From b46192a43d60a85ad4c830516bb33549100ca0d6 Mon Sep 17 00:00:00 2001
From: dev_agent <dev_agent@cubecraftlabs.com>
Date: Fri, 9 Oct 2026 18:17:41 -0400
Subject: [PATCH 1/3] Add the scheduler core with period keys and catch-up

internal/scheduler runs registered schedules (a name, a daily or weekly
rule read from the household settings, and a run function) in the
household zone with an injectable clock. Each fire time's period is
claimed by inserting schedule_runs(schedule, period_key) with ON
CONFLICT DO NOTHING, so a period runs at most once across restarts and
instances. At boot only each schedule's most recent fire time is
considered: within 24 h it runs once, older it is claimed as missed so
it never runs later. Settings changes move the next fire time through
SettingsService.Subscribe. DST follows household.LocalTime. The server
starts it with the worker context and waits for it before closing the
database; the concrete schedules come in later Feature 20 tasks.

Maestro task f7024755-c3d3-46f3-b798-e242ad676c1e.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---
 cmd/server/main.go                   |  16 +-
 docs/ARCHITECTURE.md                 |  14 +
 internal/db/querier.go               |   3 +
 internal/db/schedule_runs.sql.go     |  38 ++
 internal/scheduler/export_test.go    |   9 +
 internal/scheduler/scheduler.go      | 502 ++++++++++++++++++++
 internal/scheduler/scheduler_test.go | 653 +++++++++++++++++++++++++++
 migrations/queries/schedule_runs.sql |   6 +
 8 files changed, 1239 insertions(+), 2 deletions(-)
 create mode 100644 internal/db/schedule_runs.sql.go
 create mode 100644 internal/scheduler/export_test.go
 create mode 100644 internal/scheduler/scheduler.go
 create mode 100644 internal/scheduler/scheduler_test.go
 create mode 100644 migrations/queries/schedule_runs.sql

diff --git a/cmd/server/main.go b/cmd/server/main.go
index 7103695..a97ca26 100644
--- a/cmd/server/main.go
+++ b/cmd/server/main.go
@@ -30,6 +30,7 @@ import (
 	"code.cubecraftlabs.com/Maniacs_Mansion/Meal-Quest/internal/plan"
 	"code.cubecraftlabs.com/Maniacs_Mansion/Meal-Quest/internal/push"
 	"code.cubecraftlabs.com/Maniacs_Mansion/Meal-Quest/internal/recipes"
+	"code.cubecraftlabs.com/Maniacs_Mansion/Meal-Quest/internal/scheduler"
 	"code.cubecraftlabs.com/Maniacs_Mansion/Meal-Quest/internal/secretbox"
 	"code.cubecraftlabs.com/Maniacs_Mansion/Meal-Quest/internal/server"
 	"code.cubecraftlabs.com/Maniacs_Mansion/Meal-Quest/internal/shopping"
@@ -187,17 +188,23 @@ func run(cfg config.Config, logger *slog.Logger) error {
 		return err
 	}
 
+	sched, err := scheduler.New(db, settings, logger)
+	if err != nil {
+		return err
+	}
+
 	router, err := server.NewRouter(logger, version, cfg.TrustedProxy, authFlow, sessions, settings, search, ai, chatSvc, dinnerSvc, suggestSvc, planSvc, shopSvc, googleSvc, pushSvc, ideasSvc, runner, photos, frontend.Assets())
 	if err != nil {
 		return err
 	}
 
 	// Start fails the jobs a previous process left running before any
-	// worker runs. The job workers and the notification worker stop, and
-	// are waited for, before the database closes.
+	// worker runs. The job workers, the notification worker and the
+	// scheduler stop, and are waited for, before the database closes.
 	workersCtx, stopWorkers := context.WithCancel(ctx)
 	defer func() {
 		stopWorkers()
+		sched.Wait()
 		runner.Wait()
 		notifier.Wait()
 	}()
@@ -213,6 +220,11 @@ func run(cfg config.Config, logger *slog.Logger) error {
 	if _, err := dinnerSvc.FailInterruptedGroundings(ctx); err != nil {
 		return err
 	}
+	// The scheduler starts last: its catch-up may enqueue jobs and
+	// notifications, so their workers must already run.
+	if err := sched.Start(workersCtx); err != nil {
+		return err
+	}
 	srv := &http.Server{
 		Addr:              cfg.HTTPAddr,
 		Handler:           router,
diff --git a/docs/ARCHITECTURE.md b/docs/ARCHITECTURE.md
index 0ebae68..190022c 100644
--- a/docs/ARCHITECTURE.md
+++ b/docs/ARCHITECTURE.md
@@ -785,6 +785,20 @@ purge are fixed and not configurable.
 
 Runs are recorded in `schedule_runs` and executed through `jobs`.
 
+`internal/scheduler` runs them. A schedule registers a name, a rule read
+from the settings (`scheduler.Daily`, `Weekly` or `WeeklyForNextWeek`) and a
+run function; settings changes reach it through `SettingsService.Subscribe`
+and move the next fire time without a restart. Each fire time has a period
+key (`Rule.PeriodKey`): the local date for a daily rule, the ISO week of the
+fire date for a weekly one, or of the week after for `WeeklyForNextWeek`
+(`plan_draft`). Inserting `schedule_runs(schedule, period_key)` is the claim;
+a period already claimed is skipped, so a period runs at most once and a
+failed run is not retried. At boot only each schedule's most recent fire
+time is considered: it runs if it passed within 24 hours, and is otherwise
+claimed without running (logged as missed) so it never runs later. DST
+follows `household.LocalTime`: a skipped time fires later by the gap, a
+repeated time fires once, at its first occurrence.
+
 ## Notifications
 
 - Events: `plan_ready`, `list_ready`, `day_of_reminder`, `new_suggestions`,
diff --git a/internal/db/querier.go b/internal/db/querier.go
index c626259..6cd1842 100644
--- a/internal/db/querier.go
+++ b/internal/db/querier.go
@@ -24,6 +24,9 @@ type Querier interface {
 	// Marks the oldest queued job running and returns it. Returns sql.ErrNoRows
 	// when no job is queued.
 	ClaimNextJob(ctx context.Context, now sql.NullString) (Job, error)
+	// Claims one period of a schedule. The insert is the claim: 1 row means
+	// this caller owns the period, 0 rows means it was already claimed.
+	ClaimScheduleRun(ctx context.Context, arg ClaimScheduleRunParams) (int64, error)
 	// Clears every day's calendar event id: none is on the chosen calendar.
 	ClearCalendarEvents(ctx context.Context, updatedAt string) error
 	// Records that Google refuses the stored grant: clears the refresh token and
diff --git a/internal/db/schedule_runs.sql.go b/internal/db/schedule_runs.sql.go
new file mode 100644
index 0000000..0370485
--- /dev/null
+++ b/internal/db/schedule_runs.sql.go
@@ -0,0 +1,38 @@
+// Code generated by sqlc. DO NOT EDIT.
+// versions:
+//   sqlc v1.31.1
+// source: schedule_runs.sql
+
+package db
+
+import (
+	"context"
+)
+
+const claimScheduleRun = `-- name: ClaimScheduleRun :execrows
+INSERT INTO schedule_runs (id, schedule, period_key, created_at)
+VALUES (?, ?, ?, ?)
+ON CONFLICT (schedule, period_key) DO NOTHING
+`
+
+type ClaimScheduleRunParams struct {
+	ID        string
+	Schedule  string
+	PeriodKey string
+	CreatedAt string
+}
+
+// Claims one period of a schedule. The insert is the claim: 1 row means
+// this caller owns the period, 0 rows means it was already claimed.
+func (q *Queries) ClaimScheduleRun(ctx context.Context, arg ClaimScheduleRunParams) (int64, error) {
+	result, err := q.db.ExecContext(ctx, claimScheduleRun,
+		arg.ID,
+		arg.Schedule,
+		arg.PeriodKey,
+		arg.CreatedAt,
+	)
+	if err != nil {
+		return 0, err
+	}
+	return result.RowsAffected()
+}
diff --git a/internal/scheduler/export_test.go b/internal/scheduler/export_test.go
new file mode 100644
index 0000000..38c8057
--- /dev/null
+++ b/internal/scheduler/export_test.go
@@ -0,0 +1,9 @@
+package scheduler
+
+import "time"
+
+// SetClock makes now the clock and after the timer s fires by.
+func SetClock(s *Scheduler, now func() time.Time, after func(time.Duration) <-chan time.Time) {
+	s.now = now
+	s.after = after
+}
diff --git a/internal/scheduler/scheduler.go b/internal/scheduler/scheduler.go
new file mode 100644
index 0000000..77c5ca5
--- /dev/null
+++ b/internal/scheduler/scheduler.go
@@ -0,0 +1,502 @@
+// Package scheduler runs the household's timed work: each registered
+// schedule fires once per period at a household-local time read from the
+// household settings.
+//
+// # Periods and the claim
+//
+// Every fire time belongs to one period, named by a period key derived in
+// the household zone from the schedule's cadence (see Rule.PeriodKey): the
+// local date "2026-10-09" for a daily rule, and the ISO week "2026-W41" of
+// the fire date, or of the week after it, for a weekly rule. Before running,
+// the scheduler inserts schedule_runs(schedule, period_key); the insert is
+// the claim, and a period already claimed, by an earlier run, an earlier
+// process or another instance, is skipped. A period therefore runs at most
+// once: a Run that fails is logged and its period is not retried.
+//
+// # Catch-up
+//
+// When the scheduler starts, and whenever it wakes late, it looks only at
+// each schedule's most recent fire time. If that time passed within
+// CatchUpWindow and the period is unclaimed, the period runs once. If it
+// passed longer ago, the period is claimed without running and logged as
+// missed, so it can never run later, not after a restart and not after a
+// settings change moves the fire time. Fire times before the most recent
+// one are never looked at.
+//
+// # Time zone and DST
+//
+// Fire times are wall-clock times in the household zone, never the process
+// zone, turned into instants by household.LocalTime. Its rule is the
+// scheduler's DST rule: a time that does not exist on a spring-forward day
+// fires later by the length of the gap (02:30 fires at 03:30 local), and a
+// time that occurs twice on a fall-back day fires once, at its first
+// occurrence. Either way the period key is that local date's, so the
+// period still runs exactly once.
+//
+// # Settings changes
+//
+// The scheduler subscribes to the household settings. A change takes effect
+// at once without a restart: fire times up to the change follow the old
+// settings and later ones the new, so moving a fire time to earlier the
+// same day does not fire the passed time.
+package scheduler
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"log/slog"
+	"sync"
+	"time"
+
+	"code.cubecraftlabs.com/Maniacs_Mansion/Meal-Quest/internal/db"
+	"code.cubecraftlabs.com/Maniacs_Mansion/Meal-Quest/internal/household"
+	"code.cubecraftlabs.com/Maniacs_Mansion/Meal-Quest/internal/store"
+)
+
+// Schedule names, as the schedule_runs.schedule CHECK constraint allows.
+const (
+	SuggestionsWeekly    = "suggestions_weekly"
+	PlanDraft            = "plan_draft"
+	ShoppingChatAutosend = "shopping_chat_autosend"
+	DayOfReminder        = "day_of_reminder"
+	RotationCheck        = "rotation_check"
+	WeekRollover         = "week_rollover"
+	AIRunsPurge          = "ai_runs_purge"
+)
+
+// names are the schedule names Register accepts.
+var names = map[string]bool{
+	SuggestionsWeekly:    true,
+	PlanDraft:            true,
+	ShoppingChatAutosend: true,
+	DayOfReminder:        true,
+	RotationCheck:        true,
+	WeekRollover:         true,
+	AIRunsPurge:          true,
+}
+
+// CatchUpWindow is how long after its fire time a period that has not run
+// still runs. A period found later than that is recorded as missed.
+const CatchUpWindow = 24 * time.Hour
+
+// maxWait bounds one sleep, so the scheduler checks the wall clock at least
+// this often even if the host was suspended or its clock stepped.
+const maxWait = time.Hour
+
+// Rule is when a schedule fires: at household-local Time ("HH:MM") every
+// day, or on Weekday only when Weekly is set. Build one with Daily, Weekly
+// or WeeklyForNextWeek.
+type Rule struct {
+	Weekly  bool
+	Weekday time.Weekday
+	Time    string
+	// NextWeek keys a weekly period by the ISO week after the fire date,
+	// for work done ahead for the coming week.
+	NextWeek bool
+}
+
+// Daily fires every day at hhmm; its period is the local date.
+func Daily(hhmm string) Rule {
+	return Rule{Time: hhmm}
+}
+
+// Weekly fires on day at hhmm; its period is the ISO week of the fire date.
+func Weekly(day time.Weekday, hhmm string) Rule {
+	return Rule{Weekly: true, Weekday: day, Time: hhmm}
+}
+
+// WeeklyForNextWeek fires on day at hhmm; its period is the ISO week after
+// the fire date's.
+func WeeklyForNextWeek(day time.Weekday, hhmm string) Rule {
+	return Rule{Weekly: true, Weekday: day, Time: hhmm, NextWeek: true}
+}
+
+// PeriodKey returns the period key of a fire on household-local date
+// fireDate: the date ("2026-10-09") for a daily rule, the ISO week
+// ("2026-W53") for a weekly one, and the ISO week after that ("2027-W01")
+// when NextWeek is set. It panics if fireDate is the zero LocalDate.
+func (r Rule) PeriodKey(fireDate store.LocalDate) string {
+	if !r.Weekly {
+		return fireDate.String()
+	}
+	if r.NextWeek {
+		return household.PeriodKey(addDays(fireDate, 7))
+	}
+	return household.PeriodKey(fireDate)
+}
+
+// validate reports whether r can fire.
+func (r Rule) validate() error {
+	if r.Weekly && (r.Weekday < time.Sunday || r.Weekday > time.Saturday) {
+		return fmt.Errorf("invalid weekday %d (want 0-6, Sunday = 0)", r.Weekday)
+	}
+	if !r.Weekly && (r.Weekday != time.Sunday || r.NextWeek) {
+		return errors.New("a daily rule has no weekday or next week")
+	}
+	// Any date checks the time's form.
+	_, err := household.LocalTime(time.UTC, store.LocalDateOf(time.Unix(0, 0).UTC()), r.Time)
+	return err
+}
+
+// Period is one period of a schedule, as given to its Run.
+type Period struct {
+	// Key is the period key claimed in schedule_runs.
+	Key string
+	// FireAt is the instant the period fired, in the household zone.
+	FireAt time.Time
+	// Date is the household-local date of FireAt.
+	Date store.LocalDate
+	// Week is the ISO Monday of the week Key names; for a daily rule, of
+	// the week containing Date.
+	Week store.LocalDate
+}
+
+// Schedule is registered work that runs once per period.
+type Schedule struct {
+	// Name is one of the schedule name constants.
+	Name string
+	// Rule returns when the schedule fires under settings. It is called
+	// again after every settings change.
+	Rule func(settings household.Settings) Rule
+	// Run does the period's work. It runs on the scheduler's goroutine,
+	// one schedule at a time, so it must return promptly: long work is
+	// enqueued as a job. ctx is cancelled when the scheduler stops. An
+	// error is logged; the period stays claimed and is not retried.
+	Run func(ctx context.Context, p Period) error
+}
+
+// Settings is the household settings source: *household.SettingsService.
+type Settings interface {
+	Get(ctx context.Context) (household.Settings, error)
+	Subscribe(fn func(household.Settings)) (unsubscribe func())
+}
+
+// Scheduler runs the registered schedules.
+type Scheduler struct {
+	st       *store.Store
+	settings Settings
+	logger   *slog.Logger
+
+	// now is the clock and after the timer; tests replace both.
+	now   func() time.Time
+	after func(d time.Duration) <-chan time.Time
+
+	// wake holds one token when the settings changed since the loop last
+	// looked; latest is the settings to apply.
+	wake     chan struct{}
+	latestMu sync.Mutex
+	latest   household.Settings
+
+	mu        sync.Mutex
+	schedules []Schedule
+	started   bool
+
+	wg sync.WaitGroup
+}
+
+// New returns a scheduler over st reading the household settings from
+// settings.
+func New(st *store.Store, settings Settings, logger *slog.Logger) (*Scheduler, error) {
+	if st == nil {
+		return nil, errors.New("scheduler: nil store")
+	}
+	if settings == nil {
+		return nil, errors.New("scheduler: nil settings")
+	}
+	if logger == nil {
+		return nil, errors.New("scheduler: nil logger")
+	}
+	return &Scheduler{
+		st:       st,
+		settings: settings,
+		logger:   logger,
+		now:      time.Now,
+		after:    time.After,
+		wake:     make(chan struct{}, 1),
+	}, nil
+}
+
+// Register adds sch. It must be called before Start, once per name.
+func (s *Scheduler) Register(sch Schedule) error {
+	if !names[sch.Name] {
+		return fmt.Errorf("scheduler: register: unknown schedule %q", sch.Name)
+	}
+	if sch.Rule == nil {
+		return fmt.Errorf("scheduler: register %s: nil rule", sch.Name)
+	}
+	if sch.Run == nil {
+		return fmt.Errorf("scheduler: register %s: nil run", sch.Name)
+	}
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	if s.started {
+		return fmt.Errorf("scheduler: register %s: already started", sch.Name)
+	}
+	for _, have := range s.schedules {
+		if have.Name == sch.Name {
+			return fmt.Errorf("scheduler: register %s: already registered", sch.Name)
+		}
+	}
+	s.schedules = append(s.schedules, sch)
+	return nil
+}
+
+// Start reads the settings and starts the scheduler, which first catches
+// up on each schedule's most recent period and then fires schedules as
+// their times come. Settings whose zone or rules are invalid are an error
+// and nothing starts. The scheduler stops when ctx is cancelled; Wait waits
+// for it.
+func (s *Scheduler) Start(ctx context.Context) error {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	if s.started {
+		return errors.New("scheduler: already started")
+	}
+	// Subscribing first means a change saved after the read below still
+	// wakes the loop.
+	unsubscribe := s.settings.Subscribe(s.changed)
+	settings, err := s.settings.Get(ctx)
+	if err != nil {
+		unsubscribe()
+		return fmt.Errorf("scheduler: %w", err)
+	}
+	p, err := compile(s.schedules, settings)
+	if err != nil {
+		unsubscribe()
+		return err
+	}
+	s.started = true
+	s.wg.Add(1)
+	go func() {
+		defer s.wg.Done()
+		defer unsubscribe()
+		s.loop(ctx, p)
+	}()
+	return nil
+}
+
+// Wait blocks until the scheduler has stopped.
+func (s *Scheduler) Wait() {
+	s.wg.Wait()
+}
+
+// changed is the settings subscriber. It runs on the updating goroutine, so
+// it only records the settings and wakes the loop.
+func (s *Scheduler) changed(settings household.Settings) {
+	s.latestMu.Lock()
+	s.latest = settings
+	s.latestMu.Unlock()
+	select {
+	case s.wake <- struct{}{}:
+	default: // the loop already has a token and reads latest then
+	}
+}
+
+// plan is the schedules compiled against one version of the settings.
+type plan struct {
+	loc   *time.Location
+	items []item
+}
+
+type item struct {
+	sch  Schedule
+	rule Rule
+}
+
+// compile builds the plan for settings.
+func compile(schedules []Schedule, settings household.Settings) (*plan, error) {
+	loc, err := household.LoadZone(settings.Timezone)
+	if err != nil {
+		return nil, fmt.Errorf("scheduler: household zone: %w", err)
+	}
+	p := &plan{loc: loc, items: make([]item, len(schedules))}
+	for i, sch := range schedules {
+		rule := sch.Rule(settings)
+		if err := rule.validate(); err != nil {
+			return nil, fmt.Errorf("scheduler: %s: %w", sch.Name, err)
+		}
+		p.items[i] = item{sch: sch, rule: rule}
+	}
+	return p, nil
+}
+
+// loop fires schedules until ctx is cancelled. done is the instant up to
+// which fire times have been handled; it starts at the zero time so the
+// first pass catches up.
+func (s *Scheduler) loop(ctx context.Context, p *plan) {
+	var done time.Time
+	for {
+		now := s.now()
+		if p != nil {
+			s.fire(ctx, p, done, now)
+		}
+		done = now
+		if ctx.Err() != nil {
+			return
+		}
+
+		var timeout <-chan time.Time
+		if p != nil {
+			if next, ok := p.next(now); ok {
+				timeout = s.after(min(next.Sub(now), maxWait))
+			}
+		}
+		select {
+		case <-ctx.Done():
+			return
+		case <-timeout:
+		case <-s.wake:
+			// Fire times up to the change follow the old settings.
+			now := s.now()
+			if p != nil {
+				s.fire(ctx, p, done, now)
+			}
+			done = now
+			s.latestMu.Lock()
+			settings := s.latest
+			s.latestMu.Unlock()
+			np, err := compile(s.schedules, settings)
+			if err != nil {
+				// Settings are validated before they are saved, so this
+				// is a bug; firing at a guessed time would be worse than
+				// not firing until the settings are fixed.
+				s.logger.ErrorContext(ctx, "scheduler: settings unusable; no schedule fires until they change", "error", err)
+			}
+			p = np
+		}
+	}
+}
+
+// fire handles, for every schedule, its most recent fire time at or before
+// now if it is after done.
+func (s *Scheduler) fire(ctx context.Context, p *plan, done, now time.Time) {
+	for _, it := range p.items {
+		if ctx.Err() != nil {
+			return
+		}
+		at, ok := fireBefore(it.rule, p.loc, now)
+		if !ok || !at.After(done) {
+			continue
+		}
+		s.handle(ctx, it, p.loc, at, now)
+	}
+}
+
+// handle claims the period of it fired at at and runs it, or records it as
+// missed when at is more than CatchUpWindow before now.
+func (s *Scheduler) handle(ctx context.Context, it item, loc *time.Location, at, now time.Time) {
+	date := household.Today(at, loc)
+	period := Period{Key: it.rule.PeriodKey(date), FireAt: at, Date: date}
+	if it.rule.NextWeek {
+		period.Week = household.WeekStart(addDays(date, 7))
+	} else {
+		period.Week = household.WeekStart(date)
+	}
+	log := s.logger.With("schedule", it.sch.Name, "period", period.Key, "fire_at", at.Format(time.RFC3339))
+
+	claimed, err := s.claim(ctx, it.sch.Name, period.Key, now)
+	if err != nil {
+		log.ErrorContext(ctx, "scheduler: claim period failed; period not run", "error", err)
+		return
+	}
+	if !claimed {
+		return
+	}
+	if late := now.Sub(at); late > CatchUpWindow {
+		log.WarnContext(ctx, "scheduler: period missed; recorded without running", "late", late.String())
+		return
+	}
+	if err := it.sch.Run(ctx, period); err != nil {
+		log.ErrorContext(ctx, "scheduler: run failed", "error", err)
+		return
+	}
+	log.InfoContext(ctx, "scheduler: ran")
+}
+
+// claim inserts schedule_runs(name, key) and reports whether this call
+// inserted it.
+func (s *Scheduler) claim(ctx context.Context, name, key string, now time.Time) (bool, error) {
+	n, err := db.New(s.st.Writer).ClaimScheduleRun(ctx, db.ClaimScheduleRunParams{
+		ID:        store.NewID(),
+		Schedule:  name,
+		PeriodKey: key,
+		CreatedAt: store.Timestamp{Time: now.UTC().Truncate(time.Millisecond)}.String(),
+	})
+	if err != nil {
+		return false, err
+	}
+	return n == 1, nil
+}
+
+// next returns the earliest fire time of any schedule after now, or false
+// when there is no schedule.
+func (p *plan) next(now time.Time) (time.Time, bool) {
+	var next time.Time
+	for _, it := range p.items {
+		at, ok := fireAfter(it.rule, p.loc, now)
+		if ok && (next.IsZero() || at.Before(next)) {
+			next = at
+		}
+	}
+	return next, !next.IsZero()
+}
+
+// window is how many days either side of a date fireBefore and fireAfter
+// search: more than a week, so a weekly rule always has a fire time on
+// each side.
+const window = 8
+
+// fireBefore returns the latest fire time of r in zone loc at or before t.
+func fireBefore(r Rule, loc *time.Location, t time.Time) (time.Time, bool) {
+	var best time.Time
+	for _, at := range fireTimes(r, loc, t) {
+		if !at.After(t) && (best.IsZero() || at.After(best)) {
+			best = at
+		}
+	}
+	return best, !best.IsZero()
+}
+
+// fireAfter returns the earliest fire time of r in zone loc after t.
+func fireAfter(r Rule, loc *time.Location, t time.Time) (time.Time, bool) {
+	var best time.Time
+	for _, at := range fireTimes(r, loc, t) {
+		if at.After(t) && (best.IsZero() || at.Before(best)) {
+			best = at
+		}
+	}
+	return best, !best.IsZero()
+}
+
+// fireTimes returns r's fire times on the local dates within window days
+// of t's. A rule that fails to resolve was validated at compile, so it
+// cannot.
+func fireTimes(r Rule, loc *time.Location, t time.Time) []time.Time {
+	today := household.Today(t, loc)
+	var out []time.Time
+	for i := -window; i <= window; i++ {
+		d := addDays(today, i)
+		if r.Weekly && weekday(d) != r.Weekday {
+			continue
+		}
+		at, err := household.LocalTime(loc, d, r.Time)
+		if err != nil {
+			panic(fmt.Sprintf("scheduler: validated rule %+v: %v", r, err))
+		}
+		out = append(out, at)
+	}
+	return out
+}
+
+// addDays returns d plus n calendar days.
+func addDays(d store.LocalDate, n int) store.LocalDate {
+	y, m, day := d.Date()
+	return store.LocalDateOf(time.Date(y, m, day+n, 0, 0, 0, 0, time.UTC))
+}
+
+// weekday returns the day of the week of d.
+func weekday(d store.LocalDate) time.Weekday {
+	y, m, day := d.Date()
+	return time.Date(y, m, day, 0, 0, 0, 0, time.UTC).Weekday()
+}
diff --git a/internal/scheduler/scheduler_test.go b/internal/scheduler/scheduler_test.go
new file mode 100644
index 0000000..2bb6183
--- /dev/null
+++ b/internal/scheduler/scheduler_test.go
@@ -0,0 +1,653 @@
+package scheduler_test
+
+import (
+	"context"
+	"errors"
+	"io"
+	"log/slog"
+	"slices"
+	"sync"
+	"testing"
+	"time"
+
+	"code.cubecraftlabs.com/Maniacs_Mansion/Meal-Quest/internal/household"
+	"code.cubecraftlabs.com/Maniacs_Mansion/Meal-Quest/internal/scheduler"
+	"code.cubecraftlabs.com/Maniacs_Mansion/Meal-Quest/internal/store"
+	"code.cubecraftlabs.com/Maniacs_Mansion/Meal-Quest/migrations"
+)
+
+const zone = "America/Chicago"
+
+// chicago is the household zone of every test.
+var chicago = mustZone(zone)
+
+func mustZone(name string) *time.Location {
+	loc, err := household.LoadZone(name)
+	if err != nil {
+		panic(err)
+	}
+	return loc
+}
+
+// at returns the instant clocks in Chicago read the given time.
+func at(year int, month time.Month, day, hour, minute int) time.Time {
+	return time.Date(year, month, day, hour, minute, 0, 0, chicago)
+}
+
+// fakeClock is a clock whose time moves only when the test moves it.
+type fakeClock struct {
+	mu     sync.Mutex
+	now    time.Time
+	timers []fakeTimer
+	// waits receives one value per After call, so the test knows the
+	// scheduler finished a pass and is waiting.
+	waits chan struct{}
+}
+
+type fakeTimer struct {
+	deadline time.Time
+	ch       chan time.Time
+}
+
+func newFakeClock(now time.Time) *fakeClock {
+	return &fakeClock{now: now, waits: make(chan struct{}, 100)}
+}
+
+func (c *fakeClock) Now() time.Time {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	return c.now
+}
+
+func (c *fakeClock) After(d time.Duration) <-chan time.Time {
+	c.mu.Lock()
+	ch := make(chan time.Time, 1)
+	c.timers = append(c.timers, fakeTimer{deadline: c.now.Add(d), ch: ch})
+	c.mu.Unlock()
+	c.waits <- struct{}{}
+	return ch
+}
+
+// set moves the clock to t and fires the timers due by then, reporting
+// whether any fired.
+func (c *fakeClock) set(t time.Time) bool {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	c.now = t
+	fired := false
+	kept := c.timers[:0]
+	for _, tm := range c.timers {
+		if tm.deadline.After(t) {
+			kept = append(kept, tm)
+			continue
+		}
+		tm.ch <- t
+		fired = true
+	}
+	c.timers = kept
+	return fired
+}
+
+// nextDeadline returns the earliest pending timer deadline.
+func (c *fakeClock) nextDeadline() (time.Time, bool) {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	var next time.Time
+	for _, tm := range c.timers {
+		if next.IsZero() || tm.deadline.Before(next) {
+			next = tm.deadline
+		}
+	}
+	return next, !next.IsZero()
+}
+
+// recorder records the periods a schedule ran.
+type recorder struct {
+	mu   sync.Mutex
+	runs []scheduler.Period
+	err  error
+}
+
+func (r *recorder) run(_ context.Context, p scheduler.Period) error {
+	r.mu.Lock()
+	defer r.mu.Unlock()
+	r.runs = append(r.runs, p)
+	return r.err
+}
+
+func (r *recorder) periods() []scheduler.Period {
+	r.mu.Lock()
+	defer r.mu.Unlock()
+	return slices.Clone(r.runs)
+}
+
+func (r *recorder) keys() []string {
+	var keys []string
+	for _, p := range r.periods() {
+		keys = append(keys, p.Key)
+	}
+	return keys
+}
+
+// env is a migrated database with household settings in Chicago.
+type env struct {
+	t        *testing.T
+	st       *store.Store
+	settings *household.SettingsService
+}
+
+func newEnv(t *testing.T) *env {
+	t.Helper()
+	ctx := context.Background()
+	st, err := store.Open(ctx, t.TempDir())
+	if err != nil {
+		t.Fatal(err)
+	}
+	t.Cleanup(func() {
+		if err := st.Close(); err != nil {
+			t.Error(err)
+		}
+	})
+	if err := store.Migrate(ctx, st.Writer, migrations.FS); err != nil {
+		t.Fatal(err)
+	}
+	if err := household.SeedSettings(ctx, st, zone); err != nil {
+		t.Fatal(err)
+	}
+	return &env{t: t, st: st, settings: household.NewSettingsService(st)}
+}
+
+// update changes the household settings.
+func (e *env) update(patch household.SettingsPatch) {
+	e.t.Helper()
+	if _, err := e.settings.Update(context.Background(), patch, "member"); err != nil {
+		e.t.Fatal(err)
+	}
+}
+
+// claimed returns the period keys schedule_runs holds for name.
+func (e *env) claimed(name string) []string {
+	e.t.Helper()
+	rows, err := e.st.Reader.Query(`SELECT period_key FROM schedule_runs WHERE schedule = ? ORDER BY period_key`, name)
+	if err != nil {
+		e.t.Fatal(err)
+	}
+	defer rows.Close()
+	var keys []string
+	for rows.Next() {
+		var k string
+		if err := rows.Scan(&k); err != nil {
+			e.t.Fatal(err)
+		}
+		keys = append(keys, k)
+	}
+	if err := rows.Err(); err != nil {
+		e.t.Fatal(err)
+	}
+	return keys
+}
+
+// running is a started scheduler on a fake clock.
+type running struct {
+	t     *testing.T
+	clock *fakeClock
+	stop  func()
+}
+
+// start starts a scheduler on e at now with the schedules, and stops it
+// when the test ends. It returns once the first pass, the catch-up, is done.
+func (e *env) start(now time.Time, schedules ...scheduler.Schedule) *running {
+	e.t.Helper()
+	s, err := scheduler.New(e.st, e.settings, slog.New(slog.NewTextHandler(io.Discard, nil)))
+	if err != nil {
+		e.t.Fatal(err)
+	}
+	clock := newFakeClock(now)
+	scheduler.SetClock(s, clock.Now, clock.After)
+	for _, sch := range schedules {
+		if err := s.Register(sch); err != nil {
+			e.t.Fatal(err)
+		}
+	}
+	ctx, cancel := context.WithCancel(context.Background())
+	if err := s.Start(ctx); err != nil {
+		cancel()
+		e.t.Fatal(err)
+	}
+	var once sync.Once
+	stop := func() {
+		once.Do(func() {
+			cancel()
+			s.Wait()
+		})
+	}
+	e.t.Cleanup(stop)
+	r := &running{t: e.t, clock: clock, stop: stop}
+	r.idle()
+	return r
+}
+
+// idle waits until the scheduler finished a pass and is waiting.
+func (r *running) idle() {
+	r.t.Helper()
+	select {
+	case <-r.clock.waits:
+	case <-time.After(10 * time.Second):
+		r.t.Fatal("scheduler did not finish its pass")
+	}
+}
+
+// runUntil moves the clock to to, stopping at every instant the scheduler
+// asked to wake at on the way, as a real clock would.
+func (r *running) runUntil(to time.Time) {
+	r.t.Helper()
+	for {
+		next, ok := r.clock.nextDeadline()
+		if !ok || next.After(to) {
+			if r.clock.set(to) {
+				r.idle()
+			}
+			return
+		}
+		if r.clock.set(next) {
+			r.idle()
+		}
+	}
+}
+
+// changed waits for the scheduler to apply a settings change.
+func (r *running) changed() {
+	r.t.Helper()
+	r.idle()
+}
+
+func reminder(rec *recorder) scheduler.Schedule {
+	return scheduler.Schedule{
+		Name: scheduler.DayOfReminder,
+		Rule: func(s household.Settings) scheduler.Rule { return scheduler.Daily(s.ReminderTime) },
+		Run:  rec.run,
+	}
+}
+
+func suggestions(rec *recorder) scheduler.Schedule {
+	return scheduler.Schedule{
+		Name: scheduler.SuggestionsWeekly,
+		Rule: func(s household.Settings) scheduler.Rule {
+			return scheduler.Weekly(time.Weekday(s.SuggestionsDow), s.SuggestionsTime)
+		},
+		Run: rec.run,
+	}
+}
+
+func planDraft(rec *recorder) scheduler.Schedule {
+	return scheduler.Schedule{
+		Name: scheduler.PlanDraft,
+		Rule: func(s household.Settings) scheduler.Rule {
+			return scheduler.WeeklyForNextWeek(time.Weekday(s.PlanDraftDow), s.PlanDraftTime)
+		},
+		Run: rec.run,
+	}
+}
+
+func ptr[T any](v T) *T { return &v }
+
+func wantKeys(t *testing.T, what string, got, want []string) {
+	t.Helper()
+	if !slices.Equal(got, want) {
+		t.Fatalf("%s = %q, want %q", what, got, want)
+	}
+}
+
+func TestFiresOncePerPeriod(t *testing.T) {
+	e := newEnv(t)
+	var rec recorder
+	// Seeded reminder time is 16:30; boot just after it on Oct 6 runs it.
+	r := e.start(at(2026, 10, 6, 16, 31), reminder(&rec))
+	wantKeys(t, "after boot", rec.keys(), []string{"2026-10-06"})
+
+	r.runUntil(at(2026, 10, 7, 16, 29))
+	wantKeys(t, "before the fire time", rec.keys(), []string{"2026-10-06"})
+	r.runUntil(at(2026, 10, 9, 23, 59))
+	wantKeys(t, "after three days", rec.keys(), []string{"2026-10-06", "2026-10-07", "2026-10-08", "2026-10-09"})
+
+	for i, p := range rec.periods()[1:] {
+		want := at(2026, 10, 7+i, 16, 30)
+		if !p.FireAt.Equal(want) {
+			t.Errorf("run %s fired at %s, want %s", p.Key, p.FireAt, want)
+		}
+		if p.FireAt.Location().String() != zone {
+			t.Errorf("run %s fired in zone %s, want %s", p.Key, p.FireAt.Location(), zone)
+		}
+	}
+	wantKeys(t, "schedule_runs", e.claimed(scheduler.DayOfReminder), rec.keys())
+}
+
+func TestRestartMidPeriodDoesNotRerun(t *testing.T) {
+	e := newEnv(t)
+	var first recorder
+	r := e.start(at(2026, 10, 7, 12, 0), suggestions(&first)) // Wednesday
+	// The seeded Sunday 08:00 of Oct 4 is days ago: missed, not run.
+	wantKeys(t, "first boot", first.keys(), nil)
+	r.runUntil(at(2026, 10, 11, 9, 0))
+	wantKeys(t, "first process", first.keys(), []string{"2026-W41"})
+	r.stop()
+
+	// The restart is an hour after the Sunday 08:00 fire, mid-period.
+	var second recorder
+	r2 := e.start(at(2026, 10, 11, 9, 0), suggestions(&second))
+	wantKeys(t, "after restart", second.keys(), nil)
+	r2.runUntil(at(2026, 10, 18, 7, 59))
+	wantKeys(t, "rest of the period", second.keys(), nil)
+	r2.runUntil(at(2026, 10, 18, 8, 0))
+	wantKeys(t, "next period", second.keys(), []string{"2026-W42"})
+	wantKeys(t, "schedule_runs", e.claimed(scheduler.SuggestionsWeekly), []string{"2026-W40", "2026-W41", "2026-W42"})
+}
+
+func TestCatchUpWithin24Hours(t *testing.T) {
+	e := newEnv(t)
+	var rec recorder
+	// Sunday Oct 11 08:00 passed 23h59m before boot.
+	e.start(at(2026, 10, 12, 7, 59), suggestions(&rec))
+	wantKeys(t, "catch-up", rec.keys(), []string{"2026-W41"})
+	if p := rec.periods()[0]; !p.FireAt.Equal(at(2026, 10, 11, 8, 0)) {
+		t.Errorf("caught-up period fired at %s, want its fire time", p.FireAt)
+	}
+}
+
+func TestNoCatchUpBeyond24Hours(t *testing.T) {
+	e := newEnv(t)
+	var rec recorder
+	// Sunday Oct 11 08:00 passed 24h01m before boot.
+	r := e.start(at(2026, 10, 12, 8, 1), suggestions(&rec))
+	wantKeys(t, "boot", rec.keys(), nil)
+	// The missed period is recorded, so it never runs later.
+	wantKeys(t, "schedule_runs", e.claimed(scheduler.SuggestionsWeekly), []string{"2026-W41"})
+
+	// Moving the fire time to Sunday 23:00 puts that week's fire time
+	// within 24 h of now. The change itself never fires a passed time...
+	e.update(household.SettingsPatch{SuggestionsTime: ptr("23:00")})
+	r.changed()
+	wantKeys(t, "after a settings change", rec.keys(), nil)
+	r.stop()
+
+	// ...and the restart's catch-up finds the period already recorded.
+	var again recorder
+	r2 := e.start(at(2026, 10, 12, 8, 2), suggestions(&again))
+	wantKeys(t, "after restart", again.keys(), nil)
+	r2.runUntil(at(2026, 10, 18, 23, 0))
+	wantKeys(t, "next period", again.keys(), []string{"2026-W42"})
+}
+
+func TestCatchUpOnlyMostRecentPeriod(t *testing.T) {
+	e := newEnv(t)
+	var rec recorder
+	// Days of downtime: only Oct 9 16:30, 30 minutes ago, runs.
+	e.start(at(2026, 10, 9, 17, 0), reminder(&rec))
+	wantKeys(t, "catch-up", rec.keys(), []string{"2026-10-09"})
+	wantKeys(t, "schedule_runs", e.claimed(scheduler.DayOfReminder), []string{"2026-10-09"})
+}
+
+func TestSpringForward(t *testing.T) {
+	// Clocks in Chicago jump 02:00 CST → 03:00 CDT on Sunday 2026-03-08.
+	e := newEnv(t)
+	e.update(household.SettingsPatch{ReminderTime: ptr("08:00"), ReviewTime: ptr("02:30")})
+	var daily, skipped recorder
+	review := scheduler.Schedule{
+		Name: scheduler.RotationCheck,
+		Rule: func(s household.Settings) scheduler.Rule { return scheduler.Daily(s.ReviewTime) },
+		Run:  skipped.run,
+	}
+	r := e.start(at(2026, 3, 7, 9, 0), reminder(&daily), review)
+	r.runUntil(at(2026, 3, 9, 12, 0))
+
+	wantKeys(t, "08:00 runs", daily.keys(), []string{"2026-03-07", "2026-03-08", "2026-03-09"})
+	wantFired(t, daily.periods(),
+		time.Date(2026, 3, 7, 14, 0, 0, 0, time.UTC), // 08:00 CST
+		time.Date(2026, 3, 8, 13, 0, 0, 0, time.UTC), // 08:00 CDT
+		time.Date(2026, 3, 9, 13, 0, 0, 0, time.UTC),
+	)
+
+	// 02:30 does not exist on Mar 8: it fires once, an hour later at
+	// 03:30 CDT, keyed by Mar 8.
+	wantKeys(t, "02:30 runs", skipped.keys(), []string{"2026-03-07", "2026-03-08", "2026-03-09"})
+	wantFired(t, skipped.periods(),
+		time.Date(2026, 3, 7, 8, 30, 0, 0, time.UTC), // 02:30 CST
+		time.Date(2026, 3, 8, 8, 30, 0, 0, time.UTC), // 03:30 CDT
+		time.Date(2026, 3, 9, 7, 30, 0, 0, time.UTC), // 02:30 CDT
+	)
+	if got := skipped.periods()[1].FireAt.Format("15:04 MST"); got != "03:30 CDT" {
+		t.Errorf("skipped 02:30 fired at %s local, want 03:30 CDT", got)
+	}
+}
+
+func TestFallBack(t *testing.T) {
+	// Clocks in Chicago go back 02:00 CDT → 01:00 CST on Sunday 2026-11-01.
+	e := newEnv(t)
+	e.update(household.SettingsPatch{ReminderTime: ptr("08:00"), ReviewTime: ptr("01:30")})
+	var daily, repeated recorder
+	review := scheduler.Schedule{
+		Name: scheduler.RotationCheck,
+		Rule: func(s household.Settings) scheduler.Rule { return scheduler.Daily(s.ReviewTime) },
+		Run:  repeated.run,
+	}
+	r := e.start(at(2026, 10, 31, 9, 0), reminder(&daily), review)
+	r.runUntil(at(2026, 11, 2, 12, 0))
+
+	wantKeys(t, "08:00 runs", daily.keys(), []string{"2026-10-31", "2026-11-01", "2026-11-02"})
+	wantFired(t, daily.periods(),
+		time.Date(2026, 10, 31, 13, 0, 0, 0, time.UTC), // 08:00 CDT
+		time.Date(2026, 11, 1, 14, 0, 0, 0, time.UTC),  // 08:00 CST
+		time.Date(2026, 11, 2, 14, 0, 0, 0, time.UTC),
+	)
+
+	// 01:30 happens twice on Nov 1: it fires once, at the first (CDT).
+	wantKeys(t, "01:30 runs", repeated.keys(), []string{"2026-10-31", "2026-11-01", "2026-11-02"})
+	wantFired(t, repeated.periods(),
+		time.Date(2026, 10, 31, 6, 30, 0, 0, time.UTC), // 01:30 CDT
+		time.Date(2026, 11, 1, 6, 30, 0, 0, time.UTC),  // first 01:30, CDT
+		time.Date(2026, 11, 2, 7, 30, 0, 0, time.UTC),  // 01:30 CST
+	)
+}
+
+func TestFallBackSecondOccurrenceDoesNotRerun(t *testing.T) {
+	e := newEnv(t)
+	e.update(household.SettingsPatch{ReviewTime: ptr("01:30")})
+	var rec recorder
+	review := scheduler.Schedule{
+		Name: scheduler.RotationCheck,
+		Rule: func(s household.Settings) scheduler.Rule { return scheduler.Daily(s.ReviewTime) },
+		Run:  rec.run,
+	}
+	// Boot between the two 01:30s of Nov 1: the first ran in an earlier
+	// process, so the second is not a new period.
+	first := time.Date(2026, 11, 1, 6, 30, 0, 0, time.UTC)
+	r := e.start(first.Add(-time.Minute), review)
+	r.runUntil(first)
+	wantKeys(t, "first 01:30", rec.keys(), []string{"2026-10-31", "2026-11-01"})
+	r.stop()
+
+	var again recorder
+	review.Run = again.run
+	r2 := e.start(first.Add(30*time.Minute), review)
+	r2.runUntil(first.Add(2 * time.Hour)) // past the second 01:30, CST
+	wantKeys(t, "second 01:30", again.keys(), nil)
+}
+
+func wantFired(t *testing.T, got []scheduler.Period, want ...time.Time) {
+	t.Helper()
+	if len(got) != len(want) {
+		t.Fatalf("%d runs, want %d", len(got), len(want))
+	}
+	for i := range want {
+		if !got[i].FireAt.Equal(want[i]) {
+			t.Errorf("run %s fired at %s (%s), want %s", got[i].Key, got[i].FireAt.UTC(), got[i].FireAt, want[i])
+		}
+	}
+}
+
+func TestISOWeek53Boundary(t *testing.T) {
+	e := newEnv(t)
+	var sugg, draft recorder
+	// Suggestions Sunday 08:00 keys the ISO week ending that Sunday; the
+	// plan draft Friday 09:00 keys the week after.
+	r := e.start(at(2026, 12, 20, 9, 0), suggestions(&sugg), planDraft(&draft))
+	r.runUntil(at(2027, 1, 11, 0, 0))
+
+	wantKeys(t, "suggestions", sugg.keys(), []string{"2026-W51", "2026-W52", "2026-W53", "2027-W01"})
+	// Sunday 2027-01-03 is the last day of 2026-W53.
+	if p := sugg.periods()[2]; p.Date.String() != "2027-01-03" || p.Week.String() != "2026-12-28" {
+		t.Errorf("2026-W53 suggestions ran on %s for week %s, want 2027-01-03 for 2026-12-28", p.Date, p.Week)
+	}
+	// Friday Dec 18, two days before boot, was missed.
+	wantKeys(t, "plan drafts", draft.keys(), []string{"2026-W53", "2027-W01", "2027-W02"})
+	// Friday 2026-12-25 drafts 2026-W53; Friday 2027-01-01 drafts 2027-W01.
+	if p := draft.periods()[0]; p.Date.String() != "2026-12-25" || p.Week.String() != "2026-12-28" {
+		t.Errorf("2026-W53 draft ran on %s for week %s, want 2026-12-25 for 2026-12-28", p.Date, p.Week)
+	}
+	if p := draft.periods()[1]; p.Date.String() != "2027-01-01" || p.Week.String() != "2027-01-04" {
+		t.Errorf("2027-W01 draft ran on %s for week %s, want 2027-01-01 for 2027-01-04", p.Date, p.Week)
+	}
+}
+
+func TestPeriodKey(t *testing.T) {
+	date := func(s string) store.LocalDate {
+		d, err := store.ParseLocalDate(s)
+		if err != nil {
+			t.Fatal(err)
+		}
+		return d
+	}
+	tests := []struct {
+		rule scheduler.Rule
+		date string
+		want string
+	}{
+		{scheduler.Daily("16:30"), "2026-12-31", "2026-12-31"},
+		{scheduler.Weekly(time.Monday, "00:00"), "2026-12-28", "2026-W53"},
+		{scheduler.Weekly(time.Sunday, "08:00"), "2027-01-03", "2026-W53"},
+		{scheduler.Weekly(time.Monday, "00:00"), "2027-01-04", "2027-W01"},
+		{scheduler.WeeklyForNextWeek(time.Friday, "09:00"), "2026-12-25", "2026-W53"},
+		{scheduler.WeeklyForNextWeek(time.Friday, "09:00"), "2027-01-01", "2027-W01"},
+	}
+	for _, tt := range tests {
+		if got := tt.rule.PeriodKey(date(tt.date)); got != tt.want {
+			t.Errorf("%+v.PeriodKey(%s) = %s, want %s", tt.rule, tt.date, got, tt.want)
+		}
+	}
+}
+
+func TestSettingsChangeMovesNextFire(t *testing.T) {
+	e := newEnv(t)
+	var rec recorder
+	r := e.start(at(2026, 10, 7, 12, 0), reminder(&rec))
+	wantKeys(t, "boot", rec.keys(), []string{"2026-10-06"})
+
+	// Earlier but still ahead: fires at the new time, not the old.
+	e.update(household.SettingsPatch{ReminderTime: ptr("13:00")})
+	r.changed()
+	if next, _ := r.clock.nextDeadline(); !next.Equal(at(2026, 10, 7, 13, 0)) {
+		t.Errorf("next wake %s, want the new 13:00", next)
+	}
+	r.runUntil(at(2026, 10, 7, 13, 0))
+	wantKeys(t, "at the new time", rec.keys(), []string{"2026-10-06", "2026-10-07"})
+	r.runUntil(at(2026, 10, 7, 23, 0))
+	wantKeys(t, "at the old time", rec.keys(), []string{"2026-10-06", "2026-10-07"})
+
+	// Moved to a time already passed today: nothing fires now, and
+	// tomorrow fires at the new time.
+	r.runUntil(at(2026, 10, 8, 12, 0))
+	e.update(household.SettingsPatch{ReminderTime: ptr("11:00")})
+	r.changed()
+	wantKeys(t, "moved into the past", rec.keys(), []string{"2026-10-06", "2026-10-07"})
+	r.runUntil(at(2026, 10, 9, 10, 59))
+	wantKeys(t, "before the new time", rec.keys(), []string{"2026-10-06", "2026-10-07"})
+	r.runUntil(at(2026, 10, 9, 11, 0))
+	wantKeys(t, "the new time", rec.keys(), []string{"2026-10-06", "2026-10-07", "2026-10-09"})
+}
+
+func TestSettingsChangeTimezone(t *testing.T) {
+	e := newEnv(t)
+	var rec recorder
+	r := e.start(at(2026, 10, 7, 12, 0), reminder(&rec))
+	e.update(household.SettingsPatch{Timezone: ptr("America/New_York")})
+	r.changed()
+	// 16:30 in New York is 15:30 in Chicago.
+	r.runUntil(at(2026, 10, 7, 15, 30))
+	wantKeys(t, "after the zone change", rec.keys(), []string{"2026-10-06", "2026-10-07"})
+	if got := rec.periods()[1].FireAt.Location().String(); got != "America/New_York" {
+		t.Errorf("fired in zone %s, want America/New_York", got)
+	}
+}
+
+func TestRunErrorKeepsPeriodClaimed(t *testing.T) {
+	e := newEnv(t)
+	rec := recorder{err: errors.New("boom")}
+	r := e.start(at(2026, 10, 7, 16, 0), reminder(&rec))
+	r.runUntil(at(2026, 10, 7, 23, 0))
+	wantKeys(t, "runs", rec.keys(), []string{"2026-10-06", "2026-10-07"})
+	r.stop()
+	var again recorder
+	e.start(at(2026, 10, 7, 23, 0), reminder(&again))
+	wantKeys(t, "after restart", again.keys(), nil)
+}
+
+func TestRegister(t *testing.T) {
+	e := newEnv(t)
+	s, err := scheduler.New(e.st, e.settings, slog.New(slog.NewTextHandler(io.Discard, nil)))
+	if err != nil {
+		t.Fatal(err)
+	}
+	var rec recorder
+	bad := []scheduler.Schedule{
+		{Name: "hourly", Rule: reminder(&rec).Rule, Run: rec.run},
+		{Name: scheduler.DayOfReminder, Run: rec.run},
+		{Name: scheduler.DayOfReminder, Rule: reminder(&rec).Rule},
+	}
+	for _, sch := range bad {
+		if err := s.Register(sch); err == nil {
+			t.Errorf("Register(%q) accepted an invalid schedule", sch.Name)
+		}
+	}
+	if err := s.Register(reminder(&rec)); err != nil {
+		t.Fatal(err)
+	}
+	if err := s.Register(reminder(&rec)); err == nil {
+		t.Error("Register accepted a second schedule with the same name")
+	}
+}
+
+func TestStartRejectsInvalidRule(t *testing.T) {
+	e := newEnv(t)
+	s, err := scheduler.New(e.st, e.settings, slog.New(slog.NewTextHandler(io.Discard, nil)))
+	if err != nil {
+		t.Fatal(err)
+	}
+	var rec recorder
+	if err := s.Register(scheduler.Schedule{
+		Name: scheduler.WeekRollover,
+		Rule: func(household.Settings) scheduler.Rule { return scheduler.Weekly(7, "00:00") },
+		Run:  rec.run,
+	}); err != nil {
+		t.Fatal(err)
+	}
+	if err := s.Start(context.Background()); err == nil {
+		s.Wait()
+		t.Fatal("Start accepted weekday 7")
+	}
+}
+
+func TestStopEndsTheLoop(t *testing.T) {
+	e := newEnv(t)
+	var rec recorder
+	r := e.start(at(2026, 10, 7, 12, 0), reminder(&rec))
+	done := make(chan struct{})
+	go func() {
+		r.stop()
+		close(done)
+	}()
+	select {
+	case <-done:
+	case <-time.After(10 * time.Second):
+		t.Fatal("scheduler did not stop")
+	}
+}
diff --git a/migrations/queries/schedule_runs.sql b/migrations/queries/schedule_runs.sql
new file mode 100644
index 0000000..e0b3a93
--- /dev/null
+++ b/migrations/queries/schedule_runs.sql
@@ -0,0 +1,6 @@
+-- name: ClaimScheduleRun :execrows
+-- Claims one period of a schedule. The insert is the claim: 1 row means
+-- this caller owns the period, 0 rows means it was already claimed.
+INSERT INTO schedule_runs (id, schedule, period_key, created_at)
+VALUES (?, ?, ?, ?)
+ON CONFLICT (schedule, period_key) DO NOTHING;
-- 
2.55.0


From 6f7e1d23420b5aed0a51a91c7d6142ad17758480 Mon Sep 17 00:00:00 2001
From: dev_agent <dev_agent@cubecraftlabs.com>
Date: Fri, 9 Oct 2026 18:41:09 -0400
Subject: [PATCH 2/3] Apply settings changes at their save time and retry
 failed claims

A settings change is now queued with the instant it was saved, and the
loop handles fire times up to that instant under the old rule and later
ones under the new, so a fire time the change made due is not skipped
when the scheduler gets to the change late (SCHED-01).

A failed claim no longer advances the loop past its period: the pass is
repeated every ClaimRetry until the claim succeeds, and periods already
claimed in that pass are skipped by the claim, so each runs once
(SCHED-02).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---
 docs/ARCHITECTURE.md                 |   6 +-
 internal/scheduler/scheduler.go      | 153 ++++++++++++++++++---------
 internal/scheduler/scheduler_test.go |  70 ++++++++++++
 3 files changed, 180 insertions(+), 49 deletions(-)

diff --git a/docs/ARCHITECTURE.md b/docs/ARCHITECTURE.md
index 190022c..18f0724 100644
--- a/docs/ARCHITECTURE.md
+++ b/docs/ARCHITECTURE.md
@@ -793,7 +793,11 @@ key (`Rule.PeriodKey`): the local date for a daily rule, the ISO week of the
 fire date for a weekly one, or of the week after for `WeeklyForNextWeek`
 (`plan_draft`). Inserting `schedule_runs(schedule, period_key)` is the claim;
 a period already claimed is skipped, so a period runs at most once and a
-failed run is not retried. At boot only each schedule's most recent fire
+failed run is not retried. A claim that fails is not skipped: the pass is
+repeated every `scheduler.ClaimRetry` until the claim succeeds. A settings
+change is recorded with the instant it was saved, and fire times up to that
+instant follow the old rule and later ones the new, however late the
+scheduler applies it. At boot only each schedule's most recent fire
 time is considered: it runs if it passed within 24 hours, and is otherwise
 claimed without running (logged as missed) so it never runs later. DST
 follows `household.LocalTime`: a skipped time fires later by the gap, a
diff --git a/internal/scheduler/scheduler.go b/internal/scheduler/scheduler.go
index 77c5ca5..38d3b57 100644
--- a/internal/scheduler/scheduler.go
+++ b/internal/scheduler/scheduler.go
@@ -23,6 +23,13 @@
 // settings change moves the fire time. Fire times before the most recent
 // one are never looked at.
 //
+// # Claim failures
+//
+// A claim that fails (the database refused the insert) is never skipped:
+// the scheduler logs it and does not advance past it, but repeats the
+// whole pass every ClaimRetry until the claim succeeds. Periods the pass
+// already claimed are claimed again and skipped, so each still runs once.
+//
 // # Time zone and DST
 //
 // Fire times are wall-clock times in the household zone, never the process
@@ -35,10 +42,12 @@
 //
 // # Settings changes
 //
-// The scheduler subscribes to the household settings. A change takes effect
-// at once without a restart: fire times up to the change follow the old
-// settings and later ones the new, so moving a fire time to earlier the
-// same day does not fire the passed time.
+// The scheduler subscribes to the household settings and records each change
+// with the instant it was saved. A change takes effect at once without a
+// restart: fire times up to that instant follow the old settings and later
+// ones the new, even when the scheduler gets to the change late. So moving
+// a fire time to earlier the same day does not fire the passed time, and a
+// new fire time that was still ahead when the change was saved fires.
 package scheduler
 
 import (
@@ -80,6 +89,10 @@ var names = map[string]bool{
 // still runs. A period found later than that is recorded as missed.
 const CatchUpWindow = 24 * time.Hour
 
+// ClaimRetry is how long the scheduler waits before repeating a pass in
+// which a claim failed.
+const ClaimRetry = time.Minute
+
 // maxWait bounds one sleep, so the scheduler checks the wall clock at least
 // this often even if the host was suspended or its clock stepped.
 const maxWait = time.Hour
@@ -183,10 +196,10 @@ type Scheduler struct {
 	after func(d time.Duration) <-chan time.Time
 
 	// wake holds one token when the settings changed since the loop last
-	// looked; latest is the settings to apply.
-	wake     chan struct{}
-	latestMu sync.Mutex
-	latest   household.Settings
+	// looked; pending is the changes to apply, oldest first.
+	wake      chan struct{}
+	pendingMu sync.Mutex
+	pending   []change
 
 	mu        sync.Mutex
 	schedules []Schedule
@@ -281,18 +294,33 @@ func (s *Scheduler) Wait() {
 	s.wg.Wait()
 }
 
+// change is a settings change and the instant it was saved.
+type change struct {
+	at       time.Time
+	settings household.Settings
+}
+
 // changed is the settings subscriber. It runs on the updating goroutine, so
-// it only records the settings and wakes the loop.
+// it only records the change and wakes the loop.
 func (s *Scheduler) changed(settings household.Settings) {
-	s.latestMu.Lock()
-	s.latest = settings
-	s.latestMu.Unlock()
+	s.pendingMu.Lock()
+	s.pending = append(s.pending, change{at: s.now(), settings: settings})
+	s.pendingMu.Unlock()
 	select {
 	case s.wake <- struct{}{}:
-	default: // the loop already has a token and reads latest then
+	default: // the loop already has a token and reads pending then
 	}
 }
 
+// takeChanges returns and clears the pending changes.
+func (s *Scheduler) takeChanges() []change {
+	s.pendingMu.Lock()
+	defer s.pendingMu.Unlock()
+	out := s.pending
+	s.pending = nil
+	return out
+}
+
 // plan is the schedules compiled against one version of the settings.
 type plan struct {
 	loc   *time.Location
@@ -323,21 +351,46 @@ func compile(schedules []Schedule, settings household.Settings) (*plan, error) {
 
 // loop fires schedules until ctx is cancelled. done is the instant up to
 // which fire times have been handled; it starts at the zero time so the
-// first pass catches up.
+// first pass catches up. Each pass first applies the pending settings
+// changes in order, handling fire times up to a change under the plan it
+// replaces, then handles fire times up to now under the current plan. A
+// pass with a failed claim keeps done and its unapplied changes where they
+// were and is repeated after ClaimRetry.
 func (s *Scheduler) loop(ctx context.Context, p *plan) {
 	var done time.Time
+	var pending []change
 	for {
 		now := s.now()
-		if p != nil {
-			s.fire(ctx, p, done, now)
+		pending = append(pending, s.takeChanges()...)
+		failed := false
+		for len(pending) > 0 {
+			c := pending[0]
+			if c.at.After(done) {
+				if p != nil && !s.fire(ctx, p, done, c.at, now) {
+					failed = true
+					break
+				}
+				done = c.at
+			}
+			p = s.compileChange(ctx, c.settings)
+			pending = pending[1:]
+		}
+		if !failed {
+			if p == nil || s.fire(ctx, p, done, now, now) {
+				done = now
+			} else {
+				failed = true
+			}
 		}
-		done = now
 		if ctx.Err() != nil {
 			return
 		}
 
 		var timeout <-chan time.Time
-		if p != nil {
+		switch {
+		case failed:
+			timeout = s.after(ClaimRetry)
+		case p != nil:
 			if next, ok := p.next(now); ok {
 				timeout = s.after(min(next.Sub(now), maxWait))
 			}
@@ -347,45 +400,48 @@ func (s *Scheduler) loop(ctx context.Context, p *plan) {
 			return
 		case <-timeout:
 		case <-s.wake:
-			// Fire times up to the change follow the old settings.
-			now := s.now()
-			if p != nil {
-				s.fire(ctx, p, done, now)
-			}
-			done = now
-			s.latestMu.Lock()
-			settings := s.latest
-			s.latestMu.Unlock()
-			np, err := compile(s.schedules, settings)
-			if err != nil {
-				// Settings are validated before they are saved, so this
-				// is a bug; firing at a guessed time would be worse than
-				// not firing until the settings are fixed.
-				s.logger.ErrorContext(ctx, "scheduler: settings unusable; no schedule fires until they change", "error", err)
-			}
-			p = np
 		}
 	}
 }
 
+// compileChange returns the plan for changed settings, or nil when they
+// are unusable.
+func (s *Scheduler) compileChange(ctx context.Context, settings household.Settings) *plan {
+	p, err := compile(s.schedules, settings)
+	if err != nil {
+		// Settings are validated before they are saved, so this is a
+		// bug; firing at a guessed time would be worse than not firing
+		// until the settings are fixed.
+		s.logger.ErrorContext(ctx, "scheduler: settings unusable; no schedule fires until they change", "error", err)
+		return nil
+	}
+	return p
+}
+
 // fire handles, for every schedule, its most recent fire time at or before
-// now if it is after done.
-func (s *Scheduler) fire(ctx context.Context, p *plan, done, now time.Time) {
+// until if it is after done; now is the current instant. It reports false
+// when a claim failed, so the fire times it covered must be handled again.
+func (s *Scheduler) fire(ctx context.Context, p *plan, done, until, now time.Time) bool {
+	ok := true
 	for _, it := range p.items {
 		if ctx.Err() != nil {
-			return
+			return ok
 		}
-		at, ok := fireBefore(it.rule, p.loc, now)
-		if !ok || !at.After(done) {
+		at, due := fireBefore(it.rule, p.loc, until)
+		if !due || !at.After(done) {
 			continue
 		}
-		s.handle(ctx, it, p.loc, at, now)
+		if !s.handle(ctx, it, p.loc, at, now) {
+			ok = false
+		}
 	}
+	return ok
 }
 
 // handle claims the period of it fired at at and runs it, or records it as
-// missed when at is more than CatchUpWindow before now.
-func (s *Scheduler) handle(ctx context.Context, it item, loc *time.Location, at, now time.Time) {
+// missed when at is more than CatchUpWindow before now. It reports false
+// when the claim failed and the period is still unclaimed.
+func (s *Scheduler) handle(ctx context.Context, it item, loc *time.Location, at, now time.Time) bool {
 	date := household.Today(at, loc)
 	period := Period{Key: it.rule.PeriodKey(date), FireAt: at, Date: date}
 	if it.rule.NextWeek {
@@ -397,21 +453,22 @@ func (s *Scheduler) handle(ctx context.Context, it item, loc *time.Location, at,
 
 	claimed, err := s.claim(ctx, it.sch.Name, period.Key, now)
 	if err != nil {
-		log.ErrorContext(ctx, "scheduler: claim period failed; period not run", "error", err)
-		return
+		log.ErrorContext(ctx, "scheduler: claim period failed; retrying", "error", err, "retry_in", ClaimRetry.String())
+		return false
 	}
 	if !claimed {
-		return
+		return true
 	}
 	if late := now.Sub(at); late > CatchUpWindow {
 		log.WarnContext(ctx, "scheduler: period missed; recorded without running", "late", late.String())
-		return
+		return true
 	}
 	if err := it.sch.Run(ctx, period); err != nil {
 		log.ErrorContext(ctx, "scheduler: run failed", "error", err)
-		return
+		return true
 	}
 	log.InfoContext(ctx, "scheduler: ran")
+	return true
 }
 
 // claim inserts schedule_runs(name, key) and reports whether this call
diff --git a/internal/scheduler/scheduler_test.go b/internal/scheduler/scheduler_test.go
index 2bb6183..51b30d1 100644
--- a/internal/scheduler/scheduler_test.go
+++ b/internal/scheduler/scheduler_test.go
@@ -565,6 +565,76 @@ func TestSettingsChangeMovesNextFire(t *testing.T) {
 	wantKeys(t, "the new time", rec.keys(), []string{"2026-10-06", "2026-10-07", "2026-10-09"})
 }
 
+// A change saved at 12:00 moving the 16:30 reminder to 12:01 must fire it
+// even when the scheduler only gets to the change at 12:02, after the new
+// fire time has passed.
+func TestSettingsChangeAppliedLateStillFires(t *testing.T) {
+	e := newEnv(t)
+	var rec recorder
+	// blocker holds the scheduler in a Run at 12:00 on 2026-10-07 while
+	// the settings change and the clock moves on.
+	started := make(chan struct{})
+	release := make(chan struct{})
+	blocker := scheduler.Schedule{
+		Name: scheduler.RotationCheck,
+		Rule: func(household.Settings) scheduler.Rule { return scheduler.Daily("12:00") },
+		Run: func(_ context.Context, p scheduler.Period) error {
+			if p.Key == "2026-10-07" {
+				close(started)
+				<-release
+			}
+			return nil
+		},
+	}
+	r := e.start(at(2026, 10, 7, 11, 0), reminder(&rec), blocker)
+	wantKeys(t, "boot", rec.keys(), []string{"2026-10-06"})
+
+	r.clock.set(at(2026, 10, 7, 12, 0))
+	<-started
+	e.update(household.SettingsPatch{ReminderTime: ptr("12:01")})
+	r.clock.set(at(2026, 10, 7, 12, 2))
+	close(release)
+	r.idle() // the pass the blocker held
+	r.idle() // the pass that applies the change
+	wantKeys(t, "after the late change", rec.keys(), []string{"2026-10-06", "2026-10-07"})
+
+	r.runUntil(at(2026, 10, 7, 23, 0))
+	wantKeys(t, "end of day", rec.keys(), []string{"2026-10-06", "2026-10-07"})
+	wantKeys(t, "claims", e.claimed(scheduler.DayOfReminder), []string{"2026-10-06", "2026-10-07"})
+}
+
+// A claim that fails is retried, not skipped, and the period then runs once.
+func TestClaimFailureIsRetried(t *testing.T) {
+	e := newEnv(t)
+	for _, q := range []string{
+		`CREATE TABLE fail_claims (on_ INTEGER NOT NULL)`,
+		`INSERT INTO fail_claims VALUES (1)`,
+		`CREATE TRIGGER fail_claim BEFORE INSERT ON schedule_runs
+			WHEN (SELECT on_ FROM fail_claims) = 1
+			BEGIN SELECT RAISE(ABORT, 'injected claim failure'); END`,
+	} {
+		if _, err := e.st.Writer.Exec(q); err != nil {
+			t.Fatal(err)
+		}
+	}
+	var rec recorder
+	r := e.start(at(2026, 10, 7, 12, 0), reminder(&rec))
+	wantKeys(t, "boot with failing claims", rec.keys(), nil)
+	wantKeys(t, "claims while failing", e.claimed(scheduler.DayOfReminder), nil)
+	if next, _ := r.clock.nextDeadline(); !next.Equal(at(2026, 10, 7, 12, 0).Add(scheduler.ClaimRetry)) {
+		t.Errorf("next wake %s, want a retry after %s", next, scheduler.ClaimRetry)
+	}
+
+	if _, err := e.st.Writer.Exec(`UPDATE fail_claims SET on_ = 0`); err != nil {
+		t.Fatal(err)
+	}
+	r.runUntil(at(2026, 10, 7, 12, 0).Add(scheduler.ClaimRetry))
+	wantKeys(t, "after the retry", rec.keys(), []string{"2026-10-06"})
+	r.runUntil(at(2026, 10, 7, 23, 0))
+	wantKeys(t, "end of day", rec.keys(), []string{"2026-10-06", "2026-10-07"})
+	wantKeys(t, "claims", e.claimed(scheduler.DayOfReminder), []string{"2026-10-06", "2026-10-07"})
+}
+
 func TestSettingsChangeTimezone(t *testing.T) {
 	e := newEnv(t)
 	var rec recorder
-- 
2.55.0


From bcd875c11fa9bb58fd50a1505adc22e55d117242 Mon Sep 17 00:00:00 2001
From: dev_agent <dev_agent@cubecraftlabs.com>
Date: Fri, 9 Oct 2026 19:02:09 -0400
Subject: [PATCH 3/3] Retry a failed claim for its own period before later fire
 times

A failed claim was retried by recomputing the most recent fire time from
the new now, so a daily claim that failed on 7 October and recovered after
the 8 October fire claimed only 8 October: the 7 October period was never
claimed nor recorded as missed. The scheduler now keeps each failed
claim's fire time and claims that same period every ClaimRetry until it
succeeds, applies the catch-up rule to it then, and only afterwards
handles later fire times and settings changes.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---
 internal/scheduler/scheduler.go      | 90 ++++++++++++++++++----------
 internal/scheduler/scheduler_test.go | 48 ++++++++++++---
 2 files changed, 100 insertions(+), 38 deletions(-)

diff --git a/internal/scheduler/scheduler.go b/internal/scheduler/scheduler.go
index 38d3b57..9f8acee 100644
--- a/internal/scheduler/scheduler.go
+++ b/internal/scheduler/scheduler.go
@@ -26,9 +26,13 @@
 // # Claim failures
 //
 // A claim that fails (the database refused the insert) is never skipped:
-// the scheduler logs it and does not advance past it, but repeats the
-// whole pass every ClaimRetry until the claim succeeds. Periods the pass
-// already claimed are claimed again and skipped, so each still runs once.
+// the scheduler logs it and keeps that period's fire time, and every
+// ClaimRetry claims that same period again until the claim succeeds. Only
+// then does it apply the catch-up rule to it, running it if its fire time
+// is within CatchUpWindow of the successful claim and recording it as
+// missed otherwise, and only then does it go on to later fire times, even
+// when one has passed meanwhile. Each period is claimed by its own fire
+// time, so the later period runs once as well.
 //
 // # Time zone and DST
 //
@@ -351,36 +355,37 @@ func compile(schedules []Schedule, settings household.Settings) (*plan, error) {
 
 // loop fires schedules until ctx is cancelled. done is the instant up to
 // which fire times have been handled; it starts at the zero time so the
-// first pass catches up. Each pass first applies the pending settings
-// changes in order, handling fire times up to a change under the plan it
-// replaces, then handles fire times up to now under the current plan. A
-// pass with a failed claim keeps done and its unapplied changes where they
-// were and is repeated after ClaimRetry.
+// first pass catches up. retry is the periods whose claim failed, each
+// with its own fire time. Each pass first claims retry again; while any
+// claim there still fails, the pass stops and is repeated after
+// ClaimRetry. Otherwise it applies the pending settings changes in order,
+// handling fire times up to a change under the plan it replaces, then
+// handles fire times up to now under the current plan, stopping after the
+// first stretch in which a claim failed.
 func (s *Scheduler) loop(ctx context.Context, p *plan) {
 	var done time.Time
 	var pending []change
+	var retry []target
 	for {
 		now := s.now()
 		pending = append(pending, s.takeChanges()...)
-		failed := false
-		for len(pending) > 0 {
+		retry = s.retry(ctx, retry, now)
+		for len(retry) == 0 && len(pending) > 0 {
 			c := pending[0]
 			if c.at.After(done) {
-				if p != nil && !s.fire(ctx, p, done, c.at, now) {
-					failed = true
-					break
+				if p != nil {
+					retry = s.fire(ctx, p, done, c.at, now)
 				}
 				done = c.at
 			}
 			p = s.compileChange(ctx, c.settings)
 			pending = pending[1:]
 		}
-		if !failed {
-			if p == nil || s.fire(ctx, p, done, now, now) {
-				done = now
-			} else {
-				failed = true
+		if len(retry) == 0 {
+			if p != nil {
+				retry = s.fire(ctx, p, done, now, now)
 			}
+			done = now
 		}
 		if ctx.Err() != nil {
 			return
@@ -388,7 +393,7 @@ func (s *Scheduler) loop(ctx context.Context, p *plan) {
 
 		var timeout <-chan time.Time
 		switch {
-		case failed:
+		case len(retry) > 0:
 			timeout = s.after(ClaimRetry)
 		case p != nil:
 			if next, ok := p.next(now); ok {
@@ -418,31 +423,54 @@ func (s *Scheduler) compileChange(ctx context.Context, settings household.Settin
 	return p
 }
 
+// target is one schedule's period, named by the fire time it belongs to.
+type target struct {
+	it  item
+	loc *time.Location
+	at  time.Time
+}
+
 // fire handles, for every schedule, its most recent fire time at or before
-// until if it is after done; now is the current instant. It reports false
-// when a claim failed, so the fire times it covered must be handled again.
-func (s *Scheduler) fire(ctx context.Context, p *plan, done, until, now time.Time) bool {
-	ok := true
+// until if it is after done; now is the current instant. It returns the
+// targets whose claim failed, to be claimed again.
+func (s *Scheduler) fire(ctx context.Context, p *plan, done, until, now time.Time) []target {
+	var failed []target
 	for _, it := range p.items {
 		if ctx.Err() != nil {
-			return ok
+			return failed
 		}
 		at, due := fireBefore(it.rule, p.loc, until)
 		if !due || !at.After(done) {
 			continue
 		}
-		if !s.handle(ctx, it, p.loc, at, now) {
-			ok = false
+		if t := (target{it: it, loc: p.loc, at: at}); !s.handle(ctx, t, now) {
+			failed = append(failed, t)
 		}
 	}
-	return ok
+	return failed
 }
 
-// handle claims the period of it fired at at and runs it, or records it as
-// missed when at is more than CatchUpWindow before now. It reports false
+// retry handles targets again at now and returns those whose claim still
+// failed.
+func (s *Scheduler) retry(ctx context.Context, targets []target, now time.Time) []target {
+	var failed []target
+	for i, t := range targets {
+		if ctx.Err() != nil {
+			return append(failed, targets[i:]...)
+		}
+		if !s.handle(ctx, t, now) {
+			failed = append(failed, t)
+		}
+	}
+	return failed
+}
+
+// handle claims the period of t and runs it, or records it as missed when
+// its fire time is more than CatchUpWindow before now. It reports false
 // when the claim failed and the period is still unclaimed.
-func (s *Scheduler) handle(ctx context.Context, it item, loc *time.Location, at, now time.Time) bool {
-	date := household.Today(at, loc)
+func (s *Scheduler) handle(ctx context.Context, t target, now time.Time) bool {
+	it, at := t.it, t.at
+	date := household.Today(at, t.loc)
 	period := Period{Key: it.rule.PeriodKey(date), FireAt: at, Date: date}
 	if it.rule.NextWeek {
 		period.Week = household.WeekStart(addDays(date, 7))
diff --git a/internal/scheduler/scheduler_test.go b/internal/scheduler/scheduler_test.go
index 51b30d1..27af79a 100644
--- a/internal/scheduler/scheduler_test.go
+++ b/internal/scheduler/scheduler_test.go
@@ -603,9 +603,9 @@ func TestSettingsChangeAppliedLateStillFires(t *testing.T) {
 	wantKeys(t, "claims", e.claimed(scheduler.DayOfReminder), []string{"2026-10-06", "2026-10-07"})
 }
 
-// A claim that fails is retried, not skipped, and the period then runs once.
-func TestClaimFailureIsRetried(t *testing.T) {
-	e := newEnv(t)
+// failClaims makes every claim fail until recover is called.
+func (e *env) failClaims() (recover func()) {
+	e.t.Helper()
 	for _, q := range []string{
 		`CREATE TABLE fail_claims (on_ INTEGER NOT NULL)`,
 		`INSERT INTO fail_claims VALUES (1)`,
@@ -614,9 +614,21 @@ func TestClaimFailureIsRetried(t *testing.T) {
 			BEGIN SELECT RAISE(ABORT, 'injected claim failure'); END`,
 	} {
 		if _, err := e.st.Writer.Exec(q); err != nil {
-			t.Fatal(err)
+			e.t.Fatal(err)
 		}
 	}
+	return func() {
+		e.t.Helper()
+		if _, err := e.st.Writer.Exec(`UPDATE fail_claims SET on_ = 0`); err != nil {
+			e.t.Fatal(err)
+		}
+	}
+}
+
+// A claim that fails is retried, not skipped, and the period then runs once.
+func TestClaimFailureIsRetried(t *testing.T) {
+	e := newEnv(t)
+	recoverClaims := e.failClaims()
 	var rec recorder
 	r := e.start(at(2026, 10, 7, 12, 0), reminder(&rec))
 	wantKeys(t, "boot with failing claims", rec.keys(), nil)
@@ -625,9 +637,7 @@ func TestClaimFailureIsRetried(t *testing.T) {
 		t.Errorf("next wake %s, want a retry after %s", next, scheduler.ClaimRetry)
 	}
 
-	if _, err := e.st.Writer.Exec(`UPDATE fail_claims SET on_ = 0`); err != nil {
-		t.Fatal(err)
-	}
+	recoverClaims()
 	r.runUntil(at(2026, 10, 7, 12, 0).Add(scheduler.ClaimRetry))
 	wantKeys(t, "after the retry", rec.keys(), []string{"2026-10-06"})
 	r.runUntil(at(2026, 10, 7, 23, 0))
@@ -635,6 +645,30 @@ func TestClaimFailureIsRetried(t *testing.T) {
 	wantKeys(t, "claims", e.claimed(scheduler.DayOfReminder), []string{"2026-10-06", "2026-10-07"})
 }
 
+// A claim that keeps failing past the next daily fire is retried for its
+// own period, not the most recent one: that period, now more than
+// CatchUpWindow old, is recorded as missed, and the later period runs once.
+func TestClaimFailureRecoversAfterNextFire(t *testing.T) {
+	e := newEnv(t)
+	recoverClaims := e.failClaims()
+	var rec recorder
+	// The 7 October 16:30 reminder is due and its claim fails, past the
+	// 8 October 16:30 fire.
+	r := e.start(at(2026, 10, 7, 16, 45), reminder(&rec))
+	r.runUntil(at(2026, 10, 8, 17, 0))
+	wantKeys(t, "runs while failing", rec.keys(), nil)
+	wantKeys(t, "claims while failing", e.claimed(scheduler.DayOfReminder), nil)
+
+	// Claims recover 24h31m after the 7 October fire.
+	recoverClaims()
+	r.runUntil(at(2026, 10, 8, 17, 1))
+	wantKeys(t, "after recovery", rec.keys(), []string{"2026-10-08"})
+	wantKeys(t, "claims after recovery", e.claimed(scheduler.DayOfReminder), []string{"2026-10-07", "2026-10-08"})
+
+	r.runUntil(at(2026, 10, 9, 16, 0))
+	wantKeys(t, "next day", rec.keys(), []string{"2026-10-08"})
+}
+
 func TestSettingsChangeTimezone(t *testing.T) {
 	e := newEnv(t)
 	var rec recorder
-- 
2.55.0