Make delivery and integration failures explicit

Persist reminder presentations and retry state, atomically complete collapsed deliveries, fall back across away reaches, and block permanent failures visibly (V-715, V-678). Fail closed when enabled integrations lack credentials and keep remote arms explicitly dark (V-691). Give mavweb one sanitized, request-correlated error contract (V-689). Owner explicitly requested direct commits to master.

cmd/mavend/main.go

@@ -559,7 +559,7 @@ func run(args []string) error {
 func personaFacts(cfg *config.Config) persona.Facts {
 	f := persona.Facts{
 		// Telegram lives outside the voice block, so it counts either way.
-		Telegram: cfg.Telegram != nil && cfg.Telegram.BotToken != "" && cfg.Telegram.ChatID != "",
+		Telegram: cfg.Telegram != nil && !cfg.Telegram.Disabled && cfg.Telegram.BotToken != "" && cfg.Telegram.ChatID != "",
 	}
 	if cfg.Voice == nil {
 		return f
@@ -678,16 +678,12 @@ func wireGatherer(st *store.Store, cfg *config.Config, rules []loop.Rule) *loop.
 // reconciled to "unknown" here, before the tick loop resumes sending, so
 // nothing auto-resends into that ambiguity.
 func wireDispatcher(st *store.Store, cfg *config.Config, voiceW *voiceWiring) (*delivery.Dispatcher, error) {
-	var ntfy delivery.Sink
-	if cfg.Ntfy != nil {
-		s, err := ntfysink.New(*cfg.Ntfy)
-		if err != nil {
-			return nil, fmt.Errorf("wire ntfy sink: %w", err)
-		}
-		ntfy = s
+	ntfy, err := wireNtfySink(cfg.Ntfy)
+	if err != nil {
+		return nil, err
 	}
 	var telegram delivery.Sink
-	if cfg.Telegram != nil {
+	if cfg.Telegram != nil && !cfg.Telegram.Disabled {
 		s, err := telegramsink.New(*cfg.Telegram)
 		if err != nil {
 			return nil, fmt.Errorf("wire telegram sink: %w", err)
@@ -712,6 +708,22 @@ func wireDispatcher(st *store.Store, cfg *config.Config, voiceW *voiceWiring) (*
 	}), nil
 }
 
+// wireNtfySink keeps an optional reach optional without ever turning a missing
+// secret into anonymous publishing. A block is live unless it says disabled;
+// therefore an expanded-empty token in a live block fails startup instead of
+// spending days in a permanent 403 retry loop. Disabled is an explicit
+// operator choice and lets another away reach take over.
+func wireNtfySink(cfg *ntfysink.Config) (delivery.Sink, error) {
+	if cfg == nil || cfg.Disabled {
+		return nil, nil
+	}
+	sink, err := ntfysink.New(*cfg)
+	if err != nil {
+		return nil, fmt.Errorf("wire ntfy sink: %w", err)
+	}
+	return sink, nil
+}
+
 // wireTickLoop reads the loop's three intervals and its schedules out of the
 // config, so the two boot paths cannot disagree about them.
 func wireTickLoop(st *store.Store, gatherer *loop.Gatherer, dispatcher *delivery.Dispatcher, phr phraser.Phraser, rules []loop.Rule, cfg *config.Config) *tickLoop {

cmd/mavend/ntfy_wiring_test.go

@@ -0,0 +1,35 @@
+package main
+
+import (
+	"testing"
+
+	"github.com/kami/maven/internal/delivery/ntfysink"
+)
+
+func TestWireNtfySinkRejectsMissingCredentialWhenEnabled(t *testing.T) {
+	_, err := wireNtfySink(&ntfysink.Config{
+		BaseURL: "https://ntfy.example", Topic: "maven",
+	})
+	if err == nil {
+		t.Fatal("expanded-empty credential did not fail an enabled reach")
+	}
+}
+
+func TestWireNtfySinkLeavesExplicitlyDisabledReachDark(t *testing.T) {
+	sink, err := wireNtfySink(&ntfysink.Config{
+		Disabled: true, BaseURL: "https://ntfy.example", Topic: "maven",
+	})
+	if err != nil {
+		t.Fatalf("wireNtfySink: %v", err)
+	}
+	if sink != nil {
+		t.Fatal("disabled reach built a live sink")
+	}
+}
+
+func TestWireNtfySinkRejectsMalformedEnabledConfig(t *testing.T) {
+	_, err := wireNtfySink(&ntfysink.Config{Token: "token", Topic: "maven"})
+	if err == nil {
+		t.Fatal("malformed enabled config did not fail wiring")
+	}
+}

cmd/mavend/telegramintake.go

@@ -25,7 +25,7 @@ import (
 // already failed the boot in wireDispatcher for the same config, so a second
 // hard failure would only lose that message.
 func wireTelegramIntake(ctx context.Context, wg *sync.WaitGroup, api ipc.CoreAPI, cfg *config.Config) {
-	if cfg == nil || cfg.Telegram == nil || !cfg.Telegram.Intake || api == nil {
+	if cfg == nil || cfg.Telegram == nil || cfg.Telegram.Disabled || !cfg.Telegram.Intake || api == nil {
 		return
 	}
 	sink, err := telegramsink.New(*cfg.Telegram)

cmd/mavend/tick.go

@@ -10,11 +10,13 @@ package main
 
 import (
 	"context"
+	"crypto/sha256"
 	"errors"
 	"fmt"
 	"log"
 	"os"
 	"path/filepath"
+	"sort"
 	"sync"
 	"time"
 
@@ -226,18 +228,11 @@ func (t *tickLoop) tick(ctx context.Context, now time.Time) {
 	// detectPatterns below for how idempotence and dismissal are respected.
 	t.detectPatterns(ctx, now, state)
 
-	// reminders: gate-bypassing class. fired once, marked after a successful
-	// delivery.  a failed send leaves the reminder pending — the next tick
-	// re-gathers and re-attempts.
+	// reminders: gate-bypassing class. The presentation and retry clock live on
+	// the reminder occurrence, so a transport outage neither spends the model
+	// every tick nor changes what the reminder says after a restart.
 	for _, d := range loop.RemindDecisions(state, due) {
-		pr, err := t.phraser.PhraseReminder(ctx, d)
-		if err != nil {
-			log.Printf("tick: phrase reminder %d: %v", d.Reminder.ID, err)
-			continue
-		}
-		if _, err := t.dispatcher.DispatchReminder(ctx, pr, now); err != nil {
-			log.Printf("tick: dispatch reminder %d: %v", d.Reminder.ID, err)
-		}
+		t.deliverReminder(ctx, d, now)
 	}
 
 	// sev4-away repeats: re-send un-acked telegram nudges per repeatInterval.
@@ -263,6 +258,125 @@ func (t *tickLoop) tick(ctx context.Context, now time.Time) {
 	}
 }
 
+// deliverReminder advances one due reminder (or collapsed bundle) through the
+// durable delivery state. A phrase is cached before the first external send;
+// every definite failure advances the persisted bounded backoff.
+func (t *tickLoop) deliverReminder(ctx context.Context, d loop.ReminderDecision, now time.Time) {
+	originals := reminderOriginals(d.Reminder)
+	pr, cached := cachedReminderPhrase(d, originals)
+	if !cached {
+		var err error
+		pr, err = t.phraser.PhraseReminder(ctx, d)
+		if err == nil && pr.Body == "" {
+			err = errors.New("phraser returned an empty reminder body")
+		}
+		if err != nil {
+			log.Printf("tick: phrase reminder %d: %v", d.Reminder.ID, err)
+			t.scheduleReminderRetry(ctx, originals, now)
+			return
+		}
+		if pr.Mood == "" {
+			pr.Mood = "neutral"
+		}
+		group := reminderDeliveryGroup(originals)
+		if err := t.store.CacheReminderPhrase(
+			ctx, originals, group, pr.Body, pr.Summary, pr.Mood,
+		); err != nil {
+			// A cancellation or another completion can win while phrasing. Do
+			// not send a presentation that no longer owns every original.
+			log.Printf("tick: cache reminder %d phrase: %v", d.Reminder.ID, err)
+			return
+		}
+		// The store now owns the phrase, but this tick's value predates that
+		// write. Stamp the exact persisted occurrence identity onto the value
+		// handed to the dispatcher so its outbox row can suppress an ambiguous
+		// crash for both a real reminder and a synthetic collapsed bundle.
+		for i := range originals {
+			originals[i].DeliveryGroup = group
+			originals[i].PhraseBody = pr.Body
+			originals[i].PhraseSummary = pr.Summary
+			originals[i].PhraseMood = pr.Mood
+		}
+		if d.Reminder.ID == 0 {
+			d.Reminder.Collapsed = originals
+		} else {
+			d.Reminder = originals[0]
+		}
+	}
+	// A phraser is not allowed to substitute the reminder decision. In
+	// particular, the durable group stamped above must reach the outbox.
+	pr.Decision = d
+
+	if _, err := t.dispatcher.DispatchReminder(ctx, pr, now); err != nil {
+		log.Printf("tick: dispatch reminder %d: %v", d.Reminder.ID, err)
+		t.scheduleReminderRetry(ctx, originals, now)
+	}
+}
+
+func (t *tickLoop) scheduleReminderRetry(ctx context.Context, originals []store.Reminder, now time.Time) {
+	if err := t.store.ScheduleReminderRetry(ctx, originals, now); err != nil {
+		log.Printf("tick: schedule reminder retry: %v", err)
+	}
+}
+
+// reminderOriginals converts the synthetic ID=0 bundle back to real store
+// rows. Keeping this in one helper makes it impossible to accidentally persist
+// retry state against reminder zero.
+func reminderOriginals(r store.Reminder) []store.Reminder {
+	if r.ID == 0 {
+		return append([]store.Reminder(nil), r.Collapsed...)
+	}
+	return []store.Reminder{r}
+}
+
+// cachedReminderPhrase reconstructs a PhrasedReminder only when every original
+// agrees on one persisted group and presentation. That agreement is what lets
+// a collapsed bundle survive a restart without being re-phrased.
+func cachedReminderPhrase(d loop.ReminderDecision, originals []store.Reminder) (delivery.PhrasedReminder, bool) {
+	if len(originals) == 0 || !originals[0].HasDeliveryPhrase() {
+		return delivery.PhrasedReminder{}, false
+	}
+	first := originals[0]
+	for _, r := range originals[1:] {
+		if !r.HasDeliveryPhrase() ||
+			r.DeliveryGroup != first.DeliveryGroup ||
+			r.PhraseBody != first.PhraseBody ||
+			r.PhraseSummary != first.PhraseSummary ||
+			r.PhraseMood != first.PhraseMood {
+			return delivery.PhrasedReminder{}, false
+		}
+	}
+	mood := first.PhraseMood
+	if mood == "" {
+		mood = "neutral"
+	}
+	return delivery.PhrasedReminder{
+		Decision: d,
+		Body:     first.PhraseBody,
+		Summary:  first.PhraseSummary,
+		Mood:     mood,
+	}, true
+}
+
+// reminderDeliveryGroup deterministically names one occurrence or collapsed
+// set. The next-fire instant is part of the identity so a recurring reminder's
+// later occurrence can never inherit the previous occurrence's phrase.
+func reminderDeliveryGroup(originals []store.Reminder) string {
+	ordered := append([]store.Reminder(nil), originals...)
+	sort.Slice(ordered, func(i, j int) bool {
+		if ordered[i].ID == ordered[j].ID {
+			return ordered[i].NextFireTs.Before(ordered[j].NextFireTs)
+		}
+		return ordered[i].ID < ordered[j].ID
+	})
+	h := sha256.New()
+	for _, r := range ordered {
+		_, _ = fmt.Fprintf(h, "%d:%d;", r.ID, r.NextFireTs.UnixMilli())
+	}
+	sum := h.Sum(nil)
+	return fmt.Sprintf("reminder:%x", sum[:12])
+}
+
 // savePresence writes back the bucket GatherState just resolved.
 //
 // It lives here and not in GatherState because that method holds a read-only

cmd/mavend/tick_reminder_delivery_test.go

@@ -0,0 +1,170 @@
+package main
+
+import (
+	"context"
+	"errors"
+	"testing"
+	"time"
+
+	"github.com/kami/maven/internal/delivery"
+	"github.com/kami/maven/internal/loop"
+	"github.com/kami/maven/internal/phraser"
+	"github.com/kami/maven/internal/store"
+)
+
+type reminderCountingPhraser struct {
+	phraser.Phraser
+	calls   int
+	body    string
+	summary string
+	mood    string
+}
+
+func (p *reminderCountingPhraser) PhraseReminder(_ context.Context, d loop.ReminderDecision) (delivery.PhrasedReminder, error) {
+	p.calls++
+	return delivery.PhrasedReminder{
+		Decision: d,
+		Body:     p.body,
+		Summary:  p.summary,
+		Mood:     p.mood,
+	}, nil
+}
+
+type reminderFailSink struct {
+	sends int
+}
+
+func (s *reminderFailSink) Send(_ context.Context, _ delivery.Sendable) error {
+	s.sends++
+	return errors.New("transport unavailable")
+}
+
+func newReminderDeliveryLoop(t *testing.T, st *store.Store, sink delivery.Sink, p phraser.Phraser) *tickLoop {
+	t.Helper()
+	rules := loop.DefaultRules()
+	return newTickLoop(
+		st,
+		loop.NewGatherer(st, rules),
+		delivery.NewDispatcher(delivery.Config{
+			Voice: sink, Ntfy: sink, Telegram: sink,
+			Nudges: st, Reminders: st, Outbox: st,
+		}),
+		p,
+		rules,
+		time.Second,
+		5*time.Minute,
+		0,
+		nil, nil, nil, nil,
+	)
+}
+
+func TestTickReminderRetryUsesPersistedPhraseAfterRestart(t *testing.T) {
+	st := newTestStore(t)
+	ctx := context.Background()
+	now := refNow()
+	if _, err := st.CreateReminder(ctx, now.Add(-time.Minute), `{"text":"позвонить маме"}`, ""); err != nil {
+		t.Fatal(err)
+	}
+
+	fail := &reminderFailSink{}
+	firstPhraser := &reminderCountingPhraser{
+		Phraser: phraser.NewStub(), body: "Не забудь позвонить маме.",
+		summary: "Позвонить маме", mood: "warm",
+	}
+	tl := newReminderDeliveryLoop(t, st, fail, firstPhraser)
+	tl.tick(ctx, now)
+	if firstPhraser.calls != 1 {
+		t.Fatalf("first tick phrased %d times, want 1", firstPhraser.calls)
+	}
+	rows, err := st.ListReminders(ctx, 1)
+	if err != nil || len(rows) != 1 {
+		t.Fatalf("list = %d, err=%v", len(rows), err)
+	}
+	if !rows[0].HasDeliveryPhrase() || rows[0].DeliveryAttempts != 1 {
+		t.Fatalf("failed delivery state was not persisted: %+v", rows[0])
+	}
+	if want := now.Add(store.ReminderRetryBase); !rows[0].NextAttemptTs.Equal(want) {
+		t.Fatalf("next attempt = %s, want %s", rows[0].NextAttemptTs, want)
+	}
+
+	// A normal tick inside the wait does no transport work and no model work.
+	sendsAfterFirst := fail.sends
+	tl.tick(ctx, now.Add(30*time.Second))
+	if firstPhraser.calls != 1 || fail.sends != sendsAfterFirst {
+		t.Fatalf("retry wait did work: phrase calls=%d, sends=%d (was %d)", firstPhraser.calls, fail.sends, sendsAfterFirst)
+	}
+
+	// Constructing a new loop is the daemon-restart boundary. Its phraser would
+	// say something different if called; the stored phrase must win instead.
+	success := &fakeSink{}
+	afterRestart := &reminderCountingPhraser{
+		Phraser: phraser.NewStub(), body: "WRONG NEW PHRASE", summary: "WRONG", mood: "neutral",
+	}
+	restarted := newReminderDeliveryLoop(t, st, success, afterRestart)
+	restarted.tick(ctx, now.Add(store.ReminderRetryBase))
+	if afterRestart.calls != 0 {
+		t.Fatalf("restart re-phrased the reminder %d times", afterRestart.calls)
+	}
+	if len(success.sends) != 1 {
+		t.Fatalf("retry sends = %d, want 1", len(success.sends))
+	}
+	if got := success.sends[0].Body; got != "Позвонить маме" {
+		t.Fatalf("away retry body = %q, want persisted summary", got)
+	}
+	rows, err = st.ListReminders(ctx, 1)
+	if err != nil || rows[0].Status != store.ReminderFired {
+		t.Fatalf("successful retry did not fire reminder: rows=%+v err=%v", rows, err)
+	}
+}
+
+func TestTickCollapsedReminderRetriesOnePhraseAndCompletesOriginals(t *testing.T) {
+	st := newTestStore(t)
+	ctx := context.Background()
+	now := refNow()
+	for _, text := range []string{"полить цветы", "записаться к врачу"} {
+		if _, err := st.CreateReminder(ctx, now.Add(-time.Minute), text, ""); err != nil {
+			t.Fatal(err)
+		}
+	}
+
+	fail := &reminderFailSink{}
+	firstPhraser := &reminderCountingPhraser{
+		Phraser: phraser.NewStub(), body: "У тебя два напоминания.",
+		summary: "Два напоминания", mood: "neutral",
+	}
+	newReminderDeliveryLoop(t, st, fail, firstPhraser).tick(ctx, now)
+	if firstPhraser.calls != 1 {
+		t.Fatalf("collapsed bundle phrased %d times, want 1", firstPhraser.calls)
+	}
+	rows, err := st.ListReminders(ctx, 10)
+	if err != nil || len(rows) != 2 {
+		t.Fatalf("list = %d, err=%v", len(rows), err)
+	}
+	for _, r := range rows {
+		if r.DeliveryGroup == "" || r.DeliveryGroup != rows[0].DeliveryGroup ||
+			r.PhraseBody != "У тебя два напоминания." || r.DeliveryAttempts != 1 {
+			t.Fatalf("collapsed original lost shared state: %+v", r)
+		}
+	}
+
+	success := &fakeSink{}
+	afterRestart := &reminderCountingPhraser{
+		Phraser: phraser.NewStub(), body: "WRONG", summary: "WRONG", mood: "neutral",
+	}
+	newReminderDeliveryLoop(t, st, success, afterRestart).tick(ctx, now.Add(store.ReminderRetryBase))
+	if afterRestart.calls != 0 {
+		t.Fatalf("collapsed retry re-phrased %d times", afterRestart.calls)
+	}
+	if len(success.sends) != 1 || success.sends[0].ReminderID != 0 {
+		t.Fatalf("collapsed retry sends = %+v, want one synthetic delivery", success.sends)
+	}
+	rows, err = st.ListReminders(ctx, 10)
+	if err != nil {
+		t.Fatal(err)
+	}
+	for _, r := range rows {
+		if r.Status != store.ReminderFired {
+			t.Fatalf("collapsed original %d status = %q, want fired", r.ID, r.Status)
+		}
+	}
+}

cmd/mavend/voicewire.go

@@ -377,19 +377,19 @@ func wireVoice(cfg *config.Config, coreAPI ipc.CoreAPI, phr phraser.Phraser, mem
 // degraded mode, so the seam is nil and the cascade routes with the classifier.
 func modelSeam(cfg *config.Config, resident *llm.Client) (router.Completer, *llm.Pair) {
 	if resident == nil {
-		if cfg.Workstation != nil {
+		if cfg.Workstation != nil && !cfg.Workstation.ModelDisabled {
 			log.Printf("voice: a workstation is configured but there is no resident model to floor it with — ignoring the block")
 		}
 		return nil, nil
 	}
-	if cfg.Workstation == nil {
+	if cfg.Workstation == nil || cfg.Workstation.ModelDisabled {
 		return resident, nil
 	}
 	ws := cfg.Workstation
 	remote := llm.New(ws.URL, time.Duration(ws.Timeout))
 	remote.SetToken(ws.Token)
 	if ws.Token == "" {
-		log.Printf("voice: no workstation.token — mavgpud refuses an unauthenticated request, so this reads as a card that is always busy")
+		log.Printf("voice: unauthenticated workstation model endpoint is loopback-only")
 	}
 	pair := llm.NewPair(
 		remote,
@@ -434,7 +434,7 @@ func sttSeam(cfg *config.Config, floor stt.Transcriber) (stt.Transcriber, *stt.P
 	)
 	pair.Start(context.Background())
 	if s.Token == "" {
-		log.Print("voice: the workstation transcriber has no token, so anything on the LAN can post audio to it")
+		log.Print("voice: unauthenticated workstation transcriber endpoint is loopback-only")
 	}
 	log.Printf("voice: workstation transcriber at %s, probed every %s, mavsttd as the floor",
 		s.URL, time.Duration(s.Probe))