v0.173.0 — R-77: endpoint-drift detection, samba protected-set gate, channel log honesty

Source: felhom.eu/documentation/audits/DIAG-agent-channel-2026-07-26.md

bootstrap.DetectEndpointDrift names a controller.yaml vs bootstrap.json
local_api.endpoint divergence -- one ERROR carrying BOTH values and BOTH paths,
its own event type local_api_endpoint_drift, and its own Hungarian banner shown
ABOVE the channel banner because drift is the cause and "agent unreachable" the
symptom. It writes NOTHING: reconciling from bootstrap.json would clobber a
correct controller.yaml on any half-provisioned or hand-repaired guest, so the
authority ruling is deferred to R-78. Fail-safe silent on absent/unparseable/
incomplete bootstrap and on an empty endpoint (ensureLocalAPI's fill-if-missing
path is untouched). Fingerprint compared as a BOOLEAN only; token never
compared, logged or exposed.

EffectiveProtected now gates samba on Enabled && UserSet, mirroring BOTH of
reconcileSambaAt's early returns, and the doc comment is corrected in the same
change -- it claimed "detection and deployment agree in both directions" while
citing only !smb.Enabled, an assertion that went false when !smb.UserSet was
added. Not over-suppressed: sharing on WITH a password and a dead container
still alarms.

Channel log: the debounce placeholder is stateUnconfirmed (rendered "unseeded")
instead of "up", so a born-down channel no longer logs "up->down" and orUnseeded
stops being dead code. Logging only -- the placeholder is still matched in the
re-arm condition, so F2 born-down alerting is byte-for-byte unchanged and all
nine pre-existing channelhealth tests pass.

Tests 951 -> 959, all green. Red-proofs A (both directions), E and F.
MinAgent unchanged; felhom-agent untouched.

CHANGELOG.md

@@ -1,5 +1,55 @@
 ## Changelog
 
+### v0.173.0 — R-77: endpoint-drift detection, samba protected-set gate, channel log honesty (2026-07-26)
+
+Source: `felhom.eu/documentation/audits/DIAG-agent-channel-2026-07-26.md`.
+
+**Operational repair first (Part 0).** Both production controllers had been dialling their pre-island
+LAN address since 2026-07-25 12:44 — the island migration rewrote `bootstrap.json` and
+`controller.yaml` was never updated. `local_api.endpoint` corrected to `169.254.253.1:8443` on
+demo-felhom and demo-hp (backups at `controller.yaml.pre-r77.bak`); **fingerprint and token agreed on
+both boxes**, so only the address moved. Channel healthy since: zero `[channel]` lines and zero
+`agent_channel_*` hub events after restart.
+
+**Endpoint-drift detection — DETECT AND NAME, never write** (`bootstrap.DetectEndpointDrift`). When
+`controller.yaml` and `bootstrap.json` both carry a complete `local_api` block and their endpoints
+disagree, the controller emits one ERROR naming **both values and both paths**, raises a **new,
+dedicated event type `local_api_endpoint_drift`** (error severity — drift never self-heals), and
+shows its own Hungarian banner *above* the channel banner, because drift is the CAUSE and
+"agent unreachable" the symptom. It **does not reconcile the files**: the mirror-image failure —
+clobbering a correct `controller.yaml` from a stale `bootstrap.json` — is just as bad, fleet-wide.
+That authority ruling is **R-78**. Fail-safe to silence on an absent/unparseable/incomplete bootstrap
+(an unprovisioned guest is not drifted) and on an empty endpoint (that is `ensureLocalAPI`'s
+fill-if-missing path, untouched). The fingerprint is compared and reported as a **boolean only**; the
+token is never compared, logged or exposed.
+
+**Hub allowlist (`felhom.eu` hub v0.74.0) — required, not optional.** `allowedEventTypes` 400s an
+unknown `event_type`, so without the one-line entry the new alert would have been silently inert —
+the exact seam-wiring failure this project has hit four times. Shipped with the controller.
+
+**Samba protected-set gate.** `EffectiveProtected` now requires `smb.Enabled && smb.UserSet`,
+mirroring **both** of `reconcileSambaAt`'s early returns. Sharing enabled without a household
+password means the controller deliberately does not deploy samba, yet the health monitor reported
+`fail` for it — demo-hp reported `health=fail` to the hub from the moment sharing was switched on.
+**The doc comment was corrected in the same change**: it claimed "detection and deployment agree in
+both directions" while citing only `!smb.Enabled`, an assertion that became false when the
+`!smb.UserSet` return was added — a comment documenting a guarantee the code no longer provides is
+how the bug returns. Not over-suppressed: sharing on **with** a password and a dead container still
+alarms. `TestEffectiveProtectedTracksSharingToggle` was updated — its old fixture asserted the buggy
+behaviour.
+
+**Channel log honesty.** The debounce branch seeded an unseeded state to `"up"`, so a **born-down**
+channel logged `up->down:<reason>` and `orUnseeded` was dead code. On 2026-07-25 that implied a
+working channel degrading when neither controller had *ever* reached its agent, and it misdirected
+the first read of the incident. The placeholder is now `stateUnconfirmed`, rendered `unseeded`.
+**Logging only** — the placeholder is still matched in the re-arm condition, so F2 born-down alerting
+is byte-for-byte unchanged; the Scenario-F test asserts sink call **count and arguments**, not just
+the string, and all nine pre-existing channelhealth tests still pass.
+
+Tests 951 → 959, all green. Three red-proofs (A, E, F) recorded in REPORT.md — Scenario A in **both**
+failure directions: no-detection, and the auto-correcting variant that trips the byte-identical
+assertion. **MinAgent unchanged; felhom-agent untouched** (DIAG refuted H1 — the island is healthy).
+
 ### v0.172.0 — R-75: canonical import root, catalog-derived skeleton, import surfaces (2026-07-26)
 
 Spike: `felhom.eu/documentation/audits/SPIKE-catalog-data-paths-2026-07-26.md`.

CONTEXT.md

@@ -7,7 +7,29 @@
 >
 > Ask Claude Code: "Please update CONTEXT.md with what we did today"
 
-Last updated: 2026-07-26 (v0.172.0 — R-75 canonical import root + catalog-derived skeleton)
+Last updated: 2026-07-26 (v0.173.0 — R-77 endpoint-drift detection + samba gate + channel log honesty)
+
+> **2026-07-26 — v0.173.0 (R-77).** Source: `audits/DIAG-agent-channel-2026-07-26.md`.
+>
+> **UNRESOLVED AND DELIBERATELY DEFERRED — which file is authoritative for `local_api`?** R-77 ships
+> DETECTION ONLY. `controller.yaml` and `bootstrap.json` can disagree; the controller dials
+> `controller.yaml`. The obvious "fix" — reconcile from `bootstrap.json` on every boot — has a failure
+> mode **as severe as the bug it fixes**: on a guest whose `controller.yaml` is correct and whose
+> `bootstrap.json` is stale (a re-provision that half-completed, a hand-repaired guest, a
+> setup-wizard box), auto-reconcile would clobber a WORKING channel on the next restart — fleet-wide,
+> silently, at the moment of a routine deploy. R-77's position is that **naming the drift is enough**:
+> it would have converted the 17.5 h outage into a specific alert on the first health cycle. The
+> authority ruling is **R-78** and needs its own spike — do not resolve it opportunistically.
+>
+> Corollary for anyone editing `bootstrap.MaybeIngest`/`ensureLocalAPI`: `ensureLocalAPI` is the ONLY
+> writer, it fires only when the endpoint is EMPTY, and `DetectEndpointDrift` must stay write-free.
+> Scenario A's test asserts `controller.yaml` is byte-identical after the check, and its red-proof
+> covers the auto-correcting variant precisely because that is the tempting wrong turn.
+>
+> **Also settled here:** the samba protected-set must mirror EVERY early return in
+> `reconcileSambaAt` (currently two: `!smb.Enabled`, `!smb.UserSet`). A third would need the same
+> mirror, and the doc comment above `EffectiveProtected` must be updated with it.
+
 
 > **2026-07-26 — v0.172.0 (R-75).** Spike `felhom.eu/documentation/audits/SPIKE-catalog-data-paths-2026-07-26.md`;
 > feature doc `felhom.eu/documentation/controller/import-and-data-paths.md`.

controller/cmd/controller/main.go

@@ -1016,6 +1016,20 @@ func main() {
 		sched.Every("agent-channel-health", 60*time.Second, chChecker.Check)
 	}
 
+	// local_api endpoint drift (R-77, from the 2026-07-25 outage): controller.yaml and bootstrap.json
+	// can disagree indefinitely and silently — the island migration rewrote the latter and the
+	// controller kept dialling the former for 17.5 h, alerting only "agent unreachable". This NAMES
+	// the fault; it deliberately does not reconcile the files (R-78 owns which one wins).
+	//
+	// Startup-only is sufficient and correct: both files are read at boot and neither changes under a
+	// running controller, so a periodic re-check would add noise without adding signal.
+	if d := bootstrap.DetectEndpointDrift(*configPath, cfg, logger); d != nil {
+		alertMgr.SetEndpointDriftAlert(true, d.HungarianMessage())
+		if notifier != nil {
+			notifier.NotifyEndpointDrift(d.EnglishMessage(), d.FingerprintAgrees)
+		}
+	}
+
 	// Wire debug callbacks (only in debug mode)
 	if cfg.Logging.Level == "debug" {
 		dc := &web.DebugCallbacks{}

controller/internal/bootstrap/bootstrap.go

@@ -320,3 +320,95 @@ func writeFileAtomic(path string, b []byte) error {
 	}
 	return os.Rename(tmp, path)
 }
+
+// --- Endpoint-drift detection (R-77, from DIAG-agent-channel-2026-07-26) ---------------------
+//
+// THE OUTAGE THIS EXISTS FOR: the R-50 island migration rewrote bootstrap.json's
+// local_api.endpoint to the island address; controller.yaml kept the pre-island LAN address on the
+// whole fleet; the agent no longer binds that address. Both controllers went dark for ~17.5 h, and
+// the only alert said "agent unreachable" — indistinguishable from a dead agent or a network blip,
+// so it read as infrastructure noise rather than a config fault. ensureLocalAPI could not catch it:
+// it fills an ABSENT local_api block and returns early on a present one, stale or not.
+//
+// This is DETECTION AND NAMING ONLY. It deliberately does NOT reconcile the two files:
+//
+//	the mirror-image failure is just as bad — on a guest whose controller.yaml is correct and whose
+//	bootstrap.json is stale, auto-reconcile would clobber a WORKING channel, fleet-wide, on the next
+//	restart. Which file is authoritative is a real, unresolved question and is tracked as R-78.
+//
+// Naming it is enough to have converted that outage into a specific, actionable alert on the first
+// health cycle, which is the whole lesson of the incident.
+
+// EndpointDrift is a detected divergence between the two local_api sources. It carries no secrets:
+// the fingerprint is reported as an agreement BOOLEAN and the token is not compared or exposed at
+// all (a token mismatch is a different failure — see the field comment).
+type EndpointDrift struct {
+	ConfigPath        string // controller.yaml
+	BootstrapPath     string // bootstrap.json
+	ConfigEndpoint    string // what the controller is actually dialling
+	BootstrapEndpoint string // what the provisioning side last wrote
+	// FingerprintAgrees is false when the pin ALSO moved. That is a materially different (and worse)
+	// situation than a moved address — fixing the endpoint alone would then fail closed on the pin —
+	// so it is surfaced, as a boolean, never as a value.
+	FingerprintAgrees bool
+}
+
+// DetectEndpointDrift compares controller.yaml's live local_api.endpoint against bootstrap.json's.
+// Returns nil (silent, no alert) in every ambiguous or not-applicable case:
+//
+//   - cfg is nil, or its endpoint is EMPTY — that is the fill-if-missing path ensureLocalAPI owns,
+//     not drift;
+//   - bootstrap.json is absent, unreadable or unparseable — a legacy / manually-configured /
+//     unprovisioned guest is not a drifted one;
+//   - the bootstrap local_api block is incomplete (any of endpoint/fingerprint/token empty) — the
+//     same completeness bar ensureLocalAPI applies before it will merge;
+//   - the endpoints agree.
+//
+// It reads two files and writes NOTHING. Emitting the ERROR here (rather than at the call site)
+// keeps the diagnosis in one line of log even when the alert path is unavailable.
+func DetectEndpointDrift(configPath string, cfg *config.Config, logger *log.Logger) *EndpointDrift {
+	if cfg == nil || cfg.LocalAPI.Endpoint == "" {
+		return nil
+	}
+	data, err := os.ReadFile(Path())
+	if err != nil {
+		return nil
+	}
+	var b Bootstrap
+	if err := json.Unmarshal(data, &b); err != nil {
+		return nil
+	}
+	if b.LocalAPI.Endpoint == "" || b.LocalAPI.Fingerprint == "" || b.LocalAPI.Token == "" {
+		return nil
+	}
+	if cfg.LocalAPI.Endpoint == b.LocalAPI.Endpoint {
+		return nil
+	}
+	d := &EndpointDrift{
+		ConfigPath:        configPath,
+		BootstrapPath:     Path(),
+		ConfigEndpoint:    cfg.LocalAPI.Endpoint,
+		BootstrapEndpoint: b.LocalAPI.Endpoint,
+		FingerprintAgrees: cfg.LocalAPI.Fingerprint == b.LocalAPI.Fingerprint,
+	}
+	if logger != nil {
+		logger.Printf("[ERROR] bootstrap: local_api endpoint DRIFT — %s says %q but %s says %q; "+
+			"the controller is dialling the FORMER. Pin agrees: %v. Not auto-corrected (R-78 owns the "+
+			"authority ruling) — fix the intended file and restart the controller.",
+			d.ConfigPath, d.ConfigEndpoint, d.BootstrapPath, d.BootstrapEndpoint, d.FingerprintAgrees)
+	}
+	return d
+}
+
+// EnglishMessage is the operator-tier alert body (operator events are English by convention).
+func (d *EndpointDrift) EnglishMessage() string {
+	return fmt.Sprintf("local_api endpoint drift: controller.yaml=%s bootstrap.json=%s (pin agrees: %v) "+
+		"— the controller is dialling controller.yaml's value; the agent may be listening on the other.",
+		d.ConfigEndpoint, d.BootstrapEndpoint, d.FingerprintAgrees)
+}
+
+// HungarianMessage is the customer-facing dashboard line, matching channelhealth's tone (short,
+// no addresses — the operator gets those in the event and the log).
+func (d *EndpointDrift) HungarianMessage() string {
+	return "A tárolókezelő ügynök címe elavult a beállításokban."
+}

controller/internal/bootstrap/drift_test.go

@@ -0,0 +1,223 @@
+package bootstrap
+
+import (
+	"bytes"
+	"encoding/json"
+	"io"
+	"log"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+
+	"gitea.dooplex.hu/admin/felhom-controller/internal/config"
+)
+
+// R-77 endpoint-drift detection. The outage this guards is DIAG-agent-channel-2026-07-26: the island
+// migration rewrote bootstrap.json, controller.yaml kept the pre-island LAN address, and the fleet's
+// controllers dialled the wrong host for 17.5 h while alerting only "agent unreachable".
+
+const driftYAML = `customer:
+    id: demo-hp
+local_api:
+    endpoint: %ENDPOINT%
+    fingerprint: aaaa1111
+    token: tok-secret
+`
+
+// writeDriftFixture lays out a controller.yaml + bootstrap.json pair and points Path() at the latter.
+func writeDriftFixture(t *testing.T, cfgEndpoint, bsEndpoint, bsFingerprint, bsToken string) (cfgPath string, cfg *config.Config) {
+	t.Helper()
+	dir := t.TempDir()
+	cfgPath = filepath.Join(dir, "controller.yaml")
+	if err := os.WriteFile(cfgPath, []byte(strings.ReplaceAll(driftYAML, "%ENDPOINT%", cfgEndpoint)), 0o600); err != nil {
+		t.Fatal(err)
+	}
+	bsPath := filepath.Join(dir, "bootstrap.json")
+	b := Bootstrap{}
+	b.LocalAPI = BootstrapLocalAPI{Endpoint: bsEndpoint, Fingerprint: bsFingerprint, Token: bsToken}
+	raw, _ := json.Marshal(b)
+	if err := os.WriteFile(bsPath, raw, 0o600); err != nil {
+		t.Fatal(err)
+	}
+	t.Setenv("FELHOM_BOOTSTRAP_PATH", bsPath)
+
+	var err error
+	cfg, err = config.LoadPermissive(cfgPath)
+	if err != nil {
+		t.Fatal(err)
+	}
+	return cfgPath, cfg
+}
+
+func sha(t *testing.T, p string) []byte {
+	t.Helper()
+	b, err := os.ReadFile(p)
+	if err != nil {
+		t.Fatal(err)
+	}
+	return b
+}
+
+// Scenario A — drift is DETECTED, NAMED, and nothing is written.
+//
+// The "nothing is written" half is the load-bearing assertion: auto-reconcile is R-78 and would
+// clobber a working channel on any guest whose controller.yaml is the correct one. "An error was
+// logged" alone would be a hollow test.
+func TestScenarioA_DriftDetectedAndNamed_NoWrite(t *testing.T) {
+	cfgPath, cfg := writeDriftFixture(t,
+		"192.168.0.87:8443",  // the live demo-hp value
+		"169.254.253.1:8443", // the island value the migration wrote
+		"aaaa1111", "tok-secret")
+	before := sha(t, cfgPath)
+
+	var buf bytes.Buffer
+	d := DetectEndpointDrift(cfgPath, cfg, log.New(&buf, "", 0))
+
+	if d == nil {
+		t.Fatal("drift must be DETECTED — this is the exact live shape from the 2026-07-25 outage")
+	}
+	if d.ConfigEndpoint != "192.168.0.87:8443" || d.BootstrapEndpoint != "169.254.253.1:8443" {
+		t.Errorf("wrong values captured: %+v", d)
+	}
+	if !d.FingerprintAgrees {
+		t.Error("fingerprints are identical in this fixture — must report agreement")
+	}
+
+	// The log line alone must diagnose it: BOTH values AND BOTH paths.
+	logged := buf.String()
+	for _, want := range []string{"192.168.0.87:8443", "169.254.253.1:8443", cfgPath, "bootstrap.json", "ERROR"} {
+		if !strings.Contains(logged, want) {
+			t.Errorf("the ERROR line must contain %q; got:\n%s", want, logged)
+		}
+	}
+	// Never log a secret.
+	if strings.Contains(logged, "tok-secret") {
+		t.Error("the token must NEVER be logged")
+	}
+
+	// THE assertion: controller.yaml is byte-identical.
+	if after := sha(t, cfgPath); !bytes.Equal(before, after) {
+		t.Errorf("controller.yaml was MODIFIED — detection must never write (that is R-78)\nbefore:\n%s\nafter:\n%s", before, after)
+	}
+}
+
+// Scenario B — agreement is silent. A spurious alert on every healthy boot would be worse than the
+// bug: it trains the operator to ignore the banner.
+func TestScenarioB_AgreementIsSilent(t *testing.T) {
+	cfgPath, cfg := writeDriftFixture(t, "169.254.253.1:8443", "169.254.253.1:8443", "aaaa1111", "tok-secret")
+	before := sha(t, cfgPath)
+	var buf bytes.Buffer
+	if d := DetectEndpointDrift(cfgPath, cfg, log.New(&buf, "", 0)); d != nil {
+		t.Errorf("matching endpoints must not report drift: %+v", d)
+	}
+	if strings.Contains(buf.String(), "ERROR") {
+		t.Errorf("no ERROR on agreement; got: %s", buf.String())
+	}
+	if after := sha(t, cfgPath); !bytes.Equal(before, after) {
+		t.Error("controller.yaml must not be touched")
+	}
+}
+
+// Scenario D — an absent / unparseable / incomplete bootstrap is NOT drift. An unprovisioned or
+// legacy guest must stay silent, not alarm.
+func TestScenarioD_IncompleteBootstrapIsFailSafe(t *testing.T) {
+	t.Run("bootstrap absent", func(t *testing.T) {
+		cfgPath, cfg := writeDriftFixture(t, "192.168.0.87:8443", "169.254.253.1:8443", "aaaa1111", "tok")
+		os.Remove(os.Getenv("FELHOM_BOOTSTRAP_PATH"))
+		if d := DetectEndpointDrift(cfgPath, cfg, log.New(io.Discard, "", 0)); d != nil {
+			t.Errorf("a missing bootstrap is not drift: %+v", d)
+		}
+	})
+	t.Run("bootstrap unparseable", func(t *testing.T) {
+		cfgPath, cfg := writeDriftFixture(t, "192.168.0.87:8443", "169.254.253.1:8443", "aaaa1111", "tok")
+		os.WriteFile(os.Getenv("FELHOM_BOOTSTRAP_PATH"), []byte("{not json"), 0o600)
+		if d := DetectEndpointDrift(cfgPath, cfg, log.New(io.Discard, "", 0)); d != nil {
+			t.Errorf("an unparseable bootstrap is not drift: %+v", d)
+		}
+	})
+	// Incomplete = any of endpoint/fingerprint/token empty — the same completeness bar ensureLocalAPI
+	// applies before it will merge.
+	for _, tc := range []struct{ name, ep, fp, tok string }{
+		{"no endpoint", "", "aaaa1111", "tok"},
+		{"no fingerprint", "169.254.253.1:8443", "", "tok"},
+		{"no token", "169.254.253.1:8443", "aaaa1111", ""},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			cfgPath, cfg := writeDriftFixture(t, "192.168.0.87:8443", tc.ep, tc.fp, tc.tok)
+			if d := DetectEndpointDrift(cfgPath, cfg, log.New(io.Discard, "", 0)); d != nil {
+				t.Errorf("an incomplete bootstrap is not drift: %+v", d)
+			}
+		})
+	}
+	t.Run("controller.yaml has no endpoint — that is the fill-if-missing path, not drift", func(t *testing.T) {
+		cfgPath, cfg := writeDriftFixture(t, `""`, "169.254.253.1:8443", "aaaa1111", "tok")
+		if cfg.LocalAPI.Endpoint != "" {
+			t.Skipf("fixture did not produce an empty endpoint (got %q)", cfg.LocalAPI.Endpoint)
+		}
+		if d := DetectEndpointDrift(cfgPath, cfg, log.New(io.Discard, "", 0)); d != nil {
+			t.Errorf("an absent local_api endpoint is ensureLocalAPI's job, not drift: %+v", d)
+		}
+	})
+	t.Run("nil cfg", func(t *testing.T) {
+		if d := DetectEndpointDrift("/nonexistent", nil, log.New(io.Discard, "", 0)); d != nil {
+			t.Error("nil cfg must be silent")
+		}
+	})
+}
+
+// A moved PIN alongside a moved address is a materially different failure — fixing the endpoint
+// alone would then fail closed on the pin. It must be surfaced, as a boolean, never as a value.
+func TestDrift_FingerprintDisagreementIsSurfacedNotLeaked(t *testing.T) {
+	cfgPath, cfg := writeDriftFixture(t, "192.168.0.87:8443", "169.254.253.1:8443", "bbbb2222", "tok-secret")
+	var buf bytes.Buffer
+	d := DetectEndpointDrift(cfgPath, cfg, log.New(&buf, "", 0))
+	if d == nil {
+		t.Fatal("drift expected")
+	}
+	if d.FingerprintAgrees {
+		t.Error("fingerprints differ in this fixture — must report DISagreement")
+	}
+	for _, secret := range []string{"aaaa1111", "bbbb2222", "tok-secret"} {
+		if strings.Contains(buf.String(), secret) {
+			t.Errorf("a fingerprint/token value leaked into the log: %q", secret)
+		}
+		if strings.Contains(d.EnglishMessage(), secret) {
+			t.Errorf("a fingerprint/token value leaked into the operator message: %q", secret)
+		}
+	}
+	if !strings.Contains(d.EnglishMessage(), "false") {
+		t.Errorf("the operator message must carry the pin-agreement boolean: %q", d.EnglishMessage())
+	}
+}
+
+// Scenario C — the fill-if-missing path is untouched by any of this (it is the ONLY writer).
+func TestScenarioC_AbsentBlockStillMerges(t *testing.T) {
+	dir := t.TempDir()
+	cfgPath := filepath.Join(dir, "controller.yaml")
+	if err := os.WriteFile(cfgPath, []byte("customer:\n    id: demo-hp\n"), 0o600); err != nil {
+		t.Fatal(err)
+	}
+	bsPath := filepath.Join(dir, "bootstrap.json")
+	b := Bootstrap{}
+	b.LocalAPI = BootstrapLocalAPI{Endpoint: "169.254.253.1:8443", Fingerprint: "aaaa1111", Token: "tok"}
+	raw, _ := json.Marshal(b)
+	os.WriteFile(bsPath, raw, 0o600)
+	t.Setenv("FELHOM_BOOTSTRAP_PATH", bsPath)
+
+	cfg, err := config.LoadPermissive(cfgPath)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if cfg.LocalAPI.Endpoint != "" {
+		t.Fatalf("precondition: fixture should have no local_api, got %q", cfg.LocalAPI.Endpoint)
+	}
+	got := ensureLocalAPI(cfgPath, cfg, log.New(io.Discard, "", 0))
+	if got == nil || got.LocalAPI.Endpoint != "169.254.253.1:8443" {
+		t.Fatalf("the fill-if-missing merge must still work; got %+v", got)
+	}
+	// And the drift check stays silent on the now-merged config.
+	if d := DetectEndpointDrift(cfgPath, got, log.New(io.Discard, "", 0)); d != nil {
+		t.Errorf("after a successful merge the two files agree — no drift: %+v", d)
+	}
+}

controller/internal/channelhealth/checker.go

@@ -105,6 +105,11 @@ func classify(constructionErr bool, err error) classification {
 	}
 }
 
+// stateUnconfirmed is the debounce placeholder for a checker that has never observed a healthy
+// probe. It is deliberately NOT "up": it must not be reported as an observation. See Check's
+// debounce branch and orUnseeded.
+const stateUnconfirmed = "unconfirmed"
+
 // Checker holds the in-memory channel state. No persistence — the state is re-derived each run
 // (mirrors the AlertManager's state-based model). Safe for the single scheduler caller; the mutex
 // guards against an overlapping run.
@@ -114,7 +119,7 @@ type Checker struct {
 	logger *log.Logger
 
 	mu              sync.Mutex
-	state           string // "" (unseeded) | "up" | "down:<reason>"
+	state           string // "" (unseeded) | stateUnconfirmed (debounce placeholder) | "up" | "down:<reason>"
 	consecutiveDown int
 	alerted         bool // have we emitted a down alert for the CURRENT down-spell? (F2: drives
 	// alerting instead of `prev==""`, so a BORN-down — broken at startup/reseed — alerts too, not
@@ -156,10 +161,17 @@ func (c *Checker) Check(ctx context.Context) error {
 	c.consecutiveDown++
 	if cls.debounce && c.consecutiveDown < debounceThreshold {
 		// A transient blip (e.g. the ~1s agent-restart socket gap, or the agent not yet up on a cold
-		// boot). Hold the previous state — do NOT flip the dashboard or notify. Unseeded → assume up
+		// boot). Hold the previous state — do NOT flip the dashboard or notify.
-		// until confirmed (so a transient born-down still needs N>=2 before it alerts).
+		//
+		// An unseeded checker is held as "not yet confirmed down" so a transient born-down still needs
+		// N>=2 before it alerts — the debounce must apply to a cold boot exactly as it does to a live
+		// blip. R-77: that hold used to be spelled `c.state = "up"`, which made a BORN-DOWN channel log
+		// `up->down` and left orUnseeded dead code. During the 2026-07-25 outage the log therefore
+		// implied a working channel degrading, when in truth neither controller had EVER reached its
+		// agent — which actively misdirected the first read of the incident. The debounce semantics are
+		// unchanged; only the state label is honest now.
 		if c.state == "" {
-			c.state = "up"
+			c.state = stateUnconfirmed
 		}
 		c.logger.Printf("[DEBUG] [channel] transient down (%s, %d/%d) — suppressed pending confirmation: %v",
 			cls.reason, c.consecutiveDown, debounceThreshold, perr)
@@ -172,7 +184,10 @@ func (c *Checker) Check(ctx context.Context) error {
 	newState := "down:" + string(cls.reason)
 	c.sink.SetDashboard(true, cls.reason, cls.hungarian) // dashboard reflects current state always
 	prev := c.state
-	if prev == "" || prev == "up" || prev != newState {
+	// stateUnconfirmed counts as unseeded for BOTH the re-arm decision and the log label: it is the
+	// debounce placeholder, never an observed up. Keeping it in this condition preserves the F2
+	// born-down alerting behaviour byte-for-byte (it used to be spelled "up" and matched here).
+	if prev == "" || prev == "up" || prev == stateUnconfirmed || prev != newState {
 		c.alerted = false
 	}
 	c.state = newState
@@ -185,8 +200,11 @@ func (c *Checker) Check(ctx context.Context) error {
 	return nil
 }
 
+// orUnseeded renders a state for the log. Both the never-observed state ("") and the debounce
+// placeholder render as "unseeded" — a born-down channel must never be logged as `up->down`, which
+// is what R-77 fixed. Before that, the placeholder was literally "up" and this function was dead code.
 func orUnseeded(s string) string {
-	if s == "" {
+	if s == "" || s == stateUnconfirmed {
 		return "unseeded"
 	}
 	return s

controller/internal/channelhealth/checker_test.go

@@ -1,10 +1,12 @@
 package channelhealth
 
 import (
+	"bytes"
 	"context"
 	"errors"
 	"io"
 	"log"
+	"strings"
 	"testing"
 )
 
@@ -272,3 +274,99 @@ func TestProbe_SeamOnly(t *testing.T) {
 		t.Fatalf("checker should call the probe seam exactly once, got %d", p.calls)
 	}
 }
+
+// R-77 Scenario F — the confirmed-down line must distinguish BORN-DOWN from a real transition.
+//
+// The debounce branch used to seed an unseeded state to "up", so a channel that had NEVER reached
+// its agent logged `up->down:unreachable` and orUnseeded was dead code. On 2026-07-25 that made the
+// log imply a working channel degrading, when in truth neither controller had ever been up — it
+// actively misdirected the first read of the incident.
+//
+// THE OTHER HALF OF THIS TEST IS THE POINT: alerting, debounce and dashboard behaviour must be
+// byte-for-byte unchanged. A logging fix that shifts alerting is a regression in a cosmetic disguise,
+// so the sink calls are asserted in COUNT and ARGUMENTS, not just the log string.
+func TestScenarioF_BornDownLogsUnseededNotUp(t *testing.T) {
+	var logbuf bytes.Buffer
+	sink := &fakeSink{}
+	c := New(nil, sink, log.New(&logbuf, "", 0))
+	// A debounced reason (connection refused → unreachable), failing from the very first probe.
+	p := &scriptedProbe{steps: []struct {
+		cons bool
+		err  error
+	}{step(false, errors.New("dial tcp 192.168.0.87:8443: connect: connection refused"))}}
+	c.probe = p.fn
+
+	// Probe 1: debounced, suppressed — no dashboard flip, no alert.
+	if err := c.Check(context.Background()); err != nil {
+		t.Fatal(err)
+	}
+	if sink.dashDown || len(sink.downs) != 0 {
+		t.Fatalf("probe 1 must be suppressed by debounce: dashDown=%v downs=%d", sink.dashDown, len(sink.downs))
+	}
+	// The placeholder must NOT be the string "up".
+	if got := c.State(); got == "up" {
+		t.Error("an unseeded checker must not report state \"up\" after a suppressed first failure")
+	}
+
+	// Probe 2: confirmed down.
+	if err := c.Check(context.Background()); err != nil {
+		t.Fatal(err)
+	}
+
+	logged := logbuf.String()
+	if !strings.Contains(logged, "unseeded->down:unreachable") {
+		t.Errorf("born-down must log \"unseeded->down:unreachable\"; got:\n%s", logged)
+	}
+	if strings.Contains(logged, "up->down") {
+		t.Errorf("a channel that was never up must NOT log \"up->down\"; got:\n%s", logged)
+	}
+
+	// --- the non-change assertions ---
+	if len(sink.downs) != 1 {
+		t.Fatalf("exactly ONE down alert must fire (F2 born-down behaviour unchanged), got %d", len(sink.downs))
+	}
+	got := sink.downs[0]
+	if got.reason != ReasonUnreachable || got.eventType != "agent_channel_unreachable" || got.severity != "warning" {
+		t.Errorf("alert arguments changed: %+v — Part 3 is a LOGGING fix only", got)
+	}
+	if !sink.dashDown || sink.dashReason != ReasonUnreachable {
+		t.Errorf("dashboard must be set down/unreachable, got down=%v reason=%q", sink.dashDown, sink.dashReason)
+	}
+	if sink.recovered != 0 {
+		t.Errorf("no recovery must fire, got %d", sink.recovered)
+	}
+	if p.calls != 2 {
+		t.Errorf("debounce threshold unchanged: expected 2 probes, got %d", p.calls)
+	}
+}
+
+// A REAL up->down transition must still log "up->down" — the fix must not relabel everything.
+func TestScenarioF_RealTransitionStillLogsUp(t *testing.T) {
+	var logbuf bytes.Buffer
+	sink := &fakeSink{}
+	c := New(nil, sink, log.New(&logbuf, "", 0))
+	p := &scriptedProbe{steps: []struct {
+		cons bool
+		err  error
+	}{
+		step(false, nil), // observed UP for real
+		step(false, errors.New("connect: connection refused")),
+		step(false, errors.New("connect: connection refused")),
+	}}
+	c.probe = p.fn
+	for i := 0; i < 3; i++ {
+		if err := c.Check(context.Background()); err != nil {
+			t.Fatal(err)
+		}
+	}
+	logged := logbuf.String()
+	if !strings.Contains(logged, "up->down:unreachable") {
+		t.Errorf("a genuine transition must still log \"up->down:unreachable\"; got:\n%s", logged)
+	}
+	if strings.Contains(logged, "unseeded->down") {
+		t.Errorf("an observed-up channel must not log \"unseeded->down\"; got:\n%s", logged)
+	}
+	if len(sink.downs) != 1 {
+		t.Errorf("exactly one down alert, got %d", len(sink.downs))
+	}
+}

controller/internal/monitor/effective_protected_test.go

@@ -42,13 +42,19 @@ func TestEffectiveProtectedDropsCloudflaredWithoutToken(t *testing.T) {
 
 // R-7b Scenario E, BOTH directions. Sharing is a customer-toggled feature, so the samba container can
 // never be in the golden controller.yaml — the effective set must add it dynamically when sharing is
-// ON (so a dead sharing service raises the same protected-container issue as a dead traefik) and must
+// live (so a dead sharing service raises the same protected-container issue as a dead traefik) and must
 // leave it out when sharing is OFF (so a box that never enabled it never reports a missing container).
-// Red-proof: delete the `if smb.Enabled` append and the enabled case fails.
+// Red-proof: delete the samba append and the enabled case fails.
+//
+// R-77 TIGHTENED THE "ON" CASE, and this test was updated with it: "on" now means
+// Enabled AND UserSet, because reconcileSambaAt refuses to deploy without a household password. The
+// previous fixture used Enabled alone and therefore asserted the very behaviour that produced the
+// live false alarm on demo-hp (2026-07-26). The three-state matrix is in
+// TestScenarioE_SambaProtectedOnlyWhenActuallyDeployed below.
 func TestEffectiveProtectedTracksSharingToggle(t *testing.T) {
 	cfg := &config.Config{Stacks: config.StacksConfig{Protected: []string{"traefik", "felhom-controller"}}}
 
-	on := EffectiveProtected(cfg, settings.SMBSettings{Enabled: true})
+	on := EffectiveProtected(cfg, settings.SMBSettings{Enabled: true, UserSet: true})
 	if !contains(on, infra.SambaContainerName) {
 		t.Errorf("sharing ON: %q must be watched, got %v", infra.SambaContainerName, on)
 	}
@@ -65,3 +71,44 @@ func TestEffectiveProtectedTracksSharingToggle(t *testing.T) {
 		}
 	}
 }
+
+// R-77 Scenario E — the samba protected-set gate must mirror reconcileSambaAt's BOTH early returns.
+//
+// The live false alarm (demo-hp, 2026-07-26): sharing was enabled without a household password, so
+// reconcileSambaAt deliberately did not deploy the container, but EffectiveProtected added it anyway
+// and the box reported health=fail for a state the controller itself had chosen.
+func TestScenarioE_SambaProtectedOnlyWhenActuallyDeployed(t *testing.T) {
+	cfg := &config.Config{Stacks: config.StacksConfig{
+		Protected: []string{"traefik", "cloudflared", "felhom-controller", "filebrowser"}}}
+	cfg.Infrastructure.CFTunnelToken = "tok" // keep cloudflared in, so the samba change is isolated
+
+	for _, tc := range []struct {
+		name string
+		smb  settings.SMBSettings
+		want bool
+		why  string
+	}{
+		{"(1) sharing OFF", settings.SMBSettings{Enabled: false, UserSet: false}, false,
+			"a box that never enabled sharing must stay quiet"},
+		{"(2) sharing ON, no password", settings.SMBSettings{Enabled: true, UserSet: false}, false,
+			"THE BUG: reconcileSambaAt refuses to deploy without a password — a deliberate state, not a fault"},
+		{"(3) sharing ON, password set", settings.SMBSettings{Enabled: true, UserSet: true}, true,
+			"the container really should be running — a dead one must STILL alarm (do not over-suppress)"},
+	} {
+		if got := contains(EffectiveProtected(cfg, tc.smb), infra.SambaContainerName); got != tc.want {
+			t.Errorf("%s: samba protected = %v, want %v — %s", tc.name, got, tc.want, tc.why)
+		}
+	}
+
+	// Not over-suppressed: the rest of the protected set is untouched in every sharing state.
+	for _, smb := range []settings.SMBSettings{
+		{Enabled: false}, {Enabled: true}, {Enabled: true, UserSet: true},
+	} {
+		set := EffectiveProtected(cfg, smb)
+		for _, must := range []string{"traefik", "cloudflared", "felhom-controller", "filebrowser"} {
+			if !contains(set, must) {
+				t.Errorf("smb=%+v: %q must always stay protected, got %v", smb, must, set)
+			}
+		}
+	}
+}

controller/internal/monitor/healthcheck.go

@@ -251,14 +251,28 @@ func checkDocker() error {
 //
 //   - cloudflared is dropped when no tunnel token is configured (a LAN-only node legitimately runs
 //     without it, so it must not be reported as a missing protected container forever);
-//   - the samba container is ADDED when network sharing is switched on (R-7b). Sharing is a
+//   - the samba container is ADDED only when network sharing is switched on AND the household
-//     customer-toggled feature, so it can never appear in the golden controller.yaml — but once it
+//     password has been set (R-7b, tightened by R-77). Sharing is a customer-toggled feature, so it
-//     IS on, a dead sharing service is exactly as customer-visible as a dead traefik and must raise
+//     can never appear in the golden controller.yaml — but once it is actually RUNNING, a dead
-//     the same protected-container issue → alert → Hungarian degradation e-mail. When sharing is
+//     sharing service is exactly as customer-visible as a dead traefik and must raise the same
-//     off the container is absent from the set, so a box that never enabled it stays quiet.
+//     protected-container issue → alert → Hungarian degradation e-mail.
 //
-// The bring-up applies the same conditions (stacks.EnsureBaseStack for cloudflared, ensureSamba's
+// THE COUPLING, and why it is spelled out: this set must mirror EVERY early return in
-// `if !smb.Enabled { return }` for samba), so detection and deployment agree in both directions.
+// stacks.reconcileSambaAt, because that function decides whether the container exists at all. It has
+// TWO:
+//
+//	if !smb.Enabled  { return }   // feature off
+//	if !smb.UserSet  { return }   // on, but no household password yet → deliberately NOT deployed
+//
+// R-77 exists because this comment previously claimed "detection and deployment agree in both
+// directions" while citing only the first. The second was added later and never mirrored here, so
+// enabling sharing without setting a password made the box report health=fail forever for a state
+// the controller had deliberately chosen (observed live on demo-hp, 2026-07-26). A THIRD early
+// return in reconcileSambaAt would need the same mirror — and this comment must be updated with it,
+// because a comment asserting a guarantee the code no longer provides is how the bug came back.
+//
+// Deliberately NOT over-suppressed: sharing on WITH a password and a dead container still raises the
+// issue. That is the case the protected set exists for.
 //
 // NOTE: the entries are CONTAINER names (checkProtectedContainers docker-inspects them). For the
 // base stacks the container name happens to equal the stack name; for samba it does NOT — the stack
@@ -272,7 +286,8 @@ func EffectiveProtected(cfg *config.Config, smb settings.SMBSettings) []string {
 		}
 		out = append(out, name)
 	}
-	if smb.Enabled {
+	// Mirrors reconcileSambaAt's two early returns — see the coupling note above.
+	if smb.Enabled && smb.UserSet {
 		out = append(out, infra.SambaContainerName)
 	}
 	return out

controller/internal/notify/notifier.go

@@ -375,6 +375,24 @@ func (n *Notifier) NotifyAgentChannelRecovered() {
 		"Controller→agent channel recovered — local-API reachable again.", nil)
 }
 
+// NotifyEndpointDrift reports a local_api endpoint divergence (R-77). Operator-only English, its
+// OWN event type — deliberately not folded into agent_channel_*, because during the 2026-07-25
+// outage the generic channel alert was the only signal and it hid a specific, fixable config fault.
+// severity=error: unlike a transient unreachable, drift never self-heals.
+//
+// NOTE: the hub validates event_type against allowedEventTypes and 400s an unknown one, so this
+// type MUST exist there too (hub handler.go) or the alert is silently inert.
+func (n *Notifier) NotifyEndpointDrift(message string, fingerprintAgrees bool) {
+	n.PushEvent("local_api_endpoint_drift", "error", message,
+		EndpointDriftDetails{FingerprintAgrees: fingerprintAgrees})
+}
+
+// EndpointDriftDetails carries NO addresses and NO secrets — the endpoints are in the message, and
+// the pin is a boolean by design.
+type EndpointDriftDetails struct {
+	FingerprintAgrees bool `json:"fingerprint_agrees"`
+}
+
 // NotifyAppDeployed sends an app deployment event.
 func (n *Notifier) NotifyAppDeployed(stackName, displayName string) {
 	n.PushEvent("app_deployed", "info",

controller/internal/web/alerts.go

@@ -42,6 +42,10 @@ type AlertManager struct {
 	// out-of-band, state-based, self-clearing model as agentChannelAlert: passing an empty slice when
 	// every deployed app is running clears the banner with no manual dismissal.
 	deadAppAlerts []Alert
+	// endpointDriftAlert (R-77) is set/cleared at startup by the local_api drift check. Separate from
+	// agentChannelAlert on purpose: drift is usually the CAUSE and "agent unreachable" the SYMPTOM,
+	// and during the 2026-07-25 outage only the symptom was visible.
+	endpointDriftAlert *Alert
 }
 
 // NewAlertManager creates a new AlertManager.
@@ -77,6 +81,28 @@ func (am *AlertManager) SetAgentChannelAlert(down bool, msg string) {
 	}
 }
 
+// SetEndpointDriftAlert sets (drift=true) or clears the local_api endpoint-drift banner (R-77).
+//
+// It is deliberately a SEPARATE alert from SetAgentChannelAlert: during the 2026-07-25 outage the
+// generic "agent unreachable" banner was the ONLY signal, and it looked like a dead agent. The two
+// can also be true at once — a drifted endpoint usually CAUSES the channel to be down — so folding
+// them together would hide the actionable one behind the symptom.
+func (am *AlertManager) SetEndpointDriftAlert(drift bool, msg string) {
+	am.mu.Lock()
+	defer am.mu.Unlock()
+	if !drift {
+		am.endpointDriftAlert = nil
+		return
+	}
+	am.endpointDriftAlert = &Alert{
+		ID:       "local-api-endpoint-drift",
+		Level:    "error",
+		Message:  msg,
+		Link:     "/settings",
+		LinkText: "Beállítások",
+	}
+}
+
 // DeadApp is a deployed app the health loop found not-running (fix-3). State is the container-state
 // string for the display (e.g. "stopped"/"exited").
 type DeadApp struct {
@@ -244,7 +270,7 @@ func (am *AlertManager) GetAlerts(excludeIDs ...string) []Alert {
 	am.mu.RLock()
 	defer am.mu.RUnlock()
 
-	if len(am.alerts) == 0 && am.agentChannelAlert == nil && len(am.deadAppAlerts) == 0 {
+	if len(am.alerts) == 0 && am.agentChannelAlert == nil && am.endpointDriftAlert == nil && len(am.deadAppAlerts) == 0 {
 		return nil
 	}
 
@@ -254,6 +280,12 @@ func (am *AlertManager) GetAlerts(excludeIDs ...string) []Alert {
 	}
 
 	var result []Alert
+	// Endpoint drift first: it is the actionable CAUSE, and the channel-down banner below is usually
+	// just its symptom. Showing the symptom above the cause is what made the 2026-07-25 outage read
+	// as an infrastructure blip for 17.5 h.
+	if am.endpointDriftAlert != nil && !exclude[am.endpointDriftAlert.ID] {
+		result = append(result, *am.endpointDriftAlert)
+	}
 	// Channel-down is prepended (highest priority — the agent link being dead breaks disk/storage UI).
 	if am.agentChannelAlert != nil && !exclude[am.agentChannelAlert.ID] {
 		result = append(result, *am.agentChannelAlert)