R-351: the restore compares where the backup says the data lived; second press cannot start a second run

Part 3 (not droppable) and the engine half of Part 2. No version bump yet - one bump and
one bake at the end of the session.

PART 3a - a second press really did start a second run. Established with a test BEFORE any
change: both offboxReconstituteHandler and offboxPlaceHandler answered "...elindult" and
overwrote the first restore's op/stack. Cause: every restore handler gated on
backupMgr.IsRunning() - the CONCURRENCY flag, which the restore goroutine acquires AFTER the
handler returns (offbox_reconstitute.go:180, offbox_restore.go:393). Seven sites. The wizard
had read the correct flag since v0.154.0 and said so in a comment; the handlers never moved.
New Server.restoreOpBlocked() reads BOTH flags - the display flag covers the whole off-box
restore, the concurrency flag is the only one the nightly backup holds - and the refusal now
names the running app and a route.

PART 3b - the page DOES refresh; the defect was the RESULT. backups_shared.html gated the
terminal result on a page-local sawRunning flag, so a restore that finished before the page
was opened, or inside one 3s poll, was shown to nobody. The 2026-08-21 OpenGist restore took
8.666s and no screen ever said it completed - the answer existed only in docker logs.
RestoreOpStatus.LastRecent now carries the server's verdict. The 10-minute window moved to
internal/backup as RestoreResultWindow and internal/web's constant is an alias: one
expression, two surfaces. Also removed the wizard's self-contradiction, which said the state
refreshes automatically AND that you must refresh the page.

PART 2 (engine) - every recovery unit manifest has carried drive and namespace_root since
schema 1, and NO non-test code read either back. The reconstitution opened the manifest and
took only the coherence stamp, then resolved its destination from the live app. A restore
into a different destination succeeded silently under a green message. New
backup/offbox_placement.go: CheckPlacement (pure, total), PlacementMismatchMessage,
recordedPlacementFromScratch. Compared before the safety dump and before the first byte.
A mismatch is NAMED and refused; ackPlacementChange lets the customer proceed deliberately -
a separate field from confirm=1, because one click must not carry two decisions. An UNKNOWN
recording is never a mismatch: refusing on an absence would strand every pre-field unit.
The not-installed refusal (R-253) now names the drive the backup recorded.

RED-PROOFS, each mutation asserted applied and reverted to 0:
  B  both guards removed (count asserted 2) -> the restore WAS seen starting with no drive
     attached: no error, full 3.00s run, wrote into /tmp/mutant-destination
  C  Mismatch forced false -> the silent divergent restore returned
  E  Known() forced true  -> the fabricated empty prefill appeared
  D  Mismatch forced true -> 8 ordinary reconstitute tests broke, proving reachability both ways
Note on D: the existing fixtures write a schema-1 manifest with NO drive, so they are
scenario-E shaped. The matching case is covered in the scenario table, not by them.

Gates 11/11 OK. Suite 28 packages ok. Hungarian verified as hex, no BOM, no mojibake sentinels.

NOT in this commit, still open: Part 2's scenario-A prefill UI, Part 1's deploy-page
visibility line, Part 1's specification document, Part 4's measurement.

controller/internal/backup/offbox_placement.go

@@ -0,0 +1,146 @@
+package backup
+
+import (
+	"fmt"
+	"io/fs"
+	"os"
+	"path/filepath"
+	"strings"
+)
+
+// R-351 — THE RESTORE ALREADY KNOWS WHERE THE APP LIVED. IT JUST NEVER LOOKED.
+//
+// Every recovery unit's manifest.json carries `drive` and `namespace_root` (recovery_unit.go:48-49),
+// written at capture time from the app's own live placement. Measured on demo-hp 2026-08-21:
+//
+//	opengist      drive=/mnt/sys_drive             nsroot=/mnt/sys_drive/felhom-data
+//	calibre-web   drive=/mnt/felhom-drives/hdd_1   nsroot=/mnt/felhom-drives/hdd_1
+//
+// Before this file, NO non-test code in the repository read either field back. `grep -rE
+// '\.Drive\b|\.NamespaceRoot\b' --include=*.go` returned only the appbackup.NamespaceRoot FUNCTION
+// and Tier2Target's unrelated field. The reconstitution opened the manifest
+// (offbox_reconstitute.go:235) and took only the coherence stamp from it, then resolved its
+// destination from the LIVE app instead. One side wrote the fact; the other never received it —
+// the same shape as several defects closed this month.
+//
+// THE CONSEQUENCE THAT MADE THIS WORTH FIXING: a restore into a destination that differs from the
+// one the backup recorded succeeded SILENTLY, under a green message. Nothing compared them.
+//
+// WHAT THIS IS NOT. It is not a lock. A domain can legitimately change and hardware can move, so a
+// difference is NAMED and the customer decides — it is their own previous answer being shown back to
+// them, not a rule imposed on them. What must never happen is the difference passing unremarked.
+
+// RecordedPlacement is what a backup says about where an app's data lived when it was captured.
+// Empty fields mean the manifest did not record them — an older unit, or one written before the
+// field existed. That is an UNKNOWN and is never rendered as a value.
+type RecordedPlacement struct {
+	Drive         string // manifest.Drive — the in-guest mount (HDD_PATH), or the system data path
+	NamespaceRoot string // manifest.NamespaceRoot — the resolved felhom-data namespace root
+}
+
+// Known reports whether the backup recorded a destination at all. A unit whose manifest predates the
+// field, or could not be read, is NOT known — and "we cannot tell" is a different answer from "they
+// match", which is why this is a method rather than a `!= ""` scattered over the callers.
+func (p RecordedPlacement) Known() bool { return strings.TrimSpace(p.Drive) != "" }
+
+// PlacementCheck is the verdict of comparing what the backup recorded against where the restore is
+// actually about to write.
+type PlacementCheck struct {
+	// Recorded is what the backup said. Zero value when the manifest carried nothing.
+	Recorded RecordedPlacement
+	// LiveDrive / LiveNamespaceRoot are where this restore will write, resolved the way the
+	// reconstitution resolves it today.
+	LiveDrive         string
+	LiveNamespaceRoot string
+	// Known mirrors Recorded.Known(), carried on the verdict so a template never has to re-derive it.
+	Known bool
+	// Mismatch is true ONLY when the backup recorded a destination AND it differs from the live one.
+	// An unknown recording is never a mismatch: refusing on an absence would block every pre-field
+	// unit, and asserting a match we cannot see would be worse.
+	Mismatch bool
+}
+
+// CheckPlacement compares a manifest's recorded placement against the live destination.
+//
+// Deliberately PURE and total: a nil manifest, an empty manifest and a manifest whose drive is blank
+// all produce the same honest "not known, no mismatch" verdict. Paths are compared Cleaned, because
+// `/mnt/x` and `/mnt/x/` are the same destination and a trailing slash must not manufacture a
+// mismatch the customer then has to dismiss. Comparison is on the DRIVE, not the namespace root: the
+// root is derived from the drive (namespaceRoot appends felhom-data only on the system-data
+// fallback), so comparing both would report one difference twice.
+func CheckPlacement(man *RecoveryManifest, liveDrive, liveNamespaceRoot string) PlacementCheck {
+	c := PlacementCheck{
+		LiveDrive:         strings.TrimSpace(liveDrive),
+		LiveNamespaceRoot: strings.TrimSpace(liveNamespaceRoot),
+	}
+	if man != nil {
+		c.Recorded = RecordedPlacement{
+			Drive:         strings.TrimSpace(man.Drive),
+			NamespaceRoot: strings.TrimSpace(man.NamespaceRoot),
+		}
+	}
+	c.Known = c.Recorded.Known()
+	if !c.Known || c.LiveDrive == "" {
+		return c
+	}
+	c.Mismatch = filepath.Clean(c.Recorded.Drive) != filepath.Clean(c.LiveDrive)
+	return c
+}
+
+// scratchManifestMaxDepth bounds the walk below. The unit sits a handful of levels under the scratch
+// root (the restore mirrors the snapshot's absolute path), and an unbounded walk over a scratch that
+// also holds a full userdata tree would stat a customer's entire library to find one small file.
+const scratchManifestMaxDepth = 8
+
+// recordedPlacementFromScratch reads the placement out of the unit manifest inside a PREPARED
+// restore scratch, without restoring anything.
+//
+// It exists for the not-installed refusal (scenario B), which fires BEFORE the live namespace root
+// can be resolved — so it cannot use the manifest the reconstitution opens later. The scratch is
+// already on local disk by then; this is a bounded walk and a file read, never a network call.
+//
+// Returns the zero value on ANY failure — unreadable, absent, malformed. A refusal that cannot name
+// the recorded place must fall back to the plain sentence rather than print an empty path, which
+// would read as "the backup says it lived nowhere".
+func (m *Manager) recordedPlacementFromScratch(scratch string) RecordedPlacement {
+	var found RecordedPlacement
+	root := filepath.Clean(scratch)
+	rootDepth := strings.Count(root, string(os.PathSeparator))
+	_ = filepath.WalkDir(root, func(path string, d fs.DirEntry, err error) error {
+		if err != nil {
+			return nil // an unreadable subtree is not fatal: keep looking elsewhere
+		}
+		if d.IsDir() {
+			if strings.Count(filepath.Clean(path), string(os.PathSeparator))-rootDepth >= scratchManifestMaxDepth {
+				return fs.SkipDir
+			}
+			return nil
+		}
+		if d.Name() != "manifest.json" {
+			return nil
+		}
+		if man := readManifest(path); man != nil && strings.TrimSpace(man.Drive) != "" {
+			found = RecordedPlacement{
+				Drive:         strings.TrimSpace(man.Drive),
+				NamespaceRoot: strings.TrimSpace(man.NamespaceRoot),
+			}
+			return fs.SkipAll
+		}
+		return nil
+	})
+	return found
+}
+
+// PlacementMismatchMessage is the Hungarian refusal shown when the destination differs from the one
+// the backup recorded. It NAMES BOTH VALUES — which is the whole point: "a destination differs" that
+// does not say from what leaves the customer with a decision they cannot make.
+//
+// It is a refusal with a route, not a dead end: the caller re-offers the action with the
+// acknowledgement field set, so the customer can proceed deliberately (scenario C).
+func PlacementMismatchMessage(stack string, c PlacementCheck) string {
+	return fmt.Sprintf(
+		"A(z) %s mentése szerint az adatok korábban itt voltak: %s. Most viszont ide állna vissza: %s. "+
+			"Ez lehet szándékos — például ha meghajtót cseréltél —, de magától nem folytatjuk. "+
+			"Ha így jó, erősítsd meg alább, és a visszaállítás az új helyre fut.",
+		stack, c.Recorded.Drive, c.LiveDrive)
+}

controller/internal/backup/offbox_placement_refusal_test.go

@@ -0,0 +1,167 @@
+package backup
+
+import (
+	"context"
+	"io/fs"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+)
+
+// scratchManifestPath finds the unit manifest the fixture wrote into the prepared scratch, so a test
+// can rewrite it. Fails loudly rather than returning "": a silently-missing manifest would make the
+// tests below pass for the wrong reason.
+func scratchManifestPath(t *testing.T, m *Manager, stack string) string {
+	t.Helper()
+	scratch, _, err := m.offboxRestoreScratchDir(stack)
+	if err != nil {
+		t.Fatalf("fixture: scratch dir: %v", err)
+	}
+	var found string
+	_ = filepath.WalkDir(scratch, func(p string, d fs.DirEntry, err error) error {
+		if err == nil && !d.IsDir() && d.Name() == "manifest.json" {
+			found = p
+			return fs.SkipAll
+		}
+		return nil
+	})
+	if found == "" {
+		t.Fatal("fixture: no unit manifest in the prepared scratch — the test would prove nothing")
+	}
+	return found
+}
+
+// R-351 SCENARIO B — THE APP IS NOT INSTALLED AND THE BACKUP NAMES WHERE IT LIVED.
+//
+// The refusal already existed (R-253) and correctly stopped the restore. What it did NOT do was say
+// where the data belonged, so the person on 2026-08-21 had to remember the drive and the address
+// themselves — both of which the backup was holding the whole time.
+//
+// WRONG OUTCOME THIS PINS: "it starts, and writes somewhere else." The assertions below are that the
+// restore is refused AND that the app was never touched — not merely that an error came back.
+func TestReconstitute_NotInstalled_RefusalNamesTheRecordedDrive(t *testing.T) {
+	const recordedDrive = "/mnt/felhom-drives/hdd_1"
+
+	m, prov, _ := reconFixture(t, "20260719T060000Z", "2026-07-19T06:00:00Z", pgDump(1))
+
+	// The backup records a drive — the one this box no longer has attached.
+	manPath := scratchManifestPath(t, m, "immich")
+	if err := writeManifest(manPath, &RecoveryManifest{
+		SchemaVersion: 2, AppName: "immich",
+		Drive: recordedDrive, NamespaceRoot: recordedDrive,
+	}); err != nil {
+		t.Fatal(err)
+	}
+
+	// The app is not installed: no live HDD path. This is the rebuilt-machine shape.
+	prov.hdd = map[string]string{}
+	if got := prov.GetStackHDDPath("immich"); got != "" {
+		t.Fatalf("fixture: the app must look uninstalled, got hdd=%q", got)
+	}
+	callsBefore := len(prov.calls)
+
+	_, err := m.ReconstituteFromOffsite(context.Background(), "immich", false)
+	if err == nil {
+		t.Fatal("a restore with no destination must be REFUSED, not started")
+	}
+	if !strings.Contains(err.Error(), recordedDrive) {
+		t.Errorf("the refusal must name the drive the backup recorded (%s); got: %v", recordedDrive, err)
+	}
+	// The route, not just the reason. „Alkalmaz" is the ASCII stem of „Alkalmazások" — matching the
+	// stem keeps accented bytes out of the comparison (strict rule 7).
+	if !strings.Contains(err.Error(), "Alkalmaz") {
+		t.Errorf("the refusal must name a route the person can take; got: %v", err)
+	}
+	// THE OBSERVABLE THAT MATTERS: nothing was stopped, placed or started. A refusal that still
+	// touched the stack would be the defect wearing an error message.
+	if len(prov.calls) != callsBefore {
+		t.Errorf("a refused restore must not touch the app; provider calls: %v", prov.calls[callsBefore:])
+	}
+}
+
+// R-351 SCENARIO C — THE DESTINATION DIFFERS FROM THE ONE THE BACKUP RECORDED.
+//
+// Not blocked outright (a drive can legitimately change) and not silently accepted (which is how a
+// restore lands in the wrong place under a green message). Named, and then the customer's own
+// deliberate acknowledgement carries it.
+func TestReconstitute_MismatchedDestination_RefusesThenProceedsOnAcknowledgement(t *testing.T) {
+	const recordedDrive = "/mnt/felhom-drives/hdd_1"
+
+	m, prov, _ := reconFixture(t, "20260719T060000Z", "2026-07-19T06:00:00Z", pgDump(1))
+	manPath := scratchManifestPath(t, m, "immich")
+	if err := writeManifest(manPath, &RecoveryManifest{
+		SchemaVersion: 2, AppName: "immich",
+		Drive: recordedDrive, NamespaceRoot: recordedDrive,
+	}); err != nil {
+		t.Fatal(err)
+	}
+	// The app IS installed — on a different drive from the recorded one (the fixture's temp dir).
+	live := prov.GetStackHDDPath("immich")
+	if live == "" || live == recordedDrive {
+		t.Fatalf("fixture: the live drive must exist and differ from the recorded one; got %q", live)
+	}
+
+	callsBefore := len(prov.calls)
+	_, err := m.ReconstituteFromOffsite(context.Background(), "immich", false)
+	if err == nil {
+		t.Fatal("a divergent destination must be NAMED and refused, never silently accepted")
+	}
+	for _, must := range []string{recordedDrive, live} {
+		if !strings.Contains(err.Error(), must) {
+			t.Errorf("the refusal must name %q so the customer can decide; got: %v", must, err)
+		}
+	}
+	if len(prov.calls) != callsBefore {
+		t.Errorf("the refusal must not touch the app; provider calls: %v", prov.calls[callsBefore:])
+	}
+
+	// ...and the customer may go ahead deliberately. Blocking outright is the other wrong outcome.
+	res, err := m.ReconstituteFromOffsite(context.Background(), "immich", true)
+	if err != nil {
+		t.Fatalf("an acknowledged placement change must proceed: %v", err)
+	}
+	if !res.Placement.Mismatch {
+		t.Error("the outcome must still record that the destination differed — a success that forgets the difference reads as a match")
+	}
+	if res.Placement.Recorded.Drive != recordedDrive {
+		t.Errorf("the outcome must carry the recorded drive, got %q", res.Placement.Recorded.Drive)
+	}
+}
+
+// recordedPlacementFromScratch must be TOTAL: an absent scratch, an unreadable one and a manifest
+// with no drive all yield the zero value, never a fabricated empty path.
+func TestRecordedPlacementFromScratch_UnknownStaysUnknown(t *testing.T) {
+	m := &Manager{}
+	if got := m.recordedPlacementFromScratch(filepath.Join(t.TempDir(), "does-not-exist")); got.Known() {
+		t.Errorf("an absent scratch must not yield a recorded placement, got %+v", got)
+	}
+
+	empty := t.TempDir()
+	if got := m.recordedPlacementFromScratch(empty); got.Known() {
+		t.Errorf("an empty scratch must not yield a recorded placement, got %+v", got)
+	}
+
+	noDrive := t.TempDir()
+	if err := writeManifest(filepath.Join(noDrive, "manifest.json"),
+		&RecoveryManifest{SchemaVersion: 1, AppName: "immich"}); err != nil {
+		t.Fatal(err)
+	}
+	if got := m.recordedPlacementFromScratch(noDrive); got.Known() {
+		t.Errorf("a manifest with no drive is an UNKNOWN, not a value; got %+v", got)
+	}
+
+	withDrive := t.TempDir()
+	deep := filepath.Join(withDrive, "a", "b", "c", "backups", "primary", "immich")
+	if err := os.MkdirAll(deep, 0o755); err != nil {
+		t.Fatal(err)
+	}
+	if err := writeManifest(filepath.Join(deep, "manifest.json"),
+		&RecoveryManifest{SchemaVersion: 2, AppName: "immich", Drive: "/mnt/sys_drive"}); err != nil {
+		t.Fatal(err)
+	}
+	got := m.recordedPlacementFromScratch(withDrive)
+	if !got.Known() || got.Drive != "/mnt/sys_drive" {
+		t.Errorf("a nested unit manifest must be found, got %+v", got)
+	}
+}

controller/internal/backup/offbox_placement_test.go

@@ -0,0 +1,135 @@
+package backup
+
+import (
+	"strings"
+	"testing"
+)
+
+// R-351 — THE SCENARIO TABLE from the task, as a truth table.
+//
+// Each row states the WRONG outcome it exists to prevent, because a row whose expectation is only
+// "want X" tells the next reader nothing about why the value matters.
+//
+// The live values in these fixtures are the ones measured on demo-hp 2026-08-21, not invented ones:
+//
+//	calibre-web (declares a data path)      drive=/mnt/felhom-drives/hdd_1
+//	opengist    (declares NONE, 40-of-53)   drive=/mnt/sys_drive
+func TestCheckPlacement_ScenarioTable(t *testing.T) {
+	const dataDrive = "/mnt/felhom-drives/hdd_1"
+	const sysDrive = "/mnt/sys_drive"
+
+	for _, tc := range []struct {
+		name         string
+		man          *RecoveryManifest
+		liveDrive    string
+		wantKnown    bool
+		wantMismatch bool
+		wrongOutcome string
+	}{
+		{
+			name:         "A/D - recorded drive is where we are restoring",
+			man:          &RecoveryManifest{Drive: dataDrive, NamespaceRoot: dataDrive},
+			liveDrive:    dataDrive,
+			wantKnown:    true,
+			wantMismatch: false,
+			wrongOutcome: "a new question or obstacle on the ordinary path",
+		},
+		{
+			name:         "C - the destination differs from what the backup recorded",
+			man:          &RecoveryManifest{Drive: dataDrive, NamespaceRoot: dataDrive},
+			liveDrive:    "/mnt/felhom-drives/hdd_2",
+			wantKnown:    true,
+			wantMismatch: true,
+			wrongOutcome: "silently accepted - a restore into the wrong place under a green message",
+		},
+		{
+			name:         "E - the backup records no drive (older unit, missing field)",
+			man:          &RecoveryManifest{Drive: "", NamespaceRoot: ""},
+			liveDrive:    dataDrive,
+			wantKnown:    false,
+			wantMismatch: false,
+			wrongOutcome: "an empty prefill presented as if it were the recorded value",
+		},
+		{
+			name:         "E - the manifest could not be read at all",
+			man:          nil,
+			liveDrive:    dataDrive,
+			wantKnown:    false,
+			wantMismatch: false,
+			wrongOutcome: "a nil manifest treated as a match, or as a refusal that strands every old backup",
+		},
+		{
+			name:         "no declared data path - the app has no field, but the record is still the truth",
+			man:          &RecoveryManifest{Drive: sysDrive, NamespaceRoot: sysDrive + "/felhom-data"},
+			liveDrive:    sysDrive,
+			wantKnown:    true,
+			wantMismatch: false,
+			wrongOutcome: "falling into the unknown case just because the app has no storage field",
+		},
+		{
+			name:         "no declared data path - and it moved to a real drive",
+			man:          &RecoveryManifest{Drive: sysDrive, NamespaceRoot: sysDrive + "/felhom-data"},
+			liveDrive:    dataDrive,
+			wantKnown:    true,
+			wantMismatch: true,
+			wrongOutcome: "the 40-of-53 class silently exempted from the mismatch check",
+		},
+		{
+			name:         "a trailing slash is the same destination",
+			man:          &RecoveryManifest{Drive: dataDrive + "/", NamespaceRoot: dataDrive},
+			liveDrive:    dataDrive,
+			wantKnown:    true,
+			wantMismatch: false,
+			wrongOutcome: "a manufactured mismatch the customer has to dismiss for no reason",
+		},
+		{
+			name:         "live destination unresolvable",
+			man:          &RecoveryManifest{Drive: dataDrive, NamespaceRoot: dataDrive},
+			liveDrive:    "",
+			wantKnown:    true,
+			wantMismatch: false,
+			wrongOutcome: "comparing against nothing and calling it a difference; the not-installed refusal owns this case",
+		},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			c := CheckPlacement(tc.man, tc.liveDrive, tc.liveDrive)
+			if c.Known != tc.wantKnown {
+				t.Errorf("Known = %v, want %v — wrong outcome guarded: %s", c.Known, tc.wantKnown, tc.wrongOutcome)
+			}
+			if c.Mismatch != tc.wantMismatch {
+				t.Errorf("Mismatch = %v, want %v — wrong outcome guarded: %s", c.Mismatch, tc.wantMismatch, tc.wrongOutcome)
+			}
+		})
+	}
+}
+
+// The refusal must NAME BOTH VALUES. "The destination differs" without saying from what leaves the
+// customer holding a decision they have no way to make — which is the same defect in a politer form.
+func TestPlacementMismatchMessage_NamesBothPlaces(t *testing.T) {
+	c := CheckPlacement(
+		&RecoveryManifest{Drive: "/mnt/felhom-drives/hdd_1"},
+		"/mnt/sys_drive", "/mnt/sys_drive/felhom-data")
+	if !c.Mismatch {
+		t.Fatal("fixture: these differ, or the message under test is never reached")
+	}
+	msg := PlacementMismatchMessage("opengist", c)
+	for _, must := range []string{"opengist", "/mnt/felhom-drives/hdd_1", "/mnt/sys_drive"} {
+		if !strings.Contains(msg, must) {
+			t.Errorf("the refusal must name %q; got: %s", must, msg)
+		}
+	}
+}
+
+// An UNKNOWN recording must never be rendered as a value. Scenario E's wrong outcome is precisely an
+// empty string shown where a recorded path belongs, which reads as "the backup says it lived
+// nowhere" — a fabricated fact.
+func TestRecordedPlacement_UnknownIsNotAValue(t *testing.T) {
+	for _, drive := range []string{"", "   ", "\t"} {
+		if (RecordedPlacement{Drive: drive}).Known() {
+			t.Errorf("a blank drive (%q) must not count as a recorded value", drive)
+		}
+	}
+	if !(RecordedPlacement{Drive: "/mnt/sys_drive"}).Known() {
+		t.Error("a real recorded drive must count as known, or the whole check is inert")
+	}
+}

controller/internal/backup/offbox_reconstitute.go

@@ -70,6 +70,12 @@ type OffsiteReconstituteResult struct {
 	OffsiteRunID string    // "" for a pre-v0.148 snapshot — an unverified pair
 	Skewed       bool      // the snapshot carries no coherence stamp: files and DB may differ in age
 	LooksEmpty   bool      // R-44 sniff on the dump about to be replayed
+	// Placement (R-351) is what the backup recorded about where this app's data lived, compared
+	// against where this restore actually wrote. Carried on the RESULT and not only on the refusal,
+	// so a restore that proceeded into a different destination says so in its own outcome rather
+	// than reporting a bare success — a warning beside a success is read as a success, so the
+	// difference has to survive into the message.
+	Placement PlacementCheck
 }
 
 // fullPlaceCopier returns the FULL-restore file copier (nil seam → rsyncRestoreOverwrite).
@@ -166,7 +172,11 @@ func (m *Manager) dumpForSafety(ctx context.Context, db DiscoveredDB, dumpDir st
 // overwritten to the snapshot's version (extras survive, nothing deleted), then the snapshot's own
 // DB dump replayed, with a safety dump of the current database taken first. Requires a completed
 // FULL scratch restore (RestoreOffboxScratch with full=true). Single-flight.
-func (m *Manager) ReconstituteFromOffsite(ctx context.Context, stack string) (OffsiteReconstituteResult, error) {
+// ackPlacementChange (R-351) is the customer's DELIBERATE acknowledgement that the destination
+// differs from the one the backup recorded. It is a separate act from the restore's own confirm:
+// folding it into `confirm=1` would mean one click carried two decisions, which is precisely what
+// R-48 exists to prevent.
+func (m *Manager) ReconstituteFromOffsite(ctx context.Context, stack string, ackPlacementChange bool) (OffsiteReconstituteResult, error) {
 	var res OffsiteReconstituteResult
 	if !m.OffboxConfigured() {
 		return res, fmt.Errorf("off-box backup not configured")
@@ -203,6 +213,16 @@ func (m *Manager) ReconstituteFromOffsite(ctx context.Context, stack string) (Of
 		// destination is the app's own HDD path, which is a drive the CUSTOMER chooses at deploy
 		// time, and picking it for them is the decision this whole recovery path exists to leave
 		// with them.
+		// R-351: the refusal now NAMES the place the backup recorded, when it can read it. The
+		// prepared scratch already contains the unit, so this is a local file read — no network call,
+		// nothing restored, and it happens on a path that was going to refuse anyway. Telling
+		// somebody to reinstall without telling them where the data belongs is what forced the
+		// 2026-08-21 operator to remember two values the backup already held.
+		if rec := m.recordedPlacementFromScratch(scratch); rec.Known() {
+			return res, fmt.Errorf("a(z) %s nincs telepítve, ezért nincs hová visszaállítani az adatait. "+
+				"A mentése szerint az adatai itt voltak: %s. Telepítsd újra az alkalmazást (Alkalmazások) "+
+				"ugyanerre a helyre, utána ez a visszaállítás működni fog", stack, rec.Drive)
+		}
 		return res, fmt.Errorf("a(z) %s nincs telepítve, ezért nincs hová visszaállítani az adatait — "+
 			"telepítsd újra az alkalmazást (Alkalmazások), utána ez a visszaállítás működni fog", stack)
 	}
@@ -232,7 +252,8 @@ func (m *Manager) ReconstituteFromOffsite(ctx context.Context, stack string) (Of
 		return res, fmt.Errorf("a pillanatképben nincs mentési egység — a visszaállítás nem indítható")
 	}
 	scratchDumpDir := filepath.Join(scratchUnit, "db-dumps")
-	if man := readManifest(filepath.Join(scratchUnit, "manifest.json")); man != nil {
+	man := readManifest(filepath.Join(scratchUnit, "manifest.json"))
+	if man != nil {
 		res.OffsiteRunID = man.OffsiteRunID
 		if man.DumpsAt != "" {
 			if t, pErr := time.Parse(time.RFC3339, man.DumpsAt); pErr == nil {
@@ -240,6 +261,23 @@ func (m *Manager) ReconstituteFromOffsite(ctx context.Context, stack string) (Of
 			}
 		}
 	}
+
+	// --- WHERE THE BACKUP SAYS THIS DATA LIVED (R-351) ------------------------------------------
+	// The manifest we just opened has carried `drive` and `namespace_root` since schema 1, and until
+	// now nothing read them back. Compared HERE, before the safety dump and before the first byte is
+	// placed, so the refusal costs nothing and leaves the app completely untouched.
+	//
+	// An UNKNOWN recording (a pre-field unit, or one we could not read) is not a mismatch and does
+	// not refuse: blocking on an absence would strand every older backup, and CheckPlacement returns
+	// that case explicitly rather than letting it fall through as "they match".
+	res.Placement = CheckPlacement(man, hdd, liveNs)
+	if res.Placement.Mismatch && !ackPlacementChange {
+		return res, fmt.Errorf("%s", PlacementMismatchMessage(stack, res.Placement))
+	}
+	if res.Placement.Mismatch {
+		m.logger.Printf("[WARN] [offbox] %s: restoring into %s, but the backup recorded %s — the customer acknowledged the change",
+			stack, res.Placement.LiveDrive, res.Placement.Recorded.Drive)
+	}
 	// A pre-v0.148 snapshot carries no stamp: its dump was whatever the 02:30 local run left behind,
 	// so the pair's two halves may be hours or days apart. Surfaced, never blocked — the confirm
 	// dialog says so and the safety dump makes it reversible.

controller/internal/backup/offbox_reconstitute_test.go

@@ -174,7 +174,7 @@ func reconFixture(t *testing.T, runID, dumpsAt string, dumpBody string) (*Manage
 func TestReconstituteReplaysDBAndOrdersOperations(t *testing.T) {
 	m, prov, imported := reconFixture(t, "20260719T060000Z", "2026-07-19T06:00:00Z", pgDump(1))
 
-	res, err := m.ReconstituteFromOffsite(context.Background(), "immich")
+	res, err := m.ReconstituteFromOffsite(context.Background(), "immich", false)
 	if err != nil {
 		t.Fatalf("reconstitute: %v", err)
 	}
@@ -220,7 +220,7 @@ func TestReconstituteRefusesWhenSafetyDumpFails(t *testing.T) {
 	var copied bool
 	m.SetOffboxFullPlaceCopier(func(_, _ string) (int, error) { copied = true; return 1, nil })
 
-	_, err := m.ReconstituteFromOffsite(context.Background(), "immich")
+	_, err := m.ReconstituteFromOffsite(context.Background(), "immich", false)
 	if err == nil {
 		t.Fatal("expected a refusal when the safety dump cannot be taken")
 	}
@@ -246,7 +246,7 @@ func TestReconstituteNoDBAppMakesNoDumpOrImportCalls(t *testing.T) {
 		return DumpResult{DB: d}
 	})
 
-	res, err := m.ReconstituteFromOffsite(context.Background(), "immich")
+	res, err := m.ReconstituteFromOffsite(context.Background(), "immich", false)
 	if err != nil {
 		t.Fatalf("reconstitute: %v", err)
 	}
@@ -267,7 +267,7 @@ func TestReconstituteNoDBAppMakesNoDumpOrImportCalls(t *testing.T) {
 func TestReconstituteSurfacesLegacySkewedPair(t *testing.T) {
 	m, _, imported := reconFixture(t, "", "", pgDump(1))
 
-	res, err := m.ReconstituteFromOffsite(context.Background(), "immich")
+	res, err := m.ReconstituteFromOffsite(context.Background(), "immich", false)
 	if err != nil {
 		t.Fatalf("a legacy pair must still be restorable, got refusal: %v", err)
 	}
@@ -286,7 +286,7 @@ func TestReconstituteFlagsCustomerEmptyDump(t *testing.T) {
 	// A valid postgres dump whose accounts table has NO rows — the 2026-07-19 shape exactly.
 	m, _, _ := reconFixture(t, "run1", "2026-07-19T06:00:00Z", pgDump(0))
 
-	res, err := m.ReconstituteFromOffsite(context.Background(), "immich")
+	res, err := m.ReconstituteFromOffsite(context.Background(), "immich", false)
 	if err != nil {
 		t.Fatalf("the sniff must never block a restore: %v", err)
 	}

controller/internal/backup/opstatus.go

@@ -18,6 +18,15 @@ type RestoreOpResult struct {
 	FinishedAt time.Time `json:"finished_at"`
 }
 
+// RestoreResultWindow bounds how long a finished restore still counts as "what just happened".
+//
+// ONE EXPRESSION, TWO SURFACES (R-351). The wizard's phase strip and the list page's banner both
+// need this bound, and until now only the wizard had one — it lived in internal/web as an
+// unexported constant. A second copy in the JS would be exactly the "two copies that already
+// differed" shape this repo keeps paying for, so the window is defined HERE, beside the status it
+// bounds, and both surfaces read it from the payload.
+const RestoreResultWindow = 10 * time.Minute
+
 // RestoreOpStatus is the shape served at GET /api/backup/restore-status.
 type RestoreOpStatus struct {
 	Running   bool             `json:"running"`
@@ -25,6 +34,15 @@ type RestoreOpStatus struct {
 	Stack     string           `json:"stack,omitempty"`
 	StartedAt time.Time        `json:"started_at,omitempty"`
 	Last      *RestoreOpResult `json:"last,omitempty"`
+	// LastRecent reports whether Last finished recently enough to still be worth showing to someone
+	// who was NOT watching when it happened.
+	//
+	// R-351, the measured defect: the banner's JS gated the terminal result on a page-local
+	// `sawRunning` flag, so a restore that finished before the page was opened — or in under one
+	// poll interval — was shown to nobody. The 2026-08-21 OpenGist restore finished in 8.7s and the
+	// operator could not tell from any screen whether it had completed; the answer existed only in
+	// a container log. A result nobody can see is the same defect class as no result at all.
+	LastRecent bool `json:"last_recent,omitempty"`
 }
 
 // BeginRestoreOp marks a restore op in flight (called by the handler just before launching the
@@ -66,6 +84,14 @@ func (m *Manager) RestoreStatus() RestoreOpStatus {
 	if m.opLast != nil {
 		cp := *m.opLast
 		st.Last = &cp
+		// Recency is decided here, on the clock the result was stamped with, so neither surface has
+		// to hold its own copy of the window. A zero FinishedAt is never recent — an unstamped
+		// result must not be drawn as "just now".
+		if !cp.FinishedAt.IsZero() {
+			if d := time.Since(cp.FinishedAt); d >= 0 && d < RestoreResultWindow {
+				st.LastRecent = true
+			}
+		}
 	}
 	return st
 }

controller/internal/backup/opstatus_recent_test.go

@@ -0,0 +1,84 @@
+package backup
+
+import (
+	"testing"
+	"time"
+)
+
+// R-351 — A FINISHED RESTORE MUST BE VISIBLE TO SOMEONE WHO WAS NOT WATCHING.
+//
+// THE MEASURED CASE (demo-hp, 2026-08-21). An off-box reconstitution refused at 16:37:14, the person
+// reinstalled and ran the local unit restore, and it completed at 16:39:25 — in 8.666s. No screen
+// ever said so. The banner's JS gated its terminal result on a page-local `sawRunning` flag, so the
+// result was rendered only for a browser that happened to be open and polling across the transition.
+// Land on the page a moment later and the banner stayed hidden: "completed" and "never ran" looked
+// identical. The answer existed only in `docker logs felhom-controller`, which a customer cannot
+// reach.
+//
+// This pins the CONSEQUENCE — the status carries a recency verdict a late arrival can act on — not
+// the mechanism. Mutating LastRecent to stay false must fail this test.
+func TestRestoreStatus_LastRecent(t *testing.T) {
+	for _, tc := range []struct {
+		name       string
+		finishedAt time.Time
+		wantRecent bool
+		why        string
+	}{
+		{
+			name:       "just finished",
+			finishedAt: time.Now().Add(-9 * time.Second),
+			wantRecent: true,
+			why:        "the 8.7s OpenGist restore: finished before a poll could see it running",
+		},
+		{
+			name:       "inside the window",
+			finishedAt: time.Now().Add(-RestoreResultWindow + time.Minute),
+			wantRecent: true,
+			why:        "still answers \"what just happened\"",
+		},
+		{
+			name:       "outside the window",
+			finishedAt: time.Now().Add(-RestoreResultWindow - time.Minute),
+			wantRecent: false,
+			why:        "landing here later must not claim a restore just finished",
+		},
+		{
+			name:       "unstamped result",
+			finishedAt: time.Time{},
+			wantRecent: false,
+			why:        "a zero timestamp is an UNKNOWN and must never be drawn as \"just now\" (presence is not success)",
+		},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			m := &Manager{}
+			m.opLast = &RestoreOpResult{
+				Op: "restore", Stack: "opengist", OK: true,
+				Message: "Restore completed", FinishedAt: tc.finishedAt,
+			}
+			st := m.RestoreStatus()
+			if st.Last == nil {
+				t.Fatal("fixture: the status must carry the terminal result")
+			}
+			if st.LastRecent != tc.wantRecent {
+				t.Errorf("LastRecent = %v, want %v — %s", st.LastRecent, tc.wantRecent, tc.why)
+			}
+		})
+	}
+}
+
+// A RUNNING op must not be reported as a recent RESULT — the banner shows one or the other, and
+// conflating them would put a success message over an operation that is still writing.
+func TestRestoreStatus_RunningIsNotAResult(t *testing.T) {
+	m := &Manager{}
+	m.BeginRestoreOp("offbox-reconstitute", "opengist")
+	st := m.RestoreStatus()
+	if !st.Running {
+		t.Fatal("fixture: the op must be reported running")
+	}
+	if st.Last != nil {
+		t.Errorf("a fresh op has no terminal result yet, got %+v", st.Last)
+	}
+	if st.LastRecent {
+		t.Error("a running op must never carry a recent-result verdict")
+	}
+}

controller/internal/backup/r47_replay_order_test.go

@@ -71,7 +71,7 @@ func TestReconstituteReplaysWithOnlyTheDBServiceUp(t *testing.T) {
 		return nil
 	}
 
-	res, err := m.ReconstituteFromOffsite(context.Background(), "immich")
+	res, err := m.ReconstituteFromOffsite(context.Background(), "immich", false)
 	if err != nil {
 		t.Fatalf("reconstitute: %v", err)
 	}
@@ -109,7 +109,7 @@ func TestReconstituteNoDBAppNeverStartsServicesOnly(t *testing.T) {
 	prov.composePath = writeLiveCompose(t, noDBCompose)
 	m.discoverDBs = func(context.Context) ([]DiscoveredDB, error) { return nil, nil }
 
-	res, err := m.ReconstituteFromOffsite(context.Background(), "immich")
+	res, err := m.ReconstituteFromOffsite(context.Background(), "immich", false)
 	if err != nil {
 		t.Fatalf("a no-DB app must restore unchanged, got: %v", err)
 	}
@@ -141,7 +141,7 @@ func TestReconstituteRefusesWhenNoDBServiceIdentifiable(t *testing.T) {
 	var copied bool
 	m.SetOffboxFullPlaceCopier(func(_, _ string) (int, error) { copied = true; return 1, nil })
 
-	_, err := m.ReconstituteFromOffsite(context.Background(), "immich")
+	_, err := m.ReconstituteFromOffsite(context.Background(), "immich", false)
 	if err == nil {
 		t.Fatal("expected a refusal: a dump exists but no database service can be started for it")
 	}
@@ -270,7 +270,7 @@ func TestReconstituteReplayFailureStillBringsTheStackUp(t *testing.T) {
 		return context.DeadlineExceeded
 	}
 
-	res, err := m.ReconstituteFromOffsite(context.Background(), "immich")
+	res, err := m.ReconstituteFromOffsite(context.Background(), "immich", false)
 	if err == nil {
 		t.Fatal("a failed replay must be surfaced, not swallowed")
 	}
@@ -292,7 +292,7 @@ func TestReconstituteDBOnlyStartFailureStillBringsTheStackUp(t *testing.T) {
 	m, prov, imported := reconFixture(t, "run1", "2026-07-19T06:00:00Z", pgDump(1))
 	prov.startSvcErr = context.DeadlineExceeded
 
-	if _, err := m.ReconstituteFromOffsite(context.Background(), "immich"); err == nil {
+	if _, err := m.ReconstituteFromOffsite(context.Background(), "immich", false); err == nil {
 		t.Fatal("a failed DB-only start must be surfaced")
 	}
 	if !prov.fullStarted {

controller/internal/web/handlers.go

@@ -1401,8 +1401,8 @@ func (s *Server) backupRestoreHandler(w http.ResponseWriter, r *http.Request) {
 	// Part B: restore is a long SYNCHRONOUS op (F4 — through cloudflared's hard 100s cap the customer
 	// got an error page while it silently succeeded). Fast-path refuse a concurrent op, then run it in
 	// a BACKGROUND goroutine (survives the request; the poll banner shows progress → result).
-	if s.backupMgr.IsRunning() {
-		http.Redirect(w, r, "/backups/restore?flash_error="+url.QueryEscape("Egy mentési/visszaállítási művelet már fut."), http.StatusFound)
+	if msg, blocked := s.restoreOpBlocked(); blocked {
+		http.Redirect(w, r, "/backups/restore?flash_error="+url.QueryEscape(msg), http.StatusFound)
 		return
 	}
 	s.logger.Printf("[WARN] [web] Restore requested (async): stack=%s, snapshot=%s from %s", stackName, snapshotID, r.RemoteAddr)
@@ -1460,8 +1460,8 @@ func (s *Server) backupTier2RestoreHandler(w http.ResponseWriter, r *http.Reques
 		return
 	}
 	// Part B (same async shape as backupRestoreHandler): fast-path refuse, then background goroutine.
-	if s.backupMgr.IsRunning() {
-		http.Redirect(w, r, "/backups/apps?flash_error="+url.QueryEscape("Egy mentési/visszaállítási művelet már fut."), http.StatusFound)
+	if msg, blocked := s.restoreOpBlocked(); blocked {
+		http.Redirect(w, r, "/backups/apps?flash_error="+url.QueryEscape(msg), http.StatusFound)
 		return
 	}
 

controller/internal/web/offbox_handlers.go

@@ -352,10 +352,10 @@ func (s *Server) offboxRestoreHandler(w http.ResponseWriter, r *http.Request) {
 	}
 	// Fast-path refuse a concurrent op, then run async on a BACKGROUND context (a proxy read-timeout on
 	// r.Context() would CANCEL the SFTP restore mid-flight — the F4 lesson).
-	if s.backupMgr.IsRunning() {
+	if msg, blocked := s.restoreOpBlocked(); blocked {
 		// Same silence class as the size gate: a refusal that starts nothing must still be findable.
 		s.logger.Printf("[WARN] [web] off-box restore refused for %s (mode=%s): another backup/restore op is already running", app, mode)
-		offboxRedirectTo(w, r, restoreWizardPath(app), "Egy mentési/visszaállítási művelet már fut.", true)
+		offboxRedirectTo(w, r, restoreWizardPath(app), msg, true)
 		return
 	}
 	full := mode == "full"
@@ -433,15 +433,19 @@ func (s *Server) offboxReconstituteHandler(w http.ResponseWriter, r *http.Reques
 		offboxRedirectTo(w, r, restoreWizardPath(app), "A teljes visszaállítás megerősítés nélkül nem hajtható végre.", true)
 		return
 	}
-	if s.backupMgr.IsRunning() {
-		offboxRedirectTo(w, r, restoreWizardPath(app), "Egy mentési/visszaállítási művelet már fut.", true)
+	if msg, blocked := s.restoreOpBlocked(); blocked {
+		offboxRedirectTo(w, r, restoreWizardPath(app), msg, true)
 		return
 	}
+	// R-351: a SEPARATE field from `confirm`. The restore's own confirm answers "overwrite my live
+	// data"; this one answers "yes, into a different place than the backup recorded". One checkbox
+	// carrying both would be the two-decisions-one-button shape R-48 removed from this surface.
+	ackPlacement := r.FormValue("ack_placement") == "1"
 	s.backupMgr.BeginRestoreOp("offbox-reconstitute", app)
 	go func() {
 		ctx, cancel := context.WithTimeout(context.Background(), 60*time.Minute)
 		defer cancel()
-		res, err := s.backupMgr.ReconstituteFromOffsite(ctx, app)
+		res, err := s.backupMgr.ReconstituteFromOffsite(ctx, app, ackPlacement)
 		if err != nil {
 			s.logger.Printf("[ERROR] [web] off-box reconstitute %s (async): %v", app, err)
 			s.backupMgr.EndRestoreOp(false, "A teljes visszaállítás sikertelen: "+err.Error())
@@ -521,8 +525,8 @@ func (s *Server) offboxPlaceHandler(w http.ResponseWriter, r *http.Request) {
 		offboxRedirectTo(w, r, "/backups/restore", "Hiányzó alkalmazás.", true)
 		return
 	}
-	if s.backupMgr.IsRunning() {
-		offboxRedirectTo(w, r, restoreWizardPath(app), "Egy mentési/visszaállítási művelet már fut.", true)
+	if msg, blocked := s.restoreOpBlocked(); blocked {
+		offboxRedirectTo(w, r, restoreWizardPath(app), msg, true)
 		return
 	}
 	s.backupMgr.BeginRestoreOp("offbox-place", app)
@@ -554,8 +558,8 @@ func (s *Server) sharesRestoreHandler(w http.ResponseWriter, r *http.Request) {
 		offboxRedirectTo(w, r, "/backups/restore", "A távoli mentési cél nincs beállítva.", true)
 		return
 	}
-	if s.backupMgr.IsRunning() {
-		offboxRedirectTo(w, r, "/backups/restore", "Egy mentési/visszaállítási művelet már fut.", true)
+	if msg, blocked := s.restoreOpBlocked(); blocked {
+		offboxRedirectTo(w, r, "/backups/restore", msg, true)
 		return
 	}
 	s.backupMgr.BeginRestoreOp("shares-restore", backup.SharesDisplayName)
@@ -580,8 +584,8 @@ func (s *Server) sharesPlaceHandler(w http.ResponseWriter, r *http.Request) {
 		offboxRedirectTo(w, r, "/backups/restore", "A távoli mentési cél nincs beállítva.", true)
 		return
 	}
-	if s.backupMgr.IsRunning() {
-		offboxRedirectTo(w, r, "/backups/restore", "Egy mentési/visszaállítási művelet már fut.", true)
+	if msg, blocked := s.restoreOpBlocked(); blocked {
+		offboxRedirectTo(w, r, "/backups/restore", msg, true)
 		return
 	}
 	s.backupMgr.BeginRestoreOp("shares-place", backup.SharesDisplayName)

controller/internal/web/restore_inflight_guard_test.go

@@ -0,0 +1,116 @@
+package web
+
+import (
+	"net/http/httptest"
+	"net/url"
+	"strings"
+	"testing"
+
+	"gitea.dooplex.hu/admin/felhom-controller/internal/settings"
+)
+
+// R-351 — THE SECOND PRESS. Part 3 asked whether a second press really starts a second run or is
+// refused somewhere deeper. It is NOT refused: it starts a second run, and the customer is told so.
+//
+// WHY THE EXISTING GUARD DOES NOT CATCH IT. Every restore handler gates on `backupMgr.IsRunning()`
+// (offbox_handlers.go:355/436/524/557/583, handlers.go:1404/1463). That reads `m.running`, the
+// CONCURRENCY single-flight, which is acquired INSIDE the restore function on the background
+// goroutine (offbox_reconstitute.go:180, offbox_restore.go:393) — not by the handler. So between the
+// handler's check and the goroutine's acquire there is a window in which `IsRunning()` is false while
+// a restore is unmistakably in flight. `restoreOpInFlight` and the whole wizard already read the
+// other flag (`RestoreStatus().Running`, set synchronously by BeginRestoreOp) for exactly this
+// reason — see the long note on restoreOpInFlight. The handlers were never moved over.
+//
+// THE FIXTURE IS THE LIVE STATE, NOT AN INVENTED ONE: display flag set, concurrency flag NOT held.
+// That is precisely what the box looks like for the entire duration of an off-box restore.
+//
+// This test pins the CONSEQUENCE (does a second run start?), not the mechanism (which flag is read),
+// per CLAUDE.md's preference. Mutating the new guard back to `IsRunning()` must make it fail.
+func TestRestoreHandlers_SecondPressDoesNotStartASecondRun(t *testing.T) {
+	const firstApp = "alpha"
+	const secondApp = "beta"
+
+	s, sett, m := newOffboxWebServer(t)
+	if err := sett.SetOffboxTarget(&settings.OffboxTarget{
+		Enabled: true, Host: "nas.local", Port: 22, User: "felhom", RepoPath: "/srv/repo",
+		Schedule: "daily", EscrowState: "escrowed",
+	}); err != nil {
+		t.Fatal(err)
+	}
+	if err := m.WriteOffboxSecrets("PRIVATE-KEY-MATERIAL", "nas.local ssh-ed25519 AAAAhostkey"); err != nil {
+		t.Fatal(err)
+	}
+	if !m.OffboxConfigured() {
+		t.Fatal("fixture: the target must be configured, or the handler exits earlier and proves nothing")
+	}
+
+	// A restore is in flight, exactly as the live box has it.
+	m.BeginRestoreOp("offbox-restore", firstApp)
+
+	// The fixture must reproduce the GAP, or this test is vacuous: the display flag says running,
+	// the concurrency flag — the one every handler reads — says it is not.
+	if !m.RestoreStatus().Running {
+		t.Fatal("fixture: the display flag must report a running op")
+	}
+	if m.IsRunning() {
+		t.Fatal("fixture: the concurrency flag must NOT be held — that gap IS the defect under test")
+	}
+
+	post := func(path string, form url.Values) *httptest.ResponseRecorder {
+		r := httptest.NewRequest("POST", path, strings.NewReader(form.Encode()))
+		r.Header.Set("Content-Type", "application/x-www-form-urlencoded")
+		w := httptest.NewRecorder()
+		switch path {
+		case "/backup/offbox/reconstitute":
+			s.offboxReconstituteHandler(w, r)
+		case "/backup/offbox/place":
+			s.offboxPlaceHandler(w, r)
+		default:
+			t.Fatalf("unrouted path %q", path)
+		}
+		return w
+	}
+
+	for _, tc := range []struct {
+		name string
+		path string
+		form url.Values
+	}{
+		{"reconstitute", "/backup/offbox/reconstitute", url.Values{"app": {secondApp}, "confirm": {"1"}}},
+		{"place", "/backup/offbox/place", url.Values{"app": {secondApp}}},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			w := post(tc.path, tc.form)
+			if w.Code != 302 {
+				t.Fatalf("the handler redirects; got %d", w.Code)
+			}
+			loc := w.Header().Get("Location")
+
+			// CONSEQUENCE 1 — the customer must not be told a second restore started.
+			// "elind" is the ASCII stem shared by every „elindult" flash; matching it avoids putting
+			// accented bytes through a comparison (strict rule 7).
+			if strings.Contains(loc, "elind") {
+				t.Errorf("a second press while a restore runs must NOT report a started restore. Location: %q", loc)
+			}
+
+			// CONSEQUENCE 2 — the in-flight op must still be the FIRST one. If the handler ran,
+			// BeginRestoreOp overwrote the op name and stack, so the first restore's identity is
+			// gone from the status the banner reads.
+			st := m.RestoreStatus()
+			if st.Op != "offbox-restore" || st.Stack != firstApp {
+				t.Errorf("the first restore's identity was overwritten by the second press: op=%q stack=%q (want offbox-restore/%s)",
+					st.Op, st.Stack, firstApp)
+			}
+
+			// CONSEQUENCE 3 — a refusal must name a route the person can act on, not just a reason.
+			// The wizard path is the route; it is where the live status is shown.
+			if !strings.Contains(loc, "/backups/restore") {
+				t.Errorf("the refusal must route somewhere actionable; got %q", loc)
+			}
+
+			// Restore the fixture for the next subtest — a handler that (today) ran will have
+			// clobbered it.
+			m.BeginRestoreOp("offbox-restore", firstApp)
+		})
+	}
+}

controller/internal/web/restore_wizard.go

@@ -113,7 +113,12 @@ const (
 // Without a bound the last result would light that phase forever — landing on the page a week later
 // would claim you had just finished a restore. Same reasoning as escrowCeremonyGraceWindow; shorter,
 // because this answers "what just happened", not "are we still waiting".
-const restoreResultWindow = 10 * time.Minute
+//
+// R-351: this is now an ALIAS, not a second value. The list page's banner needs the same bound, and
+// the payload carries the verdict (RestoreOpStatus.LastRecent), so the window is defined once in
+// internal/backup beside the status it bounds. Keeping a separate literal here is how the two
+// surfaces would drift.
+const restoreResultWindow = backup.RestoreResultWindow
 
 // hasRecentRestoreResult reports whether THIS app has a just-finished restore to show. Pure (the
 // clock is a parameter) so the boundary and the wrong-app case are table-testable.
@@ -182,6 +187,43 @@ func restoreOpInFlight(st backup.RestoreOpStatus) bool {
 	return st.Running
 }
 
+// restoreOpBlocked reports whether a NEW restore must be refused right now, and returns the
+// Hungarian refusal to show. It reads BOTH flags, deliberately:
+//
+//   - `RestoreStatus().Running` — the DISPLAY flag, set synchronously by `BeginRestoreOp` in the
+//     handler. It is the only one that is true for the WHOLE duration of an off-box restore, which
+//     is what makes it the right flag to refuse on.
+//   - `IsRunning()` — the CONCURRENCY flag. The nightly backup run holds this one and never calls
+//     `BeginRestoreOp`, so dropping it would open a hole the old guard did close. Kept, not replaced.
+//
+// R-351, the measured defect: every restore handler read ONLY `IsRunning()`, which the restore
+// goroutine acquires AFTER the handler has already returned (offbox_reconstitute.go:180,
+// offbox_restore.go:393). A second press inside that window started a second run and was told
+// „…elindult". Pinned by TestRestoreHandlers_SecondPressDoesNotStartASecondRun, which asserts the
+// CONSEQUENCE — that the first restore's identity survives the second press — rather than which
+// flag was read.
+//
+// The refusal names a reason AND a route: the page it redirects to is the wizard, which carries the
+// live status banner, so „ezen az oldalon" is a true instruction and not a gesture.
+func (s *Server) restoreOpBlocked() (string, bool) {
+	if s.backupMgr == nil {
+		return "", false
+	}
+	if st := s.backupMgr.RestoreStatus(); restoreOpInFlight(st) {
+		subject := "Egy visszaállítási művelet"
+		if st.Stack != "" {
+			subject = "Egy visszaállítási művelet (" + st.Stack + ")"
+		}
+		return subject + " már fut, ezért most nem indítható újabb. Az állapotát ezen az oldalon " +
+			"követheted; amint befejeződik, újra indíthatsz visszaállítást.", true
+	}
+	if s.backupMgr.IsRunning() {
+		return "Egy mentési művelet már fut, ezért most nem indítható visszaállítás. Az állapotát " +
+			"ezen az oldalon követheted; amint befejeződik, újra indíthatsz visszaállítást.", true
+	}
+	return "", false
+}
+
 // backupsRestoreWizardHandler renders GET /backups/restore/app?name=<app> — the single entry the
 // list page now offers per app.
 //

controller/internal/web/templates/backups_restore_wizard.html

@@ -45,7 +45,11 @@
 <div class="settings-card">
     <h3>Végrehajtás</h3>
     <p>Jelenleg egy mentési vagy visszaállítási művelet fut{{with .RunningStack}} ({{.}}){{end}}. Amíg ez tart, új visszaállítás nem indítható.</p>
-    <p class="form-hint">Az állapot fent automatikusan frissül. A művelet befejezése után frissítsd az oldalt.</p>
+    <!-- R-351: this used to say the state refreshes automatically AND that you must refresh the page
+         when it finishes. Both cannot be true, and the second half was the one people believed. The
+         banner now carries the terminal result too (RestoreOpStatus.LastRecent), so the sentence can
+         describe what the page actually does. -->
+    <p class="form-hint">Az állapot fent automatikusan frissül, és a művelet eredménye is ott jelenik meg, amint elkészült.</p>
     <div class="form-actions">
         <a href="/backups/restore/app?name={{.App}}" class="btn btn-sm btn-outline">Állapot frissítése</a>
     </div>

controller/internal/web/templates/backups_shared.html

@@ -40,7 +40,11 @@
             banner.textContent = opLabel(st.op) + ' folyamatban' + (st.stack ? ': ' + st.stack : '') + '…';
             return;
         }
-        if (st.last && sawRunning) {
+        // R-351: `sawRunning` alone showed a terminal result ONLY to a page that watched the op
+        // happen. A restore that finished before this page was opened — or inside one poll interval
+        // — was shown to nobody, which is how a completed restore became unknowable from any screen.
+        // `last_recent` is the server's verdict, using the one window in internal/backup.
+        if (st.last && (sawRunning || st.last_recent)) {
             banner.style.display = 'block';
             if (st.last.ok) {
                 banner.className = 'flash flash-success';