optionCStory)
+}
+
+func TestR1340_LocalRestore(t *testing.T) {
+	msg, _, _ := heldStoppedOperatorText(settings.RestoreHold{Stack: "app", Reason: settings.HoldReasonRestoreMixed, RestoreKind: settings.RestoreKindLocal,
+		MixedCause: settings.MixedCausePutBack, FailedStep: "putting the live volumes back failed: tar xf failed", SafetyDump: "/srv/undo/app/1"})
+	mustHave(t, msg, "restore from its own backup failed", "putting the live volumes back failed")
+	mustNotHave(t, msg, "off-site", "database", optionCStory)
+}
+
+func TestR1340_PutBackPending(t *testing.T) {
+	msg, leg, _ := heldStoppedOperatorText(settings.RestoreHold{Stack: "immich", Reason: settings.HoldReasonPutBackPending, SafetyDump: "/mnt/hdd/undo/immich/1"})
+	mustHave(t, msg, "copy (/mnt/hdd/undo/immich/1) is on a drive that is not bound", "poured back and the app started when the drive is bound", "DROPS that put-back")
+	mustNotHave(t, msg, "database", optionCStory)
+	if leg != "putback-pending" {
+		t.Errorf("leg %q", leg)
+	}
+}
+
+func TestR1340_InterruptedPutBack(t *testing.T) {
+	msg, _, _ := heldStoppedOperatorText(settings.RestoreHold{Stack: "immich", Reason: settings.HoldReasonRestoreMixed, MixedCause: settings.MixedCauseInterruptedPutBack,
+		FailedStep: "putting the interrupted restore's copy back failed: compose down: timeout", SafetyDump: "/mnt/hdd/undo/immich/1"})
+	mustHave(t, msg, "cut by a stop could not be put back", "compose down: timeout", "kept at /mnt/hdd/undo/immich/1")
+	mustNotHave(t, msg, "database", "off-site", optionCStory)
+}
+
+// A mixed hold written before R-1340 names only what its record says.
+func TestR1340_LegacyMixedHoldAssumesNothing(t *testing.T) {
+	msg, _, _ := heldStoppedOperatorText(settings.RestoreHold{Stack: "x", At: "2026-10-01T00:00:00Z", Reason: settings.HoldReasonRestoreMixed, ReplayError: "boom"})
+	mustHave(t, msg, "stopped half-way", "The hold record says: boom")
+	mustNotHave(t, msg, "database", optionCStory)
+}
+
+// The notification reads the hold the writer persisted; when that write failed, it rebuilds one from the errors.
+func TestR1340_NotifyReadsTheHoldRecord(t *testing.T) {
+	rec := settings.RestoreHold{Stack: "radicale", Reason: settings.HoldReasonRestoreMixed, MixedCause: settings.MixedCausePutBack, FailedStep: "x"}
+	get := func(s string) (settings.RestoreHold, bool) { return rec, s == "radicale" }
+	if h := heldHoldFromNotify(get, "radicale", errors.New("e"), nil); h.MixedCause != settings.MixedCausePutBack {
+		t.Fatalf("the persisted record was not used: %+v", h)
+	}
+	h := heldHoldFromNotify(get, "other", errors.New("replay"), errors.New("rollback"))
+	if h.Reason != settings.HoldReasonRestoreFailed || h.RollbackErr != "rollback" || h.ReplayError != "replay" {
+		t.Fatalf("fallback hold %+v", h)
+	}
+	if h := heldHoldFromNotify(nil, "other", errors.New("replay"), nil); h.Reason != settings.HoldReasonRestoreMixed {
+		t.Fatalf("no rollback error = mixed, got %+v", h)
+	}
+}

controller/cmd/controller/main.go

@@ -92,8 +92,9 @@ func main() {
 	abandonStatus := flag.Bool("abandon-status", false, "R-241: print the state of this box's off-site abandonment countdown (set-aside path, due date, days left) and exit. Read-only.")
 	abandonExtend := flag.Int("abandon-extend", 0, "R-241 (operator): extend a running abandonment countdown by N days from now, then exit. Refuses when no countdown is running.")
 	abandonStop := flag.Bool("abandon-stop", false, "R-241 (operator): stop a running abandonment countdown, then exit. The set-aside history is kept and nothing is deleted. Refuses when no countdown is running.")
-	listRestoreHolds := flag.Bool("restore-holds", false, "R-379 (operator): list apps this controller is deliberately holding stopped after a failed database restore whose rollback also failed, then exit. Read-only.")
-	clearRestoreHold := flag.String("clear-restore-hold", "", "R-379 (operator): clear the restore hold on APP so it can start again, then exit. Clearing does NOT repair the database — check the app's data first; the undo copies are in its unit's db-dumps dir.")
+	listRestoreHolds := flag.Bool("restore-holds", false, "R-379 (operator): list the apps this controller is holding stopped (every reason, with why), then exit. Read-only.")
+	clearRestoreHold := flag.String("clear-restore-hold", "", "R-379/D64 (operator): clear the hold on APP (any reason), then exit. Run it inside the guest: "+holdClearCommandHint+
+		". It asks the RUNNING controller, so no restart is needed; only when no controller runs does it edit settings.json. Clearing repairs nothing and starts nothing — check the app's data first.")
 	flag.Parse()
 
 	if *showVersion {
@@ -172,21 +173,23 @@ func main() {
 	// starts, exactly like the other CLI subcommands: each loads config + settings, acts, and exits.
 	// R-379 (operator): the way OUT of a hold. A CLI subcommand rather than an API route, matching
 	// --abandon-stop above: the controller's HTTP surface is authenticated as the CUSTOMER, and
-	// clearing a hold is a decision that should follow an operator looking at the database. Reaching
+	// clearing a hold is a decision that should follow an operator looking at the data. Reaching
 	// this needs `docker exec` into the guest, which is the operator gate this project already uses.
+	// R-1343: the clear goes to the RUNNING controller over its local socket (holdclear.go); the list reads
+	// settings.json, which the running controller saves at every hold change.
 	if *listRestoreHolds || *clearRestoreHold != "" {
 		cfg, err := config.LoadPermissive(*configPath)
 		if err != nil {
 			fmt.Fprintf(os.Stderr, "restore-hold: loading config: %v\n", err)
 			os.Exit(1)
 		}
-		lg := log.New(os.Stderr, "", 0)
-		sett, err := settings.Load(cfg.Paths.DataDir+"/settings.json", lg)
-		if err != nil {
-			fmt.Fprintf(os.Stderr, "restore-hold: loading settings: %v\n", err)
-			os.Exit(1)
-		}
 		if *listRestoreHolds {
+			lg := log.New(os.Stderr, "", 0)
+			sett, err := settings.Load(cfg.Paths.DataDir+"/settings.json", lg)
+			if err != nil {
+				fmt.Fprintf(os.Stderr, "restore-hold: loading settings: %v\n", err)
+				os.Exit(1)
+			}
 			holds := sett.ListRestoreHolds()
 			if len(holds) == 0 {
 				fmt.Println("no restore holds in force")
@@ -197,34 +200,9 @@ func main() {
 			}
 			os.Exit(0)
 		}
-		cleared, err := sett.ClearRestoreHold(*clearRestoreHold)
-		if err != nil {
-			fmt.Fprintf(os.Stderr, "restore-hold: clearing %s: %v\n", *clearRestoreHold, err)
-			os.Exit(1)
-		}
-		if !cleared {
-			// "there was nothing to clear" is not success — a silent 0 here would let an operator
-			// believe they had unblocked an app whose name they mistyped.
-			fmt.Fprintf(os.Stderr, "no restore hold in force for %q — nothing was changed\n", *clearRestoreHold)
-			os.Exit(2)
-		}
-		// THE RESTART IS NOT OPTIONAL, and saying so is the difference between a route that works
-		// and one an operator believes worked. This runs as a SECOND process: it clears the hold in
-		// settings.json, but the RUNNING controller holds its own in-memory Settings and keeps
-		// refusing until it reloads. Measured on demo-hp 2026-08-22 — the clear succeeded, the file
-		// was correct, and the customer's start button still refused until the controller restarted.
-		//
-		// There is also a lost-update window while both processes hold the file: if the running
-		// controller saves settings.json from its own copy before the restart, the clear is undone.
-		// Restarting promptly closes it. A route that clears through the running controller would
-		// remove both problems and is the right shape later; it needs an operator tier the
-		// controller's HTTP surface does not have today (it authenticates as the CUSTOMER, and a
-		// customer clearing their own hold is precisely what the hold exists to prevent).
-		fmt.Printf("restore hold cleared for %s in settings.json.\n", *clearRestoreHold)
-		fmt.Printf("  NOW RESTART THE CONTROLLER, or it will keep refusing to start the app:\n")
-		fmt.Printf("    systemctl restart felhom-controller-bootstrap.service\n")
-		fmt.Printf("  Check the app's data first — the undo copies are in its unit's db-dumps dir.\n")
-		os.Exit(0)
+		// R-1343: through the RUNNING controller (holdclear.go) — the old second-process edit of settings.json was
+		// written back by the running controller's next save. Only when no controller answers is the file edited here.
+		os.Exit(clearRestoreHoldCLI(cfg.Paths.DataDir, *clearRestoreHold, os.Stdout, os.Stderr))
 	}
 
 	if *abandonStatus || *abandonExtend > 0 || *abandonStop {
@@ -485,6 +463,18 @@ func main() {
 		return stackMgr.DriveLive(hdd)
 	})
 	appStopRecovery := appStopGuard.Recover()
+	// D64 = B / R-1343: the operator's hold clear — ONE door for the hub's clear_hold and the box command, run in THIS
+	// process so memory and settings.json agree (holdclear.go). The box command reaches it over a unix socket in the
+	// data dir (0600, nothing on TCP). A socket that cannot be bound is logged; the command then says plainly that no
+	// running controller answered.
+	holdDeps := holdClearDeps{clear: appStopGuard.ClearHoldByOperator, forgetUpdateHeld: stackMgr.ForgetUpdateHeld, logger: logger}
+	if holdLn, herr := listenHoldCtl(holdCtlPath(cfg.Paths.DataDir), func(stack, who string) (string, error) {
+		return clearHoldNow(holdDeps, stack, who)
+	}, logger); herr != nil {
+		logger.Printf("[ERROR] [hold] the local hold-clear channel could not be opened: %v — `--clear-restore-hold` will edit settings.json and need a restart", herr)
+	} else {
+		defer holdLn.Close()
+	}
 	// R-1251: a put-back that waited for its drive (a hold from an earlier start) runs now if that drive is bound.
 	if started := appStopGuard.ResumeWaitingPutBacks(); len(started) > 0 {
 		logger.Printf("[WARN] [appstop] %d app(s) whose interrupted restore waited for its drive are put back and started: %v", len(started), started)
@@ -1081,6 +1071,7 @@ func main() {
 		holdReporter.Tick(time.Now())
 		return nil
 	})
+	oomLog := &oomLogGate{} // R-1341
 	sched.Every("deadapp-check", 30*time.Second, func(ctx context.Context) error {
 		if bootGrace.Within(time.Now()) {
 			return nil // still inside the startup settle window (90 s, or ~15 min after a crash boot)
@@ -1096,8 +1087,12 @@ func main() {
 		if oerr != nil {
 			logger.Printf("[WARN] [deadapp] OOM scan failed: %v", oerr)
 		} else {
-			for _, o := range ooms {
+			// R-1341: the line once per NEW kill (oomlog.go); the event call stays per scan — the notifier's cooldown
+			// keeps it to one, as before.
+			for _, o := range oomLog.fresh(ooms) {
 				logger.Printf("[WARN] [deadapp] %s: container %s was OOM-killed (started %s; kills %d, limit %s, peak %s; seen by %s)", o.Stack, o.Container, o.StartedAt, o.Kills, o.MemLimit, o.Peak, o.Source)
+			}
+			for _, o := range ooms {
 				notifier.NotifyAppOOMFrom(o.Stack, o.Container, o.StartedAt, o.Kills, o.MemLimit, o.Peak, o.Source)
 			}
 		}
@@ -1217,7 +1212,7 @@ func main() {
 		// R-314/R-279/R-177 (`09` §3 decision 185): the operator's closed list of actions, carried in the
 		// report reply and run in THIS process (opactions_wiring.go). The result rides the next report;
 		// a finished result fires an out-of-cycle report through the trigger once it exists.
-		opActions := report.NewOperatorActions(ctx, operatorActionHandlers(backupMgr, sched, func() {
+		opActions := report.NewOperatorActions(ctx, operatorActionHandlers(backupMgr, sched, holdDeps, func() {
 			if f, ok := opActionReportFire.Load().(func()); ok && f != nil {
 				f()
 			}
@@ -1548,22 +1543,18 @@ func main() {
 		// DELIBERATELY NOT `backup_failed`: customer-enabled by default, Hungarian "A biztonsági
 		// mentés sikertelen!", and this is a deliberate hold rather than a backup failure. R-171 one
 		// path over.
+		// R-1340: the text is built from the HOLD RECORD the writer persisted just before this call (holdmail.go) — its
+		// reason, which restore, which branch, which put-back step failed, whether it was a database replay — never the
+		// option-C database story for every cause.
 		backupMgr.SetRestoreHoldNotify(func(stack string, replayErr, rollbackErr error) {
-			d := notify.BackupRunFailuresDetails{RunKind: "offsite-reconstitute", Failed: 1, Attempted: 1}
-			if rollbackErr == nil { // R-893: the rollback WORKED, but files/volumes/version had already moved
-				d.Apps = append(d.Apps, notify.RunFailureDetail{App: stack, Leg: "restore-hold-mixed", Reason: "replay failed after files/volumes/version moved"})
-				msg := fmt.Sprintf("App %q is HELD STOPPED for support: its off-site database restore failed. The database was rolled back to the pre-restore copy and the live definition written back, "+
-					"but its files and named volumes are the snapshot's — a mixed state (R-893). The app will not start from any path until the hold is cleared. Replay error: %v", stack, replayErr)
-				notifier.NotifyBackupRunFailures(msg, d)
-				return
+			h := heldHoldFromNotify(sett.GetRestoreHold, stack, replayErr, rollbackErr)
+			msg, leg, reason := heldStoppedOperatorText(h)
+			runKind := "offsite-reconstitute"
+			if h.RestoreKind == settings.RestoreKindLocal {
+				runKind = "restore"
 			}
-			reason := "rollback failed"
-			if rollbackErr != nil {
-				reason = rollbackErr.Error()
-			}
-			d.Apps = append(d.Apps, notify.RunFailureDetail{App: stack, Leg: "restore-hold", Reason: reason})
-			msg := fmt.Sprintf("App %q is HELD STOPPED: its off-site database restore failed AND the rollback to the customer's own pre-restore copy also failed. "+
-				"The app will not start from any path until the hold is cleared. Replay error: %v. Rollback error: %v", stack, replayErr, rollbackErr)
+			d := notify.BackupRunFailuresDetails{RunKind: runKind, Failed: 1, Attempted: 1}
+			d.Apps = append(d.Apps, notify.RunFailureDetail{App: stack, Leg: leg, Reason: reason})
 			notifier.NotifyBackupRunFailures(msg, d)
 		})
 		// R-893 option A (D20 = A): a failed off-site restore that was PUT BACK — the live volumes and definition poured

controller/cmd/controller/oomlog.go

@@ -0,0 +1,43 @@
+package main
+
+import (
+	"strconv"
+
+	"gitea.dooplex.hu/admin/felhom-controller/internal/stacks"
+)
+
+// ── R-1341: the OOM line is logged once per NEW kill ──────────────────────────────────────────────────────────────
+//
+// ScanOOMKilled reports a container for as long as it lives after a kill (Docker's OOMKilled flag is sticky, and the
+// kernel's kill counter never goes down), so the 30-s dead-app scan logged „[WARN] [deadapp] … was OOM-killed" about
+// 2 880 times a day per container — measured 2026-10-29 on Tester 1 — while the notifier's cooldown kept the event to
+// one. The line now goes out when the kill is NEW: the key is (container, its start time, the kill count), and a line
+// is logged only when a container's key changes — a second scan of the same kill logs nothing, a new kill (the count
+// went up) or a restarted container (a new start time) logs again. The event call is unchanged (every scan, the
+// notifier decides). Pinned by TestR1341_*.
+
+// oomLogGate remembers the last logged key per container. Not safe for concurrent use: the dead-app job runs serially.
+type oomLogGate struct {
+	last map[string]string
+}
+
+func oomLogKey(o stacks.OOMContainer) string {
+	return o.StartedAt + "|" + strconv.FormatInt(o.Kills, 10)
+}
+
+// fresh returns the containers whose kill is new since the last scan, and remembers this scan's keys. A container
+// absent from this scan is forgotten (it was removed or restarted clean), which also bounds the memory.
+func (g *oomLogGate) fresh(ooms []stacks.OOMContainer) []stacks.OOMContainer {
+	next := make(map[string]string, len(ooms))
+	var out []stacks.OOMContainer
+	for _, o := range ooms {
+		k := oomLogKey(o)
+		next[o.Container] = k
+		if prev, ok := g.last[o.Container]; ok && prev == k {
+			continue
+		}
+		out = append(out, o)
+	}
+	g.last = next
+	return out
+}

controller/cmd/controller/oomlog_test.go

@@ -0,0 +1,64 @@
+package main
+
+import (
+	"testing"
+
+	"gitea.dooplex.hu/admin/felhom-controller/internal/stacks"
+)
+
+// R-1341: the „[WARN] [deadapp] … was OOM-killed" line goes out once per NEW kill, not every 30-s scan.
+//
+// RED-PROOF (2026-10-29): with oomLogGate.fresh returning its input unchanged (the pre-R-1341 loop logged every
+// container every scan), TestR1341_ASecondScanOfTheSameKillLogsNothing fails on „logged 1 line(s), want 0".
+
+func oom(container, started string, kills int64) stacks.OOMContainer {
+	return stacks.OOMContainer{Stack: "gokapi", Container: container, StartedAt: started, Kills: kills}
+}
+
+func TestR1341_ASecondScanOfTheSameKillLogsNothing(t *testing.T) {
+	g := &oomLogGate{}
+	scan := []stacks.OOMContainer{oom("gokapi", "2026-10-29T09:02:00Z", 1)}
+	if got := g.fresh(scan); len(got) != 1 {
+		t.Fatalf("the first sighting must log, logged %d line(s)", len(got))
+	}
+	for i := 0; i < 5; i++ {
+		if got := g.fresh(scan); len(got) != 0 {
+			t.Fatalf("scan %d of the same kill logged %d line(s), want 0", i+2, len(got))
+		}
+	}
+}
+
+func TestR1341_ANewKillLogsAgain(t *testing.T) {
+	g := &oomLogGate{}
+	g.fresh([]stacks.OOMContainer{oom("gokapi", "2026-10-29T09:02:00Z", 1)})
+	got := g.fresh([]stacks.OOMContainer{oom("gokapi", "2026-10-29T09:02:00Z", 2)})
+	if len(got) != 1 || got[0].Kills != 2 {
+		t.Fatalf("a new kill (count 1→2) must log once, got %+v", got)
+	}
+	if got := g.fresh([]stacks.OOMContainer{oom("gokapi", "2026-10-29T09:02:00Z", 2)}); len(got) != 0 {
+		t.Fatalf("the same new kill must not log twice, got %+v", got)
+	}
+}
+
+func TestR1341_ARestartedContainerLogsAgain(t *testing.T) {
+	g := &oomLogGate{}
+	g.fresh([]stacks.OOMContainer{oom("gokapi", "2026-10-29T09:02:00Z", 1)})
+	if got := g.fresh([]stacks.OOMContainer{oom("gokapi", "2026-10-29T09:30:00Z", 1)}); len(got) != 1 {
+		t.Fatalf("a restarted container's kill must log, got %+v", got)
+	}
+}
+
+// Two containers are judged apart; one that leaves the scan is forgotten, so its return logs again.
+func TestR1341_ContainersAreJudgedApart(t *testing.T) {
+	g := &oomLogGate{}
+	a, b := oom("a", "t1", 1), oom("b", "t1", 1)
+	if got := g.fresh([]stacks.OOMContainer{a, b}); len(got) != 2 {
+		t.Fatalf("first scan: %+v", got)
+	}
+	if got := g.fresh([]stacks.OOMContainer{a}); len(got) != 0 {
+		t.Fatalf("a seen again must not log: %+v", got)
+	}
+	if got := g.fresh([]stacks.OOMContainer{a, b}); len(got) != 1 || got[0].Container != "b" {
+		t.Fatalf("b came back after leaving the scan: %+v", got)
+	}
+}

controller/cmd/controller/opactions_wiring.go

@@ -17,18 +17,26 @@ import (
 // executor is built earlier than the trigger, so it reads it through this holder.
 var opActionReportFire atomic.Value
 
-// operatorActionHandlers wires the four operator actions (R-314/R-279/R-177, `09` §3 decision 185)
+// operatorActionHandlers wires the five operator actions (R-314/R-279/R-177, `09` §3 decision 185)
 // to the running process. ONE wiring site, called from main and from the wiring test, so the test
 // exercises the doors production uses (the seam-built-but-never-wired trap).
 //
 // offsite_backup_now makes the SAME four checks as the household's „back up now"
 // (offboxRunHandler) and then runs the same call, so an operator press can never do what the
 // household's own press cannot.
-func operatorActionHandlers(backupMgr *backup.Manager, sched *scheduler.Scheduler, resultReady func(), logger *log.Logger) report.OperatorActionHandlers {
+//
+// clear_hold (D64 = B) goes through clearHoldNow with hold — the same door the box command's socket uses
+// (holdclear.go), so the hub's button and the box command can never differ. hold.clear nil = not available.
+func operatorActionHandlers(backupMgr *backup.Manager, sched *scheduler.Scheduler, hold holdClearDeps, resultReady func(), logger *log.Logger) report.OperatorActionHandlers {
 	h := report.OperatorActionHandlers{ResultReady: resultReady}
 	if sched != nil {
 		h.RunJob = sched.RunNow
 	}
+	if hold.clear != nil {
+		h.ClearHold = func(stack string, actionID int64) (string, error) {
+			return clearHoldNow(hold, stack, fmt.Sprintf("the operator via the hub, action #%d", actionID))
+		}
+	}
 	if backupMgr == nil {
 		return h
 	}

controller/cmd/controller/opactions_wiring_test.go

@@ -57,7 +57,7 @@ func TestD1_AbandonStopActsOnTheRunningManagerAndDisk(t *testing.T) {
 		t.Fatal("fixture: a countdown must be running")
 	}
 
-	o := report.NewOperatorActions(context.Background(), operatorActionHandlers(mgr, nil, nil, logger), logger)
+	o := report.NewOperatorActions(context.Background(), operatorActionHandlers(mgr, nil, holdClearDeps{}, nil, logger), logger)
 	o.Reconcile([]report.OperatorAction{{ID: 1, Action: report.OpAbandonStop}})
 
 	if mgr.AbandonStatus().Active {
@@ -97,7 +97,7 @@ func TestD1_OffsiteBackupNowRefusesWithoutATarget(t *testing.T) {
 		t.Fatal("no restic may run")
 		return nil, nil
 	})
-	o := report.NewOperatorActions(context.Background(), operatorActionHandlers(mgr, nil, nil, logger), logger)
+	o := report.NewOperatorActions(context.Background(), operatorActionHandlers(mgr, nil, holdClearDeps{}, nil, logger), logger)
 	o.Reconcile([]report.OperatorAction{{ID: 9, Action: report.OpOffsiteBackupNow}})
 	if r := o.Results(); len(r) != 1 || r[0].Outcome != report.OutcomeRefused {
 		t.Fatalf("result = %+v", r)
@@ -115,7 +115,7 @@ func TestD1_RunJobUsesTheScheduler(t *testing.T) {
 	defer cancel()
 	sched.Start(ctx)
 	defer sched.Stop()
-	o := report.NewOperatorActions(ctx, operatorActionHandlers(nil, sched, nil, logger), logger)
+	o := report.NewOperatorActions(ctx, operatorActionHandlers(nil, sched, holdClearDeps{}, nil, logger), logger)
 	o.Reconcile([]report.OperatorAction{{ID: 1, Action: report.OpRunJob, Arg: "fill-watch"}, {ID: 2, Action: report.OpRunJob, Arg: "offsite-proof"}})
 	select {
 	case od := <-ran:

controller/internal/backup/appstop_putback_wait.go

@@ -119,7 +119,8 @@ func (g *AppStopGuard) holdPutBackFailed(stack, dir string, cause error) bool {
 		return false
 	}
 	h := settings.RestoreHold{Stack: stack, At: g.now().UTC().Format(time.RFC3339), Reason: settings.HoldReasonRestoreMixed,
-		ReplayError: "interrupted restore, put-back failed: " + cause.Error(), SafetyDump: dir}
+		ReplayError: "interrupted restore, put-back failed: " + cause.Error(), SafetyDump: dir,
+		MixedCause: settings.MixedCauseInterruptedPutBack, FailedStep: "putting the interrupted restore's copy back failed: " + cause.Error()}
 	if err := g.holds.SetRestoreHold(h); err != nil {
 		g.logger.Printf("[ERROR] [appstop] %s: the failed put-back could not be recorded as a hold: %v — the marker is kept", stack, err)
 		return false

controller/internal/backup/hold_clear.go

@@ -0,0 +1,66 @@
+package backup
+
+import (
+	"errors"
+
+	"gitea.dooplex.hu/admin/felhom-controller/internal/settings"
+)
+
+// ── D64 = B (operator ruling 2026-10-29 12:10) + R-1343: the operator clears a hold IN THE RUNNING CONTROLLER ──────
+//
+// One door for both of the operator's routes — the hub's „Clear hold" (operator action clear_hold) and the box command
+// `--clear-restore-hold`, which now asks the running controller over its local socket (cmd/controller/holdctl.go).
+//
+// WHY IN THIS PROCESS (R-1343, measured 2026-10-29 on Tester 1): the CLI used to run as a SECOND process that edited
+// settings.json while the running controller kept the hold in its own in-memory Settings — and that controller's next
+// save (after a hub report) wrote the hold back. Clearing through the running Settings removes the hold from memory
+// AND saves the file at once, so no later save of any component can bring it back.
+// Pinned by TestD64_ClearHoldStaysGoneAfterALaterSave (cmd/controller).
+//
+// EVERY REASON (D64 = B): the hold's reason does not matter — R-379 now reads „a person — on the box, or the operator
+// on the hub". One reason needs a second act: a put-back WAITING for its drive (HoldReasonPutBackPending) is also
+// recorded in the app-stop marker, and a controller start with the drive bound would pour that old copy back over
+// whatever the app wrote after the clear. So the clear also drops the app from the marker; the copy stays on disk
+// (nothing is deleted). It runs under resumeMu, so it can never interleave with a put-back of the same app.
+// Pinned by TestD64_ClearPutBackPendingDropsTheMarkerEntry.
+
+// ErrHoldBadStack: the argument is not a valid stack name. ErrNotHeld: the app has no hold. Both are refusals —
+// nothing was changed.
+var (
+	ErrHoldBadStack = errors.New("not a valid app name")
+	ErrNotHeld      = errors.New("this app has no hold")
+)
+
+// IsValidHoldStack reports whether s may name an app for the hold commands (the backup package's strict
+// isSafeStackName: letters, digits, '-' and '_', at most 64).
+func IsValidHoldStack(s string) bool { return isSafeStackName(s) }
+
+// ClearHoldByOperator lifts stack's hold, whatever its reason, in the hold store the running controller uses, and
+// returns the hold it lifted. A waiting put-back is dropped from the app-stop marker too (see above). Refusals
+// (ErrHoldBadStack, ErrNotHeld) change nothing.
+func (g *AppStopGuard) ClearHoldByOperator(stack string) (settings.RestoreHold, error) {
+	if !isSafeStackName(stack) {
+		return settings.RestoreHold{}, ErrHoldBadStack
+	}
+	if g == nil || g.holds == nil {
+		return settings.RestoreHold{}, errors.New("no hold store is wired on this box")
+	}
+	resumeMu.Lock()
+	defer resumeMu.Unlock()
+	h, ok := g.holds.GetRestoreHold(stack)
+	if !ok {
+		return settings.RestoreHold{}, ErrNotHeld
+	}
+	cleared, err := g.holds.ClearRestoreHold(stack)
+	if err != nil {
+		return h, err
+	}
+	if !cleared {
+		return settings.RestoreHold{}, ErrNotHeld
+	}
+	if h.Reason == settings.HoldReasonPutBackPending {
+		g.dropFromMarker(stack)
+		g.logger.Printf("[WARN] [appstop] %s: its waiting put-back is DROPPED with the hold — the copy stays on disk at %s and is never poured back by itself", stack, h.SafetyDump)
+	}
+	return h, nil
+}

controller/internal/backup/offbox_reconstitute.go

@@ -425,9 +425,12 @@ func (m *Manager) holdAppAfterFailedRollback(stack string, replayErr, rollbackEr
 		m.logger.Printf("[ERROR] [offbox] %s: cannot persist the restore hold — no settings wired; the app is stopped but NOTHING will refuse a restart", stack)
 		return
 	}
+	// R-1340: this branch is only ever an off-site restore's database replay (offbox_reconstitute's rollback).
 	h := settings.RestoreHold{
-		Stack: stack,
-		At:    time.Now().UTC().Format(time.RFC3339),
+		Stack:       stack,
+		At:          time.Now().UTC().Format(time.RFC3339),
+		RestoreKind: settings.RestoreKindOffsite,
+		DBReplay:    true,
 	}
 	if replayErr != nil {
 		h.ReplayError = replayErr.Error()
@@ -447,6 +450,15 @@ func (m *Manager) holdAppAfterFailedRollback(stack string, replayErr, rollbackEr
 	}
 }
 
+// mixedHoldWhy (R-1340) is what a restore_mixed hold records about its cause (settings.RestoreHold's R-1340 fields).
+type mixedHoldWhy struct {
+	kind     string // settings.RestoreKind*
+	cause    string // settings.MixedCause*
+	step     string // the step that failed, with its error
+	dbReplay bool   // the restore's own failure was its database replay
+	copyDir  string // where the copy of the live volumes is kept; "" = none kept
+}
+
 // liveDefinition is the app's definition as it ran before an off-site restore wrote the snapshot's
 // (R-893): a copy of its compose files in a temporary folder, and its full env held in memory only.
 type liveDefinition struct {
@@ -511,7 +523,11 @@ func (m *Manager) captureLiveDefinition(stack string) (*liveDefinition, error) {
 // written back (when the snapshot's was written), the database service is stopped again, and the app is
 // held with HoldReasonRestoreMixed. The operator is notified with no rollback error — that is how the
 // notification tells this case from R-379's double failure.
-func (m *Manager) holdAppAfterMixedRestore(stack string, replayErr error, live *liveDefinition) {
+//
+// why (R-1340) is what the hold records about its cause, so the operator's mail tells the real story — this hold is
+// written by three different branches now (a definition write, a put-back that failed at one of its steps), not
+// only by option C's database replay.
+func (m *Manager) holdAppAfterMixedRestore(stack string, replayErr error, live *liveDefinition, why mixedHoldWhy) {
 	// The hold is PERSISTED FIRST (security review 2026-10-08): a controller that dies between the stop and
 	// the hold would otherwise restart the app from the R-166 marker with nothing refusing it. When the hold
 	// cannot be persisted, the app-stop marker is KEPT, so nothing reads the window as finished.
@@ -519,7 +535,8 @@ func (m *Manager) holdAppAfterMixedRestore(stack string, replayErr error, live *
 	if m.settings == nil {
 		m.logger.Printf("[ERROR] [offbox] %s: cannot persist the restore hold — no settings wired; the app is stopped but NOTHING will refuse a restart", stack)
 	} else {
-		h := settings.RestoreHold{Stack: stack, At: time.Now().UTC().Format(time.RFC3339), Reason: settings.HoldReasonRestoreMixed}
+		h := settings.RestoreHold{Stack: stack, At: time.Now().UTC().Format(time.RFC3339), Reason: settings.HoldReasonRestoreMixed,
+			RestoreKind: why.kind, MixedCause: why.cause, FailedStep: why.step, DBReplay: why.dbReplay, SafetyDump: why.copyDir}
 		if replayErr != nil {
 			h.ReplayError = replayErr.Error()
 		}
@@ -557,7 +574,8 @@ func (m *Manager) restartAtLiveDefinition(stack string, live *liveDefinition, ca
 	if live != nil {
 		if err := m.stackProvider.RecreateStackDefinitionFromUnit(stack, live.dir, live.env); err != nil {
 			m.logger.Printf("[ERROR] [offbox] %s: the definition write failed (%v) and writing the live definition back failed too: %v — holding the app stopped", stack, cause, err)
-			m.holdAppAfterMixedRestore(stack, cause, nil)
+			m.holdAppAfterMixedRestore(stack, cause, nil, mixedHoldWhy{kind: settings.RestoreKindOffsite, cause: settings.MixedCauseDefinitionWrite,
+				step: "writing the live app definition back failed: " + err.Error()})
 			return
 		}
 		m.logger.Printf("[INFO] [offbox] %s: the definition write failed (%v); the live definition is written back before the start", stack, cause)
@@ -1060,7 +1078,7 @@ func (m *Manager) ReconstituteFromOffsite(ctx context.Context, stack string, ack
 			// R-893 option A: no volume has moved yet, so only the live definition is written back and the app starts.
 			// The failing copy may have placed part of its files, so the sentence names placed files (+1).
 			if definitionMoved {
-				return res, m.putBackAfterFailedRestore(stack, cErr, liveDef, undo, false, res.FilesPlaced+1, restartStack, "err.backup.restore_failed_held_mixed")
+				return res, m.putBackAfterFailedRestore(stack, cErr, liveDef, undo, false, res.FilesPlaced+1, restartStack, "err.backup.restore_failed_held_mixed", settings.RestoreKindOffsite)
 			}
 			// Nothing moved yet: best-effort bring-up — a failed restore must not also be an outage.
 			if sErr := restartStack(); sErr != nil {
@@ -1095,7 +1113,7 @@ func (m *Manager) ReconstituteFromOffsite(ctx context.Context, stack string, ack
 		// R-893 option A: the replay removes a volume BEFORE it recreates it, so even nVols == 0 may have lost one —
 		// with a copy, it is always poured back; with no copy the old rule decides (and holds when one moved).
 		if definitionMoved || nVols > 0 || undo != nil {
-			return res, m.putBackAfterFailedRestore(stack, vErr, liveDef, undo, nVols > 0 || undo != nil, res.FilesPlaced, restartStack, "err.backup.restore_failed_held_mixed")
+			return res, m.putBackAfterFailedRestore(stack, vErr, liveDef, undo, nVols > 0 || undo != nil, res.FilesPlaced, restartStack, "err.backup.restore_failed_held_mixed", settings.RestoreKindOffsite)
 		}
 		// Nothing replaced yet: bring the app back up rather than leaving an outage, then surface it.
 		if sErr := restartStack(); sErr != nil {
@@ -1114,7 +1132,7 @@ func (m *Manager) ReconstituteFromOffsite(ctx context.Context, stack string, ack
 		if err := m.stackProvider.StartStackServices(stack, dbServices); err != nil {
 			// R-893 option C (security review 2026-10-08): volumes or the definition already moved → HOLD.
 			if definitionMoved || res.VolumesReplayed > 0 {
-				return res, m.putBackAfterFailedRestore(stack, err, liveDef, undo, res.VolumesReplayed > 0, res.FilesPlaced, restartStack, "err.backup.restore_failed_held_mixed")
+				return res, m.putBackAfterFailedRestore(stack, err, liveDef, undo, res.VolumesReplayed > 0, res.FilesPlaced, restartStack, "err.backup.restore_failed_held_mixed", settings.RestoreKindOffsite)
 			}
 			// Best-effort bring-up: a failed restore must not also be an outage.
 			if sErr := restartStack(); sErr != nil {
@@ -1166,7 +1184,7 @@ func (m *Manager) ReconstituteFromOffsite(ctx context.Context, stack string, ack
 			// under them — the copy of the live volumes is poured back over the rolled-back database, the live
 			// definition written back, and the app starts exactly as it was. A put-back that fails holds, as before.
 			if definitionMoved || res.VolumesReplayed > 0 {
-				return res, m.putBackAfterFailedRestore(stack, iErr, liveDef, undo, res.VolumesReplayed > 0, res.FilesPlaced, restartStack, "err.backup.db_restore_failed_held_mixed")
+				return res, m.putBackAfterFailedRestore(stack, iErr, liveDef, undo, res.VolumesReplayed > 0, res.FilesPlaced, restartStack, "err.backup.db_restore_failed_held_mixed", settings.RestoreKindOffsite)
 			}
 			if sErr := restartStack(); sErr != nil {
 				m.logger.Printf("[WARN] [offbox] %s: start after a successful rollback failed: %v", stack, sErr)

controller/internal/backup/offbox_restore_undo.go

@@ -11,6 +11,7 @@ import (
 	"time"
 
 	"gitea.dooplex.hu/admin/felhom-controller/internal/dockerexec"
+	"gitea.dooplex.hu/admin/felhom-controller/internal/settings"
 	"gitea.dooplex.hu/admin/felhom-controller/internal/util"
 )
 
@@ -219,11 +220,23 @@ func (m *Manager) putBack(stack string, c *liveVolumeCopy) error {
 // failure's own held sentence) and the copy stays on disk. volumesMoved with no copy is such a case: nothing can
 // put them back. filesPlaced chooses the household's sentence: placed files are not taken back (as ruled for
 // option C, a file mix does not hold), so the sentence then says so instead of „nothing changed".
+//
+// restoreKind (R-1340) is settings.RestoreKindOffsite or RestoreKindLocal: the hold records it, with the step that
+// failed and whether the restore's failure was a database replay (holdKey says so — it is the callers' own choice of
+// the household's sentence), so the operator's mail names the real cause.
 func (m *Manager) putBackAfterFailedRestore(stack string, cause error, live *liveDefinition, c *liveVolumeCopy, volumesMoved bool,
-	filesPlaced int, restart func() error, holdKey string) error {
+	filesPlaced int, restart func() error, holdKey, restoreKind string) error {
+	why := func(step string) mixedHoldWhy {
+		w := mixedHoldWhy{kind: restoreKind, cause: settings.MixedCausePutBack, step: step,
+			dbReplay: holdKey == "err.backup.db_restore_failed_held_mixed"}
+		if c != nil && c.keep {
+			w.copyDir = c.dir
+		}
+		return w
+	}
 	if volumesMoved && c == nil {
 		m.logger.Printf("[ERROR] [offbox] %s: volumes were replaced and there is no copy to put back — holding the app", stack)
-		m.holdAppAfterMixedRestore(stack, cause, live)
+		m.holdAppAfterMixedRestore(stack, cause, live, why("the restore had already changed the app's data and there was no copy of the live volumes to put back"))
 		return util.MsgError(holdKey, stack)
 	}
 	// R-1256 H5: a put-back under an app that could not be stopped could mix the copy with what the app writes.
@@ -231,14 +244,14 @@ func (m *Manager) putBackAfterFailedRestore(stack string, cause error, live *liv
 	if err := m.stackProvider.StopStack(stack); err != nil {
 		c.keepIfAny()
 		m.logger.Printf("[ERROR] [offbox] %s: stopping the app before the put-back FAILED: %v — nothing is put back; the copy stays at %s; holding the app", stack, err, c.dirOrNone())
-		m.holdAppAfterMixedRestore(stack, cause, live)
+		m.holdAppAfterMixedRestore(stack, cause, live, why("stopping the app before the put-back failed, so nothing was put back: "+err.Error()))
 		return util.MsgError(holdKey, stack)
 	}
 	if volumesMoved {
 		if err := m.putBack(stack, c); err != nil {
 			c.keep = true
 			m.logger.Printf("[ERROR] [offbox] %s: putting the live volumes back FAILED: %v — the copy stays at %s; holding the app", stack, err, c.dir)
-			m.holdAppAfterMixedRestore(stack, cause, live)
+			m.holdAppAfterMixedRestore(stack, cause, live, why("putting the live volumes back failed: "+err.Error()))
 			return util.MsgError(holdKey, stack)
 		}
 	}
@@ -246,7 +259,7 @@ func (m *Manager) putBackAfterFailedRestore(stack string, cause error, live *liv
 		if err := m.stackProvider.RecreateStackDefinitionFromUnit(stack, live.dir, live.env); err != nil {
 			c.keepIfAny() // R-1256 H5 (D20): a put-back that fails keeps its copy for support
 			m.logger.Printf("[ERROR] [offbox] %s: writing the live definition back FAILED: %v — the copy stays at %s; holding the app", stack, err, c.dirOrNone())
-			m.holdAppAfterMixedRestore(stack, cause, live)
+			m.holdAppAfterMixedRestore(stack, cause, live, why("writing the live app definition back failed: "+err.Error()))
 			return util.MsgError(holdKey, stack)
 		}
 	}

controller/internal/backup/r1340_hold_cause_test.go

@@ -0,0 +1,133 @@
+package backup
+
+import (
+	"context"
+	"errors"
+	"io"
+	"log"
+	"os"
+	"strings"
+	"testing"
+
+	"gitea.dooplex.hu/admin/felhom-controller/internal/settings"
+)
+
+// R-1340: a restore_mixed hold records WHAT put it in place — which restore, which branch, which put-back step failed,
+// whether the restore's failure was a database replay, and where the copy is kept — so the operator's mail
+// (cmd/controller heldStoppedOperatorText, pinned by TestR1340_* there) tells the real story. Measured 2026-10-29 on
+// Tester 1: Radicale (no database), its volume could not be removed, the put-back's stop failed, and the mail said its
+// DATABASE restore failed and was rolled back.
+//
+// RED-PROOF (2026-10-29): with the R-1340 fields dropped from holdAppAfterMixedRestore's hold literal and from
+// holdPutBackFailed's (the pre-R-1340 record, which carried no cause), all five tests below fail on „mixed cause """ /
+// „kind """.
+
+func holdOf(t *testing.T, m *Manager, stack string) settings.RestoreHold {
+	t.Helper()
+	if m.settings == nil {
+		t.Fatal("fixture: no settings wired")
+	}
+	h, ok := m.settings.GetRestoreHold(stack)
+	if !ok {
+		t.Fatalf("%s is not held", stack)
+	}
+	return h
+}
+
+// The Radicale shape: a VOLUME restore fails (no database involved), then the put-back's stop fails.
+func TestR1340_AVolumeFailureWhosePutBackStopFailsRecordsTheStepAndNoDatabase(t *testing.T) {
+	m, vp, rig, _, notified, _ := r893PutBackFixture(t)
+	rig.snapReplayErr = errors.New("volume rm -f immich_pgdata: volume is in use")
+	p := &r1256StopFailProvider{r893Provider: vp, failAt: 2} // 1 = the restore's stop, 2 = the put-back's
+	m.SetStackProvider(p)
+	if _, err := m.ReconstituteFromOffsite(context.Background(), "immich", false); err == nil {
+		t.Fatal("a failed restore must be surfaced")
+	}
+	h := holdOf(t, m, "immich")
+	if h.Reason != settings.HoldReasonRestoreMixed || h.MixedCause != settings.MixedCausePutBack {
+		t.Fatalf("reason %q, mixed cause %q — want restore_mixed / put_back", h.Reason, h.MixedCause)
+	}
+	if !strings.Contains(h.FailedStep, "stopping the app before the put-back failed") || !strings.Contains(h.FailedStep, "compose down: timeout") {
+		t.Errorf("failed step %q — want the stop and its error", h.FailedStep)
+	}
+	if h.DBReplay {
+		t.Error("a volume failure must not be recorded as a database replay")
+	}
+	if h.RestoreKind != settings.RestoreKindOffsite {
+		t.Errorf("restore kind %q, want offsite", h.RestoreKind)
+	}
+	if h.SafetyDump == "" {
+		t.Error("the kept copy's folder must be recorded")
+	} else if _, err := os.Stat(h.SafetyDump); err != nil {
+		t.Errorf("the recorded copy %s is not on disk: %v", h.SafetyDump, err)
+	}
+	if *notified != 1 {
+		t.Errorf("hold notices %d, want 1", *notified)
+	}
+}
+
+// The database replay fails (option C's shape), then the put-back's stop fails: DBReplay is recorded.
+func TestR1340_ADatabaseReplayFailureRecordsTheDatabase(t *testing.T) {
+	m, vp, _, _, _, _ := r893PutBackFixture(t)
+	m.SetStackProvider(&r1256StopFailProvider{r893Provider: vp, failAt: 2})
+	if _, err := m.ReconstituteFromOffsite(context.Background(), "immich", false); err == nil {
+		t.Fatal("a failed restore must be surfaced")
+	}
+	h := holdOf(t, m, "immich")
+	if h.MixedCause != settings.MixedCausePutBack || !h.DBReplay {
+		t.Fatalf("mixed cause %q, db replay %v — want put_back and true", h.MixedCause, h.DBReplay)
+	}
+}
+
+// R-1256 H4: the backup's definition cannot be written and the live one cannot be written back — no data moved.
+func TestR1340_ADefinitionWriteFailureRecordsItsCause(t *testing.T) {
+	m, _, _ := r1256DefFailFixture(t, 2)
+	if _, err := m.ReconstituteFromOffsite(context.Background(), "immich", false); err == nil {
+		t.Fatal("a failed definition write must be surfaced")
+	}
+	h := holdOf(t, m, "immich")
+	if h.MixedCause != settings.MixedCauseDefinitionWrite || h.DBReplay {
+		t.Fatalf("mixed cause %q, db replay %v — want definition_write and false", h.MixedCause, h.DBReplay)
+	}
+	if !strings.Contains(h.FailedStep, "writing the live app definition back failed") {
+		t.Errorf("failed step %q", h.FailedStep)
+	}
+}
+
+// A LOCAL restore (from the app's own backup) records itself as local, never off-site.
+func TestR1340_ALocalRestoreRecordsItsKind(t *testing.T) {
+	m, _, rig, _ := r1177Fixture(t)
+	if m.settings == nil {
+		sett, err := settings.Load(t.TempDir()+"/settings.json", log.New(io.Discard, "", 0))
+		if err != nil {
+			t.Fatal(err)
+		}
+		m.settings = sett
+	}
+	rig.unitErr = errors.New("tar: short read")
+	rig.putBackErr = errors.New("tar xf failed")
+	if _, err := m.RestoreFromRecoveryUnit("app"); err == nil {
+		t.Fatal("a failed put-back must be surfaced")
+	}
+	h := holdOf(t, m, "app")
+	if h.RestoreKind != settings.RestoreKindLocal || h.MixedCause != settings.MixedCausePutBack || h.DBReplay {
+		t.Fatalf("kind %q, cause %q, db %v — want local / put_back / false", h.RestoreKind, h.MixedCause, h.DBReplay)
+	}
+	if !strings.Contains(h.FailedStep, "putting the live volumes back failed") || !strings.Contains(h.FailedStep, "tar xf failed") {
+		t.Errorf("failed step %q", h.FailedStep)
+	}
+}
+
+// A restore cut by a stop whose put-back fails at the next start (R-1251 H2) records its own cause.
+func TestR1340_AnInterruptedPutBackRecordsItsCause(t *testing.T) {
+	r := newR1251Rig(t)
+	r.putErr = errors.New("compose down: timeout")
+	r.guard().Recover()
+	h, ok := r.sett.GetRestoreHold("immich")
+	if !ok || h.Reason != settings.HoldReasonRestoreMixed || h.MixedCause != settings.MixedCauseInterruptedPutBack {
+		t.Fatalf("hold %+v (held %v) — want restore_mixed / interrupted_put_back", h, ok)
+	}
+	if !strings.Contains(h.FailedStep, "compose down: timeout") {
+		t.Errorf("failed step %q", h.FailedStep)
+	}
+}

controller/internal/backup/restore_unit.go

@@ -3,6 +3,7 @@ package backup
 import (
 	"errors"
 	"fmt"
+	"gitea.dooplex.hu/admin/felhom-controller/internal/settings"
 	"gitea.dooplex.hu/admin/felhom-controller/internal/util"
 	"os"
 	"path/filepath"
@@ -530,7 +531,7 @@ func (m *Manager) RestoreFromRecoveryUnitAtWith(stackName, unitDir string, opt U
 	// without one (the unit holds no volume) only a database replay that began can have changed data — and with no
 	// copy to pour back that is the R-379 half state: held, never started (operator ruling 2026-08-22).
 	putBack := func(cause error, holdKey string, moved bool) (UnitRestoreResult, error) {
-		return res, m.putBackAfterFailedRestore(stackName, cause, liveDef, undo, moved || undo != nil, opt.FilesPlaced, restartStack, holdKey)
+		return res, m.putBackAfterFailedRestore(stackName, cause, liveDef, undo, moved || undo != nil, opt.FilesPlaced, restartStack, holdKey, settings.RestoreKindLocal)
 	}
 	// R-353: the count is captured even when the replay errors — a partial replay is a fact the
 	// customer's sentence has to be built from, and discarding it on the error path is how Scenario C

controller/internal/report/opactions.go

@@ -21,9 +21,11 @@ import (
 // problem, main.go). The result rides the next report as `operator_action_results: [{id, outcome,
 // message}]`; the hub lists an action until its result has arrived, then stops.
 //
-// THE LIST IS CLOSED. Four actions, and run_job names one of four jobs. Nothing here deletes data,
+// THE LIST IS CLOSED. Five actions, and run_job names one of four jobs. Nothing here deletes data,
 // starts a countdown or shortens one — `03` §4 asks for a signing key only to destroy or overwrite the
-// only copy, and none of these does. An unknown action, an unknown job or an argument out of range is
+// only copy, and none of these does. clear_hold (D64 = B, operator ruling 2026-10-29 12:10) lifts an app's
+// hold of ANY reason with no operator signature: it deletes nothing and starts nothing — the app stays
+// stopped until the household (or the operator) starts it, and a waiting put-back's copy stays on disk. An unknown action, an unknown job or an argument out of range is
 // answered `refused` and NOTHING is called. Pinned by TestOpActions_ClosedList (which fails when an
 // entry is added, so a new action is a decision, not an edit) and TestOpActions_UnknownIsRefused.
 //
@@ -38,6 +40,7 @@ const (
 	OpAbandonStop      = "abandon_stop"
 	OpAbandonExtend    = "abandon_extend"
 	OpRunJob           = "run_job"
+	OpClearHold        = "clear_hold"
 
 	OutcomeDone    = "done"
 	OutcomeRefused = "refused"
@@ -57,7 +60,7 @@ const (
 )
 
 // operatorActionNames is THE closed list. Adding an entry fails TestOpActions_ClosedList by design.
-var operatorActionNames = []string{OpOffsiteBackupNow, OpAbandonStop, OpAbandonExtend, OpRunJob}
+var operatorActionNames = []string{OpOffsiteBackupNow, OpAbandonStop, OpAbandonExtend, OpRunJob, OpClearHold}
 
 // operatorJobNames is the fixed set run_job may name — the scheduler's own job names (main.go).
 var operatorJobNames = []string{"fill-watch", "offsite-integrity", "offsite-proof", "disk-health-check"}
@@ -83,6 +86,21 @@ func contains(list []string, v string) bool {
 	return false
 }
 
+// validHoldStack is clear_hold's argument check: the same alphabet as backup.isSafeStackName (report cannot import
+// backup; REUSE.md „Safe-name validators").
+func validHoldStack(s string) bool {
+	if s == "" || len(s) > 64 {
+		return false
+	}
+	for _, c := range s {
+		if (c >= 'a' && c <= 'z') || (c >= 'A' && c <= 'Z') || (c >= '0' && c <= '9') || c == '-' || c == '_' {
+			continue
+		}
+		return false
+	}
+	return true
+}
+
 // ErrActionRefused, wrapped by a handler, turns its error into a `refused` outcome (the action could
 // not START — e.g. an off-site run already in flight) instead of `failed` (it started and went wrong).
 var ErrActionRefused = errors.New("refused")
@@ -101,7 +119,7 @@ type OperatorActionResult struct {
 	Message string `json:"message"`
 }
 
-// OperatorActionHandlers are the four doors, wired by main.go. A nil handler answers `refused`
+// OperatorActionHandlers are the five doors, wired by main.go. A nil handler answers `refused`
 // („not available on this box").
 type OperatorActionHandlers struct {
 	// OffsiteBackupNow returns a refusal sentence when the run may not start (the household's
@@ -111,6 +129,9 @@ type OperatorActionHandlers struct {
 	AbandonExtend    func(days int) (time.Time, error)
 	// RunJob is Scheduler.RunNow: an error = not started (unknown/running); the channel = the job's end.
 	RunJob func(name string) (<-chan error, error)
+	// ClearHold lifts the app's hold in the RUNNING controller (D64 = B, R-1343) and returns the result
+	// sentence. actionID names the action in the box's log. An error wrapping ErrActionRefused = nothing changed.
+	ClearHold func(stack string, actionID int64) (string, error)
 	// ResultReady sends an out-of-cycle report so a result reaches the hub in seconds (optional).
 	ResultReady func()
 }
@@ -317,6 +338,23 @@ func (o *OperatorActions) dispatch(a OperatorAction) {
 			}
 			o.finish(a, outcomeOf(jerr), msg)
 		}()
+	case OpClearHold:
+		// The argument is checked HERE, before any handler is touched (the closed list's rule): one app name,
+		// the backup package's strict alphabet (letters, digits, '-' and '_', at most 64).
+		if !validHoldStack(a.Arg) {
+			o.finish(a, OutcomeRefused, "the argument must be one app name (letters, digits, '-' and '_')")
+			return
+		}
+		if o.h.ClearHold == nil {
+			o.finish(a, OutcomeRefused, "not available on this box")
+			return
+		}
+		msg, err := o.h.ClearHold(a.Arg, a.ID)
+		if err != nil {
+			o.finish(a, outcomeOf(err), err.Error())
+			return
+		}
+		o.finish(a, OutcomeDone, msg)
 	default:
 		o.finish(a, OutcomeRefused, "unknown action")
 	}

controller/internal/report/opactions_test.go

@@ -4,6 +4,7 @@ import (
 	"context"
 	"encoding/json"
 	"errors"
+	"fmt"
 	"reflect"
 	"strings"
 	"sync"
@@ -15,9 +16,11 @@ import (
 
 // The list is CLOSED. This test fails when an action or a job is added, by design: a new entry is an
 // operator decision (no action may delete data, start a countdown or shorten one), not an edit. The
-// hub pins the same four in its own test (hub internal/store opactions_test.go).
+// hub pins the same list in its own test (hub internal/store opactions_test.go).
+// 2026-10-29: clear_hold added deliberately — operator ruling D64 = B (12:10): the hub's „Clear hold" works for
+// every hold reason with no operator signature. It deletes nothing and starts nothing.
 func TestOpActions_ClosedList(t *testing.T) {
-	if got, want := OperatorActionNames(), []string{"abandon_extend", "abandon_stop", "offsite_backup_now", "run_job"}; !reflect.DeepEqual(got, want) {
+	if got, want := OperatorActionNames(), []string{"abandon_extend", "abandon_stop", "clear_hold", "offsite_backup_now", "run_job"}; !reflect.DeepEqual(got, want) {
 		t.Fatalf("operator actions = %v, want exactly %v — a new action needs the operator's word (decision 185)", got, want)
 	}
 	if got, want := OperatorJobNames(), []string{"disk-health-check", "fill-watch", "offsite-integrity", "offsite-proof"}; !reflect.DeepEqual(got, want) {
@@ -56,6 +59,10 @@ func recordingHandlers(c *calls) OperatorActionHandlers {
 			close(ch)
 			return ch, nil
 		},
+		ClearHold: func(stack string, id int64) (string, error) {
+			c.add("clear:" + stack)
+			return "the hold on " + stack + " is cleared", nil
+		},
 	}
 }
 
@@ -197,3 +204,73 @@ func TestOpActions_WireShapes(t *testing.T) {
 		t.Fatalf("report carried %s", b)
 	}
 }
+
+// clear_hold (D64 = B): the argument is one app name in the strict alphabet — never a path, never empty, never two
+// names. A bad one is refused and the handler is NOT called (the closed list's rule: validate before any door).
+// RED-PROOF (2026-10-29): with validHoldStack returning true, this fails on „a refused action called [clear:…]".
+func TestOpActions_ClearHoldBadArgumentIsRefusedAndCallsNothing(t *testing.T) {
+	c := &calls{}
+	o := NewOperatorActions(context.Background(), recordingHandlers(c), nil)
+	list := []OperatorAction{
+		{ID: 1, Action: OpClearHold, Arg: ""},
+		{ID: 2, Action: OpClearHold, Arg: "../settings"},
+		{ID: 3, Action: OpClearHold, Arg: "immich paperless"},
+		{ID: 4, Action: OpClearHold, Arg: "immich/x"},
+		{ID: 5, Action: OpClearHold, Arg: strings.Repeat("a", 65)},
+		{ID: 6, Action: OpClearHold, Arg: "immich\n"},
+		{ID: 7, Action: OpClearHold, Arg: "."},
+	}
+	o.Reconcile(list)
+	if got := c.get(); len(got) != 0 {
+		t.Fatalf("a refused action called %v", got)
+	}
+	for _, a := range list {
+		if r := resultFor(t, o, a.ID); r.Outcome != OutcomeRefused || !strings.Contains(r.Message, "one app name") {
+			t.Errorf("#%d clear_hold(%q): %+v, want refused naming the rule", a.ID, a.Arg, r)
+		}
+	}
+	// control: a good name does reach the handler
+	o.Reconcile([]OperatorAction{{ID: 8, Action: OpClearHold, Arg: "paperless-ngx_2"}})
+	if got := c.get(); len(got) != 1 || got[0] != "clear:paperless-ngx_2" {
+		t.Fatalf("a valid name must reach the handler, calls %v", got)
+	}
+}
+
+// clear_hold (D64 = B): a valid name reaches the handler with the action's id, and its sentence is the result; a
+// handler refusal (the app has no hold) is answered refused, an error that is not a refusal failed.
+// RED-PROOF (2026-10-29): with the OpClearHold case removed from dispatch, this fails on „unknown action".
+func TestOpActions_ClearHoldDoneRefusedFailed(t *testing.T) {
+	var gotStack string
+	var gotID int64
+	h := OperatorActionHandlers{ClearHold: func(stack string, id int64) (string, error) {
+		gotStack, gotID = stack, id
+		switch stack {
+		case "notheld":
+			return "", fmt.Errorf("%w: notheld has no hold — nothing was changed", ErrActionRefused)
+		case "broken":
+			return "", errors.New("saving settings: disk full")
+		}
+		return "the hold on " + stack + " is cleared (it was: held after a failed update)", nil
+	}}
+	o := NewOperatorActions(context.Background(), h, nil)
+	o.Reconcile([]OperatorAction{{ID: 7, Action: OpClearHold, Arg: "kimai"}})
+	if gotStack != "kimai" || gotID != 7 {
+		t.Fatalf("handler called with (%q, %d), want (kimai, 7)", gotStack, gotID)
+	}
+	if r := resultFor(t, o, 7); r.Outcome != OutcomeDone || r.Message != "the hold on kimai is cleared (it was: held after a failed update)" {
+		t.Fatalf("result %+v", r)
+	}
+	o.Reconcile([]OperatorAction{{ID: 8, Action: OpClearHold, Arg: "notheld"}, {ID: 9, Action: OpClearHold, Arg: "broken"}})
+	if r := resultFor(t, o, 8); r.Outcome != OutcomeRefused || !strings.Contains(r.Message, "has no hold") {
+		t.Errorf("not held: %+v", r)
+	}
+	if r := resultFor(t, o, 9); r.Outcome != OutcomeFailed {
+		t.Errorf("a save error must be failed, not refused: %+v", r)
+	}
+	// nil handler → refused, not a crash
+	o2 := NewOperatorActions(context.Background(), OperatorActionHandlers{}, nil)
+	o2.Reconcile([]OperatorAction{{ID: 1, Action: OpClearHold, Arg: "kimai"}})
+	if r := resultFor(t, o2, 1); r.Outcome != OutcomeRefused {
+		t.Errorf("nil handler: %+v", r)
+	}
+}

controller/internal/settings/settings.go

@@ -1982,8 +1982,36 @@ type RestoreHold struct {
 	// first stop or a repeat within 24 h (Trip >= 2 → the sentence says support is informed).
 	UnhealthyKind string `json:"unhealthy_kind,omitempty"`
 	Trip          int    `json:"trip,omitempty"`
+
+	// R-1340 (2026-10-29): what put a restore hold in place, so the operator's „HELD STOPPED" mail tells the real
+	// story instead of the option-C database story for every cause. All empty on a hold written before them: the mail
+	// then names only what the record carries (cmd/controller heldStoppedOperatorText).
+	//
+	// RestoreKind is which restore: RestoreKindOffsite or RestoreKindLocal.
+	RestoreKind string `json:"restore_kind,omitempty"`
+	// MixedCause (restore_mixed only) is the branch that held the app: MixedCause*.
+	MixedCause string `json:"mixed_cause,omitempty"`
+	// FailedStep is the put-back step that failed, with its error (MixedCausePutBack).
+	FailedStep string `json:"failed_step,omitempty"`
+	// DBReplay: the restore's own failure was its database replay. False = the app's database was never touched
+	// by the failure, so the mail never says „database".
+	DBReplay bool `json:"db_replay,omitempty"`
 }
 
+// Restore kinds and mixed-hold causes (R-1340). See RestoreHold.RestoreKind / MixedCause.
+const (
+	RestoreKindOffsite = "offsite"
+	RestoreKindLocal   = "local"
+
+	// MixedCauseDefinitionWrite: the backup's definition could not be written (no data moved yet) and writing the
+	// live definition back failed too (R-1256 H4).
+	MixedCauseDefinitionWrite = "definition_write"
+	// MixedCausePutBack: the restore failed and putting the live data back failed at FailedStep (D20/R-1256 H5).
+	MixedCausePutBack = "put_back"
+	// MixedCauseInterruptedPutBack: a restore cut by a stop could not be put back at the next start (R-1251 H2).
+	MixedCauseInterruptedPutBack = "interrupted_put_back"
+)
+
 // Hold reasons. See RestoreHold.Reason.
 const (
 	HoldReasonRestoreFailed = ""              // R-379/R-380: a restore AND its rollback failed

controller/internal/stacks/d64_forget_update_held_test.go

@@ -0,0 +1,46 @@
+package stacks
+
+import (
+	"io"
+	"log"
+	"testing"
+
+	"gitea.dooplex.hu/admin/felhom-controller/internal/config"
+)
+
+// D64 = B / R-1343: when the operator clears a hold, the ONE held flag kept outside the hold store — „the last update
+// ended held" — goes too. Left set, the crash-loop observer and the nightly update leg kept skipping an app whose hold
+// was gone, until the controller restarted.
+// RED-PROOF (2026-10-29): with ForgetUpdateHeld's body reduced to `return false`, this fails on „a held update must be forgotten".
+func TestD64_ForgetUpdateHeld(t *testing.T) {
+	m := &Manager{cfg: &config.Config{}, logger: log.New(io.Discard, "", 0), stacks: map[string]*Stack{
+		"kimai":   {Name: "kimai", Deployed: true},
+		"running": {Name: "running", Deployed: true, Updating: true},
+	}}
+	m.finishUpdateKey("kimai", UpdatePhaseFailed, UpdateErrorKeyHeld, "a kimai leállítva marad")
+	m.markUpdateHeld("kimai")
+	m.markUpdateHeld("running")
+
+	if !m.ForgetUpdateHeld("kimai") {
+		t.Fatal("a held update must be forgotten")
+	}
+	st, _ := m.GetStack("kimai")
+	if st.updateHeld || st.UpdateErrorKey != "" || st.UpdateError != "" || st.UpdatePhase != "" {
+		t.Fatalf("held state left: held=%v key=%q err=%q phase=%q", st.updateHeld, st.UpdateErrorKey, st.UpdateError, st.UpdatePhase)
+	}
+	// The two readers of the flag (ObserveUnhealthy's skip, unhealthy.go; legCandidate's LegSkipHeld, unattended.go)
+	// read the stored stack — not called here, because ObserveUnhealthy reaches docker (R-650).
+	m.mu.RLock()
+	skipped := m.stacks["kimai"].updateHeld || m.stacks["kimai"].UpdateErrorKey == UpdateErrorKeyHeld
+	m.mu.RUnlock()
+	if skipped {
+		t.Fatal("the observer and the update leg still skip kimai")
+	}
+	// A running update is never touched; an app with nothing held reports false.
+	if m.ForgetUpdateHeld("running") {
+		t.Fatal("a running update must not be touched")
+	}
+	if m.ForgetUpdateHeld("kimai") || m.ForgetUpdateHeld("nosuch") {
+		t.Fatal("nothing held → false")
+	}
+}

controller/internal/stacks/update.go

@@ -209,6 +209,25 @@ func (m *Manager) ClearUpdateState(name string) {
 	s.HealthProbe = nil
 }
 
+// ForgetUpdateHeld (D64 = B, R-1343) forgets that the app's last update ended HELD, once the hold itself is cleared
+// by the operator. updateHeld is the ONE held flag kept outside the hold store: it makes the crash-loop observer
+// (ObserveUnhealthy) and the nightly update leg (legCandidate) skip the app, and keeps the hold's sentence as the
+// update error. Left set, an app whose hold was cleared would stay unobserved and never be updated by night until
+// the controller restarts. A running update is not touched. Returns whether anything was forgotten.
+// Pinned by TestD64_ForgetUpdateHeld (this package) and TestD64_ClearHoldForgetsTheUpdateHeldFlag (cmd/controller).
+func (m *Manager) ForgetUpdateHeld(name string) bool {
+	m.mu.Lock()
+	defer m.mu.Unlock()
+	s, ok := m.stacks[name]
+	if !ok || s.Updating || (!s.updateHeld && s.UpdateErrorKey != UpdateErrorKeyHeld) {
+		return false
+	}
+	s.updateHeld = false
+	s.UpdatePhase, s.UpdatePhaseLabel = "", ""
+	s.UpdateError, s.UpdateErrorKey, s.UpdateErrorArgs = "", "", nil
+	return true
+}
+
 func (m *Manager) markUpdateHeld(name string) {
 	m.mu.Lock()
 	if s, ok := m.stacks[name]; ok {