Set the chart's placeholder version to 0.14.0

Cosmetic, as in 4c85e76 and 041e159. `make helm-package` passes --version
and --app-version from the tag, so neither field decides anything about
what release.yaml publishes.

Done anyway because a tree heading for v0.14.0 that still says 0.13.0
tells a reader something false. appVersion keeps the v, per
APPVERSION_PREFIX in .release.conf.

README.md

@@ -85,6 +85,16 @@ How a browser stays signed in:
 With `TERDUT_PUBLIC_URL` set, tapping a push notification opens the incident in
 the web UI (`/incidents/{id}`).
 
+A **Team** tab holds everything a team owns: the on-call rota, the escalation
+ladder, the alert sources with their keys, the dead man's switches and the
+membership. An owner edits it; a member sees the same page read-only, because
+the server refuses their writes anyway. Somebody in more than one team picks
+between them at the top.
+
+The **Admin** tab appears only for a system administrator, and holds what
+belongs to the whole server rather than to one team: every team, every user, and
+the settings that used to be environment variables.
+
 ### Docker
 
 ```bash
@@ -377,6 +387,50 @@ exhausts its retries. Written from the result rather than at enqueue, so the
 timeline says what actually happened — and a page that never landed is visible
 instead of looking the same as one that did.
 
+### Escalation
+
+Without a ladder, an unacknowledged incident re-pages the same topic every
+`notify_repeat` forever. That is a louder version of the same silence: if the
+person on call is asleep, out of signal, or has left, nothing else happens.
+
+A team can configure an ordered ladder instead. Each level has a timeout and a
+set of targets, and a target is either a named person or **whoever the team's
+rota says is on call today** — the target that keeps working when the rota
+changes and nobody remembers to edit the policy.
+
+```
+level 1   5m    oncall            the rota gets first refusal
+level 2   5m    user:bob          then a named second
+                                  then repeat_count more rounds
+                                  then the team's fallback topic, once
+```
+
+When a level's timeout passes with the incident still `triggered`, the next
+level is paged. Off the end of the ladder the whole thing runs again
+`repeat_count` times, and after that the team's `fallback_topic` is paged once
+as the end of the line. The incident stays open throughout: running out of
+people to wake is not the same as somebody answering.
+
+**Acknowledging or resolving stops it**, which is the point — continuing to wake
+people after somebody has said "I have this" is how a tool teaches people to
+mute it. **Snoozing pauses it**: a deliberate "not now" holds the ladder where
+it is, and it resumes when the snooze runs out.
+
+Every step is on the incident's timeline with the level and the names it woke,
+so somebody reading it afterwards can tell why their phone rang at 04:00. A
+level whose targets are all unreachable — no ntfy topic, a disabled account, an
+empty rota — is recorded as `nobody reachable` and the ladder moves on rather
+than stalling on a rung that cannot ring.
+
+**Reminders and escalation never both run.** A team with a ladder gets
+escalation; a team without keeps the reminder behaviour exactly as it was. Two
+pages for one silence is the surest way to get a tool muted.
+
+The ladder's `fallback_topic` is per team, unlike `TERDUT_NTFY_FALLBACK_TOPIC`,
+which is the install-wide topic used when an incident opens with nobody on call.
+They answer different questions: one is "nobody was scheduled", the other is
+"everybody scheduled has been tried".
+
 ### Stale alert expiry
 
 A resolved webhook is the only signal that an alert has stopped firing, so a
@@ -578,6 +632,8 @@ and was removed in v0.13.0 once senders had moved onto keys.
 | `GET` | `/api/teams/{teamID}/integrations` | member | List integrations. Never returns keys |
 | `POST` | `/api/teams/{teamID}/integrations` | **owner** | Mint an integration `{"name","kind"}` — key and URL shown once |
 | `DELETE` | `/api/teams/{teamID}/integrations/{integrationID}` | **owner** | Revoke an integration |
+| `GET` | `/api/teams/{teamID}/escalation` | member | The team's [escalation ladder](#escalation) `{repeat_count, fallback_topic, levels[]}`. Empty levels means the team has none |
+| `PUT` | `/api/teams/{teamID}/escalation` | **owner** | Replace it wholesale. `400` for a level with no targets or no timeout — a rung that pages nobody is a silence with a number on it |
 | `GET` | `/api/teams/{teamID}/deadman` | member | The team's [dead man's switch](#dead-mans-switch) configuration `{matchers, timeout_seconds, severity}` |
 | `PUT` | `/api/teams/{teamID}/deadman` | **owner** | Replace it. `400` when no matcher names an `alertname`, because a switch that silently watches nothing is the failure this feature exists to prevent |
 

charts/terdut-server/Chart.yaml

@@ -15,5 +15,5 @@ type: application
 # appVersion and image.tag in values.yaml no longer agree, and that is not an oversight:
 # image.tag stays "latest", which is what a local install actually pulls. appVersion is
 # metadata and drives nothing.
-version: 0.12.0
+version: 0.14.0
-appVersion: "v0.12.0"
+appVersion: "v0.14.0"

internal/api/alertmanager.go

@@ -403,12 +403,18 @@ func openIncident(ctx context.Context, q querier, notify NotifyConfig, teamID in
 		}
 	}
 
-	// Queue the page, but do not send it here: this runs inside a transaction on
+	// Queue the page, but do not send it here: this runs inside the webhook's
-	// a single-connection pool, so an HTTP call would hold up every other
+	// transaction, and an HTTP call would hold a connection open across a
-	// request. The notifier picks the row up within a tick.
+	// network round trip. The notifier picks the row up within a tick.
 	if err := enqueueOpened(ctx, q, notify, id, onCall); err != nil {
 		return 0, err
 	}
+
+	// And start the escalation clock, if the team keeps one. In the same
+	// transaction, so an incident is never briefly open with nobody counting.
+	if err := startEscalation(ctx, q, id, teamID); err != nil {
+		return 0, err
+	}
 	return id, nil
 }
 

internal/api/escalation.go

@@ -0,0 +1,506 @@
+package api
+
+import (
+	"context"
+	"database/sql"
+	"log"
+	"net/http"
+	"strconv"
+	"strings"
+	"time"
+)
+
+// evEscalated records a rung of the ladder on the incident's timeline: which
+// level, and who it woke.
+const evEscalated = "escalated"
+
+// escalationPolicy is a team's ladder, loaded whole. It is small — a handful of
+// levels with a few targets each — and every use needs all of it, so there is
+// no point reading it a level at a time.
+type escalationPolicy struct {
+	teamID        int64
+	repeatCount   int64
+	fallbackTopic string
+	levels        []escalationLevel
+}
+
+type escalationLevel struct {
+	id       int64
+	position int64
+	timeout  time.Duration
+	targets  []escalationTarget
+}
+
+type escalationTarget struct {
+	kind   string // "user" or "oncall"
+	userID *int64
+}
+
+// configured reports whether this team has anything to escalate through. A
+// policy row with no levels is the same as no policy: the team gets the
+// pre-escalation behaviour, which is reminders on the assignee's topic.
+func (p *escalationPolicy) configured() bool { return p != nil && len(p.levels) > 0 }
+
+// level returns the level at a 1-based position.
+func (p *escalationPolicy) level(pos int64) (escalationLevel, bool) {
+	for _, l := range p.levels {
+		if l.position == pos {
+			return l, true
+		}
+	}
+	return escalationLevel{}, false
+}
+
+// loadEscalationPolicy reads one team's ladder. A team with no policy row
+// returns nil, which every caller treats as "not configured" rather than as an
+// error: most teams will never set one up.
+func loadEscalationPolicy(ctx context.Context, q querier, teamID int64) (*escalationPolicy, error) {
+	p := &escalationPolicy{teamID: teamID}
+	err := q.QueryRowContext(ctx,
+		"SELECT repeat_count, fallback_topic FROM escalation_policies WHERE team_id = $1",
+		teamID).Scan(&p.repeatCount, &p.fallbackTopic)
+	if err == sql.ErrNoRows {
+		return nil, nil
+	}
+	if err != nil {
+		return nil, err
+	}
+
+	rows, err := q.QueryContext(ctx, `
+		SELECT l.id, l.position, l.timeout_seconds, t.kind, t.user_id
+		FROM   escalation_levels l
+		LEFT   JOIN escalation_targets t ON t.level_id = l.id
+		WHERE  l.team_id = $1
+		ORDER  BY l.position, t.id`, teamID)
+	if err != nil {
+		return nil, err
+	}
+	defer rows.Close()
+
+	byPosition := map[int64]int{} // position -> index in p.levels
+	for rows.Next() {
+		var id, position, timeout int64
+		var kind *string
+		var userID *int64
+		if err := rows.Scan(&id, &position, &timeout, &kind, &userID); err != nil {
+			return nil, err
+		}
+		idx, seen := byPosition[position]
+		if !seen {
+			p.levels = append(p.levels, escalationLevel{
+				id:       id,
+				position: position,
+				timeout:  time.Duration(timeout) * time.Second,
+			})
+			idx = len(p.levels) - 1
+			byPosition[position] = idx
+		}
+		// LEFT JOIN: a level with no targets yet still produces a row, with a
+		// NULL kind. It is a rung that pages nobody, which the API refuses to
+		// store but an older row could still hold.
+		if kind != nil {
+			p.levels[idx].targets = append(p.levels[idx].targets,
+				escalationTarget{kind: *kind, userID: userID})
+		}
+	}
+	return p, rows.Err()
+}
+
+// escalate advances every incident whose current level has run out of time.
+//
+// Runs on the notifier's tick, beside the reminder pass, because it is the same
+// question asked differently: reminders ask "has this been ignored long
+// enough to say it again", escalation asks "long enough to say it to somebody
+// else". Sharing the tick means one query cadence and one outbox.
+func escalate(ctx context.Context, db *sql.DB, cfg NotifyConfig) {
+	rows, err := db.QueryContext(ctx, `
+		SELECT i.id, i.team_id, i.escalation_level, i.escalation_level_at, i.escalation_round
+		FROM   incidents i
+		JOIN   escalation_policies p ON p.team_id = i.team_id
+		WHERE  i.resolved_at IS NULL
+		  AND  i.archived_at IS NULL
+		  AND  i.status = 'triggered'
+		  AND  (i.snoozed_until IS NULL OR i.snoozed_until <= $1)
+		  AND  i.escalation_level > 0`, time.Now().Unix())
+	if err != nil {
+		log.Printf("escalation: find due: %v", err)
+		return
+	}
+
+	type pending struct {
+		incidentID, teamID, level, round int64
+		levelAt                          int64
+	}
+	var due []pending
+	for rows.Next() {
+		var p pending
+		var levelAt *int64
+		if err := rows.Scan(&p.incidentID, &p.teamID, &p.level, &levelAt, &p.round); err != nil {
+			rows.Close()
+			log.Printf("escalation: scan: %v", err)
+			return
+		}
+		if levelAt == nil {
+			continue
+		}
+		p.levelAt = *levelAt
+		due = append(due, p)
+	}
+	rows.Close()
+	if err := rows.Err(); err != nil {
+		log.Printf("escalation: iterate: %v", err)
+		return
+	}
+
+	now := time.Now()
+	for _, d := range due {
+		policy, err := loadEscalationPolicy(ctx, db, d.teamID)
+		if err != nil {
+			log.Printf("escalation: load policy for team %d: %v", d.teamID, err)
+			continue
+		}
+		if !policy.configured() {
+			continue
+		}
+		current, ok := policy.level(d.level)
+		if !ok {
+			continue
+		}
+		if now.Sub(time.Unix(d.levelAt, 0)) < current.timeout {
+			continue
+		}
+		if err := advanceEscalation(ctx, db, cfg, policy, d.incidentID, d.level, d.round, now); err != nil {
+			log.Printf("escalation: advance incident %d: %v", d.incidentID, err)
+		}
+	}
+}
+
+// advanceEscalation moves one incident to its next rung, or off the end of the
+// ladder.
+//
+// The whole move is one transaction: the level, the page and the timeline entry
+// are one event, and an incident recorded as being at level 3 that nobody at
+// level 3 was told about is the worst of the possible half-states.
+func advanceEscalation(ctx context.Context, db *sql.DB, cfg NotifyConfig, policy *escalationPolicy, incidentID, level, round int64, now time.Time) error {
+	tx, err := db.BeginTx(ctx, nil)
+	if err != nil {
+		return err
+	}
+	defer tx.Rollback() //nolint:errcheck
+
+	next := level + 1
+	nextRound := round
+	if _, ok := policy.level(next); !ok {
+		// Off the end. Either start the chain again, or make the last call.
+		if round < policy.repeatCount {
+			next, nextRound = 1, round+1
+		} else {
+			if err := escalationExhausted(ctx, tx, policy, incidentID, now); err != nil {
+				return err
+			}
+			return tx.Commit()
+		}
+	}
+
+	target, ok := policy.level(next)
+	if !ok {
+		return nil
+	}
+	paged, err := pageLevel(ctx, tx, cfg, policy, incidentID, target)
+	if err != nil {
+		return err
+	}
+
+	if _, err := tx.ExecContext(ctx, `
+		UPDATE incidents
+		   SET escalation_level = $1, escalation_level_at = $2, escalation_round = $3
+		 WHERE id = $4`, next, now.Unix(), nextRound, incidentID); err != nil {
+		return err
+	}
+
+	detail := "level " + strconv.FormatInt(next, 10)
+	if nextRound > round {
+		detail += " (round " + strconv.FormatInt(nextRound+1, 10) + ")"
+	}
+	if len(paged) > 0 {
+		detail += ": " + strings.Join(paged, ", ")
+	} else {
+		// Worth recording loudly: the rung exists, its turn came, and it woke
+		// nobody. That is a policy that looks configured and is not.
+		detail += ": nobody reachable"
+	}
+	if err := logEvent(ctx, tx, incidentID, evEscalated, nil, nil, &detail); err != nil {
+		return err
+	}
+	return tx.Commit()
+}
+
+// escalationExhausted is the end of the line: the fallback topic, once, and a
+// timeline entry saying the ladder is finished. The incident stays triggered —
+// escalation running out is not the same as somebody answering.
+func escalationExhausted(ctx context.Context, tx *sql.Tx, policy *escalationPolicy, incidentID int64, now time.Time) error {
+	detail := "escalation exhausted"
+	if policy.fallbackTopic != "" {
+		if err := enqueueNotification(ctx, tx, incidentID, nil, policy.fallbackTopic, notifyEscalated); err != nil {
+			return err
+		}
+		detail += ": paged " + policy.fallbackTopic
+	} else {
+		detail += ": no fallback topic configured"
+	}
+
+	// Level 0 again, so the sweep stops considering it. The round counter is
+	// left where it is, as the record of how far it got.
+	if _, err := tx.ExecContext(ctx,
+		"UPDATE incidents SET escalation_level = 0, escalation_level_at = NULL WHERE id = $1",
+		incidentID); err != nil {
+		return err
+	}
+	return logEvent(ctx, tx, incidentID, evEscalated, nil, nil, &detail)
+}
+
+// pageLevel notifies every target of one level and reports who was woken.
+//
+// Each target gets its own outbox row, so each gets its own Acknowledge token:
+// the button in a notification must acknowledge as the person holding the
+// phone, not as whoever was paged first.
+func pageLevel(ctx context.Context, tx *sql.Tx, cfg NotifyConfig, policy *escalationPolicy, incidentID int64, level escalationLevel) ([]string, error) {
+	var paged []string
+	seen := map[int64]bool{}
+
+	for _, t := range level.targets {
+		userID := t.userID
+		if t.kind == "oncall" {
+			onCall, err := currentOnCall(ctx, tx, policy.teamID)
+			if err != nil {
+				return nil, err
+			}
+			if onCall == nil {
+				continue
+			}
+			userID = onCall
+		}
+		if userID == nil || seen[*userID] {
+			continue
+		}
+		seen[*userID] = true
+
+		var topic *string
+		var username string
+		if err := tx.QueryRowContext(ctx,
+			"SELECT ntfy_topic, username FROM users WHERE id = $1 AND disabled_at IS NULL",
+			*userID).Scan(&topic, &username); err != nil {
+			// A disabled or deleted account is not an error in the middle of an
+			// escalation: it is a target that cannot be woken, and the next
+			// level is the answer to that.
+			continue
+		}
+		if topic == nil || *topic == "" {
+			continue
+		}
+		if err := enqueueNotification(ctx, tx, incidentID, userID, *topic, notifyEscalated); err != nil {
+			return nil, err
+		}
+		paged = append(paged, username)
+	}
+	return paged, nil
+}
+
+// startEscalation puts a newly opened incident on the first rung, when its team
+// has a ladder. Called from openIncident, inside the same transaction, so an
+// incident is never briefly open with no escalation clock running.
+func startEscalation(ctx context.Context, q querier, incidentID, teamID int64) error {
+	policy, err := loadEscalationPolicy(ctx, q, teamID)
+	if err != nil || !policy.configured() {
+		return err
+	}
+	_, err = q.ExecContext(ctx,
+		"UPDATE incidents SET escalation_level = 1, escalation_level_at = $1 WHERE id = $2",
+		time.Now().Unix(), incidentID)
+	return err
+}
+
+// stopEscalation takes an incident off the ladder. Acknowledging or resolving
+// is somebody saying "I have this", and continuing to wake people after that is
+// the behaviour that teaches people to ignore the tool.
+func stopEscalation(ctx context.Context, q querier, incidentID int64) error {
+	_, err := q.ExecContext(ctx,
+		"UPDATE incidents SET escalation_level = 0, escalation_level_at = NULL WHERE id = $1",
+		incidentID)
+	return err
+}
+
+// handleGetEscalation returns a team's ladder.
+func handleGetEscalation(db *sql.DB) http.HandlerFunc {
+	return func(w http.ResponseWriter, r *http.Request) {
+		teamID, ok := teamParam(w, r)
+		if !ok {
+			return
+		}
+		if !requireTeamMember(w, r, teamID) {
+			return
+		}
+
+		policy, err := loadEscalationPolicy(r.Context(), db, teamID)
+		if err != nil {
+			respond(w, http.StatusInternalServerError, errResp("internal error"))
+			return
+		}
+		respond(w, http.StatusOK, escalationResponse(policy, teamID))
+	}
+}
+
+type escalationLevelJSON struct {
+	Position       int64                  `json:"position"`
+	TimeoutSeconds int64                  `json:"timeout_seconds"`
+	Targets        []escalationTargetJSON `json:"targets"`
+}
+
+type escalationTargetJSON struct {
+	Kind   string `json:"kind"`
+	UserID *int64 `json:"user_id,omitempty"`
+}
+
+type escalationJSON struct {
+	TeamID        int64                 `json:"team_id"`
+	RepeatCount   int64                 `json:"repeat_count"`
+	FallbackTopic string                `json:"fallback_topic"`
+	Levels        []escalationLevelJSON `json:"levels"`
+}
+
+func escalationResponse(p *escalationPolicy, teamID int64) escalationJSON {
+	out := escalationJSON{TeamID: teamID, Levels: []escalationLevelJSON{}}
+	if p == nil {
+		return out
+	}
+	out.RepeatCount = p.repeatCount
+	out.FallbackTopic = p.fallbackTopic
+	for _, l := range p.levels {
+		level := escalationLevelJSON{
+			Position:       l.position,
+			TimeoutSeconds: int64(l.timeout.Seconds()),
+			Targets:        []escalationTargetJSON{},
+		}
+		for _, t := range l.targets {
+			level.Targets = append(level.Targets, escalationTargetJSON{Kind: t.kind, UserID: t.userID})
+		}
+		out.Levels = append(out.Levels, level)
+	}
+	return out
+}
+
+// handleSetEscalation replaces a team's ladder wholesale.
+//
+// Replace rather than patch: the levels are an order, and an API that edits one
+// rung has to answer what happens to the numbering of the others. Sending the
+// whole ladder makes the order the client's to decide and the server's to
+// store, and makes an edit atomic — there is no moment where level 2 exists
+// twice.
+func handleSetEscalation(db *sql.DB) http.HandlerFunc {
+	return func(w http.ResponseWriter, r *http.Request) {
+		teamID, ok := teamParam(w, r)
+		if !ok {
+			return
+		}
+		if !requireTeamOwner(w, r, teamID) {
+			return
+		}
+
+		var req escalationJSON
+		if err := decodeJSON(r, &req); err != nil {
+			respond(w, http.StatusBadRequest, errResp("invalid request body"))
+			return
+		}
+		if req.RepeatCount < 0 || req.RepeatCount > 10 {
+			respond(w, http.StatusBadRequest, errResp("repeat_count must be between 0 and 10"))
+			return
+		}
+		for i, l := range req.Levels {
+			if l.TimeoutSeconds <= 0 {
+				respond(w, http.StatusBadRequest, errResp("every level needs a timeout"))
+				return
+			}
+			if len(l.Targets) == 0 {
+				// A rung that pages nobody is not a delay, it is a silence with
+				// a number on it.
+				respond(w, http.StatusBadRequest,
+					errResp("level "+strconv.FormatInt(int64(i+1), 10)+" has no targets"))
+				return
+			}
+			for _, t := range l.Targets {
+				switch t.Kind {
+				case "oncall":
+					if t.UserID != nil {
+						respond(w, http.StatusBadRequest, errResp("an oncall target takes no user_id"))
+						return
+					}
+				case "user":
+					if t.UserID == nil {
+						respond(w, http.StatusBadRequest, errResp("a user target needs a user_id"))
+						return
+					}
+				default:
+					respond(w, http.StatusBadRequest, errResp("target kind must be user or oncall"))
+					return
+				}
+			}
+		}
+
+		tx, err := db.BeginTx(r.Context(), nil)
+		if err != nil {
+			respond(w, http.StatusInternalServerError, errResp("internal error"))
+			return
+		}
+		defer tx.Rollback() //nolint:errcheck
+
+		if _, err := tx.ExecContext(r.Context(), `
+			INSERT INTO escalation_policies (team_id, repeat_count, fallback_topic, updated_at)
+			VALUES ($1, $2, $3, `+nowEpoch+`)
+			ON CONFLICT (team_id) DO UPDATE SET
+				repeat_count = excluded.repeat_count,
+				fallback_topic = excluded.fallback_topic,
+				updated_at = excluded.updated_at`,
+			teamID, req.RepeatCount, req.FallbackTopic); err != nil {
+			respond(w, http.StatusInternalServerError, errResp("internal error"))
+			return
+		}
+		// The levels are replaced, not merged; the cascade takes the targets.
+		if _, err := tx.ExecContext(r.Context(),
+			"DELETE FROM escalation_levels WHERE team_id = $1", teamID); err != nil {
+			respond(w, http.StatusInternalServerError, errResp("internal error"))
+			return
+		}
+
+		for i, l := range req.Levels {
+			var levelID int64
+			if err := tx.QueryRowContext(r.Context(), `
+				INSERT INTO escalation_levels (team_id, position, timeout_seconds)
+				VALUES ($1, $2, $3) RETURNING id`,
+				teamID, int64(i+1), l.TimeoutSeconds).Scan(&levelID); err != nil {
+				respond(w, http.StatusInternalServerError, errResp("internal error"))
+				return
+			}
+			for _, t := range l.Targets {
+				if _, err := tx.ExecContext(r.Context(), `
+					INSERT INTO escalation_targets (level_id, kind, user_id)
+					VALUES ($1, $2, $3)`, levelID, t.Kind, t.UserID); err != nil {
+					// The only foreign key here is the user.
+					respond(w, http.StatusBadRequest, errResp("unknown user in targets"))
+					return
+				}
+			}
+		}
+
+		if err := tx.Commit(); err != nil {
+			respond(w, http.StatusInternalServerError, errResp("internal error"))
+			return
+		}
+
+		policy, err := loadEscalationPolicy(r.Context(), db, teamID)
+		if err != nil {
+			respond(w, http.StatusInternalServerError, errResp("internal error"))
+			return
+		}
+		respond(w, http.StatusOK, escalationResponse(policy, teamID))
+	}
+}

internal/api/escalation_test.go

@@ -0,0 +1,385 @@
+package api_test
+
+import (
+	"net/http"
+	"strings"
+	"testing"
+	"time"
+
+	"git.ryuvia.com/niklas/terdut-server/internal/api"
+)
+
+// teamUser creates a user in the default team with an ntfy topic, so they can
+// actually be paged.
+func teamUser(t *testing.T, s *ts, username, topic string) int64 {
+	t.Helper()
+	var user struct {
+		ID int64 `json:"id"`
+	}
+	decode(t, s.req(t, http.MethodPost, "/api/users",
+		map[string]string{"username": username, "email": username + "@test.com"}), &user)
+	resp := s.req(t, http.MethodPost, "/api/teams/"+defaultTeam+"/members",
+		map[string]any{"user_id": user.ID, "role": "member"})
+	resp.Body.Close()
+	setTopic(t, s, int(user.ID), topic)
+	return user.ID
+}
+
+// Escalation is all timeouts, and there is no fake clock in this package. The
+// tests back-date escalation_level_at instead, which is the same trick the dead
+// man's switch tests use on received_at: the sweeper reads a stored timestamp,
+// so moving the timestamp is moving the clock.
+
+// ladder configures the default team with two levels: the rota first, then a
+// named person, then the fallback topic.
+func ladder(t *testing.T, s *ts, secondUserID int64, repeat int64, fallback string) {
+	t.Helper()
+	resp := s.req(t, http.MethodPut, "/api/teams/"+defaultTeam+"/escalation", map[string]any{
+		"repeat_count":   repeat,
+		"fallback_topic": fallback,
+		"levels": []map[string]any{
+			{"timeout_seconds": 300, "targets": []map[string]any{{"kind": "oncall"}}},
+			{"timeout_seconds": 300, "targets": []map[string]any{{"kind": "user", "user_id": secondUserID}}},
+		},
+	})
+	defer resp.Body.Close()
+	if resp.StatusCode != http.StatusOK {
+		t.Fatalf("configure the ladder: %d", resp.StatusCode)
+	}
+}
+
+// overdue back-dates an incident's current level so its timeout has passed.
+func overdue(t *testing.T, s *ts, incidentID int64) {
+	t.Helper()
+	s.exec(t, "UPDATE incidents SET escalation_level_at = $1 WHERE id = $2",
+		time.Now().Add(-time.Hour).Unix(), incidentID)
+}
+
+func escalationLevel(t *testing.T, s *ts, incidentID int64) (level, round int64) {
+	t.Helper()
+	if err := s.db.QueryRow(
+		"SELECT escalation_level, escalation_round FROM incidents WHERE id = $1",
+		incidentID).Scan(&level, &round); err != nil {
+		t.Fatalf("read escalation state: %v", err)
+	}
+	return level, round
+}
+
+// The whole point: nobody answers, so somebody else is woken.
+func TestEscalation_PagesTheNextLevel(t *testing.T) {
+	s, f := notifyTS(t, api.NotifyConfig{PublicURL: "https://terdut.example.com", RepeatEvery: 15 * time.Minute})
+	second := teamUser(t, s, "second", "terdut-second")
+	ladder(t, s, second, 0, "terdut-fallback")
+
+	postWebhook(t, s, []map[string]any{
+		amAlert("fp-esc", "DiskFull", "firing", "2026-05-20T10:00:00Z", zeroTime, nil),
+	})
+	s.sweepNotify(t)
+
+	// Level 1 is the rota, so the first page went to the admin.
+	if level, _ := escalationLevel(t, s, 1); level != 1 {
+		t.Fatalf("a new incident should start at level 1, got %d", level)
+	}
+	if got := f.topicsSince(t); len(got) == 0 || got[0] != "terdut-admin" {
+		t.Fatalf("the first page should go to the on-call user, went to %v", got)
+	}
+
+	// Time passes with no acknowledgement.
+	f.forget()
+	overdue(t, s, 1)
+	s.sweepNotify(t)
+
+	if level, _ := escalationLevel(t, s, 1); level != 2 {
+		t.Errorf("expected level 2, got %d", level)
+	}
+	if got := f.topicsSince(t); len(got) != 1 || got[0] != "terdut-second" {
+		t.Errorf("level 2 should page the named user, paged %v", got)
+	}
+
+	// And the timeline says so, which is what somebody reads afterwards to
+	// understand why their phone rang at 04:00.
+	timeline := list(t, s.req(t, http.MethodGet, "/api/incidents/1/timeline", nil))
+	found := ""
+	for _, e := range timeline {
+		if e["type"] == "escalated" {
+			found, _ = e["detail"].(string)
+		}
+	}
+	if found == "" {
+		t.Error("the timeline should record the escalation")
+	} else if !strings.HasPrefix(found, "level 2") || !strings.Contains(found, "second") {
+		t.Errorf("the escalation entry should say which level and who: %q", found)
+	}
+}
+
+// Acknowledging is somebody saying "I have this". Nobody else should be woken.
+func TestEscalation_AcknowledgementStopsIt(t *testing.T) {
+	s, f := notifyTS(t, api.NotifyConfig{PublicURL: "https://terdut.example.com", RepeatEvery: 15 * time.Minute})
+	second := teamUser(t, s, "second", "terdut-second")
+	ladder(t, s, second, 0, "terdut-fallback")
+
+	postWebhook(t, s, []map[string]any{
+		amAlert("fp-ack", "DiskFull", "firing", "2026-05-20T10:00:00Z", zeroTime, nil),
+	})
+	s.sweepNotify(t)
+
+	s.req(t, http.MethodPost, "/api/incidents/1/acknowledge", nil).Body.Close()
+	if level, _ := escalationLevel(t, s, 1); level != 0 {
+		t.Errorf("acknowledging should take the incident off the ladder, level is %d", level)
+	}
+
+	f.forget()
+	overdue(t, s, 1) // no-op: level is 0, so there is nothing due
+	s.sweepNotify(t)
+	if got := f.topicsSince(t); len(got) != 0 {
+		t.Errorf("an acknowledged incident should page nobody, paged %v", got)
+	}
+}
+
+// Resolving stops it too, and by the same mechanism.
+func TestEscalation_ResolutionStopsIt(t *testing.T) {
+	s, f := notifyTS(t, api.NotifyConfig{PublicURL: "https://terdut.example.com", RepeatEvery: 15 * time.Minute})
+	second := teamUser(t, s, "second", "terdut-second")
+	ladder(t, s, second, 0, "terdut-fallback")
+
+	postWebhook(t, s, []map[string]any{
+		amAlert("fp-res", "DiskFull", "firing", "2026-05-20T10:00:00Z", zeroTime, nil),
+	})
+	s.sweepNotify(t)
+	s.req(t, http.MethodPost, "/api/incidents/1/resolve", nil).Body.Close()
+
+	f.forget()
+	overdue(t, s, 1)
+	s.sweepNotify(t)
+	if level, _ := escalationLevel(t, s, 1); level != 0 {
+		t.Errorf("a resolved incident should be off the ladder, level is %d", level)
+	}
+}
+
+// Snoozing is a deliberate "not now", so the ladder waits rather than carrying
+// on without the person who asked for quiet.
+func TestEscalation_SnoozePausesIt(t *testing.T) {
+	s, f := notifyTS(t, api.NotifyConfig{PublicURL: "https://terdut.example.com", RepeatEvery: 15 * time.Minute})
+	second := teamUser(t, s, "second", "terdut-second")
+	ladder(t, s, second, 0, "terdut-fallback")
+
+	postWebhook(t, s, []map[string]any{
+		amAlert("fp-snooze", "DiskFull", "firing", "2026-05-20T10:00:00Z", zeroTime, nil),
+	})
+	s.sweepNotify(t)
+
+	resp := s.req(t, http.MethodPost, "/api/incidents/1/snooze", map[string]any{"duration": "1h"})
+	resp.Body.Close()
+
+	f.forget()
+	overdue(t, s, 1)
+	s.sweepNotify(t)
+
+	if level, _ := escalationLevel(t, s, 1); level != 1 {
+		t.Errorf("a snoozed incident should stay where it is, level is %d", level)
+	}
+	if got := f.topicsSince(t); len(got) != 0 {
+		t.Errorf("a snoozed incident should page nobody, paged %v", got)
+	}
+
+	// When the snooze ends, the ladder picks up where it left off.
+	s.exec(t, "UPDATE incidents SET snoozed_until = $1 WHERE id = 1", time.Now().Add(-time.Minute).Unix())
+	s.sweepNotify(t)
+	if level, _ := escalationLevel(t, s, 1); level != 2 {
+		t.Errorf("after the snooze the ladder should resume, level is %d", level)
+	}
+}
+
+// Running out of ladder pages the team's fallback topic once, and says so.
+func TestEscalation_ExhaustionPagesTheFallback(t *testing.T) {
+	s, f := notifyTS(t, api.NotifyConfig{PublicURL: "https://terdut.example.com", RepeatEvery: 15 * time.Minute})
+	second := teamUser(t, s, "second", "terdut-second")
+	ladder(t, s, second, 0, "terdut-fallback")
+
+	postWebhook(t, s, []map[string]any{
+		amAlert("fp-end", "DiskFull", "firing", "2026-05-20T10:00:00Z", zeroTime, nil),
+	})
+	s.sweepNotify(t)
+
+	overdue(t, s, 1)
+	s.sweepNotify(t) // level 2
+	f.forget()
+	overdue(t, s, 1)
+	s.sweepNotify(t) // off the end
+
+	if got := f.topicsSince(t); len(got) != 1 || got[0] != "terdut-fallback" {
+		t.Errorf("exhaustion should page the fallback topic once, paged %v", got)
+	}
+	level, _ := escalationLevel(t, s, 1)
+	if level != 0 {
+		t.Errorf("an exhausted ladder should stop asking, level is %d", level)
+	}
+
+	// The incident is still open: running out of people is not an answer.
+	var status string
+	if err := s.db.QueryRow("SELECT status FROM incidents WHERE id = 1").Scan(&status); err != nil {
+		t.Fatal(err)
+	}
+	if status != "triggered" {
+		t.Errorf("exhaustion must not resolve the incident, status is %q", status)
+	}
+}
+
+// repeat_count walks the whole ladder again before giving up.
+func TestEscalation_RepeatsTheChain(t *testing.T) {
+	s, f := notifyTS(t, api.NotifyConfig{PublicURL: "https://terdut.example.com", RepeatEvery: 15 * time.Minute})
+	second := teamUser(t, s, "second", "terdut-second")
+	ladder(t, s, second, 1, "terdut-fallback") // one extra round
+
+	postWebhook(t, s, []map[string]any{
+		amAlert("fp-repeat", "DiskFull", "firing", "2026-05-20T10:00:00Z", zeroTime, nil),
+	})
+	s.sweepNotify(t)
+
+	overdue(t, s, 1)
+	s.sweepNotify(t) // level 2
+	f.forget()
+	overdue(t, s, 1)
+	s.sweepNotify(t) // back to level 1, round 2
+
+	level, round := escalationLevel(t, s, 1)
+	if level != 1 || round != 1 {
+		t.Errorf("expected level 1 round 1, got level %d round %d", level, round)
+	}
+	if got := f.topicsSince(t); len(got) != 1 || got[0] != "terdut-admin" {
+		t.Errorf("the second round should start at the top again, paged %v", got)
+	}
+}
+
+// A team without a ladder keeps exactly the behaviour it had, and never gets
+// both a reminder and an escalation for the same silence.
+func TestEscalation_WithoutAPolicyRemindersStillRun(t *testing.T) {
+	s, f := notifyTS(t, api.NotifyConfig{PublicURL: "https://terdut.example.com", RepeatEvery: 15 * time.Minute})
+	postWebhook(t, s, []map[string]any{
+		amAlert("fp-noesc", "DiskFull", "firing", "2026-05-20T10:00:00Z", zeroTime, nil),
+	})
+	s.sweepNotify(t)
+
+	// Age the first notification past the repeat interval.
+	f.forget()
+	s.exec(t, "UPDATE notifications SET created_at = $1, sent_at = $1",
+		time.Now().Add(-time.Hour).Unix())
+	s.sweepNotify(t)
+
+	if got := f.topicsSince(t); len(got) != 1 || got[0] != "terdut-admin" {
+		t.Errorf("without a ladder the reminder should still fire, paged %v", got)
+	}
+	if level, _ := escalationLevel(t, s, 1); level != 0 {
+		t.Errorf("an incident in a team with no ladder should not be on one, level is %d", level)
+	}
+}
+
+// With a ladder, reminders stop: two pages for one silence is how people learn
+// to mute the tool.
+func TestEscalation_WithAPolicyRemindersDoNotAlsoFire(t *testing.T) {
+	s, f := notifyTS(t, api.NotifyConfig{PublicURL: "https://terdut.example.com", RepeatEvery: 15 * time.Minute})
+	second := teamUser(t, s, "second", "terdut-second")
+	ladder(t, s, second, 0, "terdut-fallback")
+
+	postWebhook(t, s, []map[string]any{
+		amAlert("fp-both", "DiskFull", "firing", "2026-05-20T10:00:00Z", zeroTime, nil),
+	})
+	s.sweepNotify(t)
+
+	f.forget()
+	// Old enough for a reminder, but not yet due for escalation.
+	s.exec(t, "UPDATE notifications SET created_at = $1, sent_at = $1",
+		time.Now().Add(-time.Hour).Unix())
+	s.sweepNotify(t)
+
+	if got := f.topicsSince(t); len(got) != 0 {
+		t.Errorf("a team with a ladder should not also get reminders, paged %v", got)
+	}
+}
+
+// The API refuses a ladder that cannot page anybody.
+func TestEscalation_RejectsAnUnusablePolicy(t *testing.T) {
+	s := newTS(t)
+
+	for _, c := range []struct {
+		name string
+		body map[string]any
+	}{
+		{"a level with no targets", map[string]any{
+			"levels": []map[string]any{{"timeout_seconds": 300, "targets": []map[string]any{}}},
+		}},
+		{"a level with no timeout", map[string]any{
+			"levels": []map[string]any{{"timeout_seconds": 0, "targets": []map[string]any{{"kind": "oncall"}}}},
+		}},
+		{"a user target with no user", map[string]any{
+			"levels": []map[string]any{{"timeout_seconds": 300, "targets": []map[string]any{{"kind": "user"}}}},
+		}},
+		{"an unknown target kind", map[string]any{
+			"levels": []map[string]any{{"timeout_seconds": 300, "targets": []map[string]any{{"kind": "everybody"}}}},
+		}},
+		{"an absurd repeat count", map[string]any{
+			"repeat_count": 99,
+			"levels":       []map[string]any{{"timeout_seconds": 300, "targets": []map[string]any{{"kind": "oncall"}}}},
+		}},
+	} {
+		resp := s.req(t, http.MethodPut, "/api/teams/"+defaultTeam+"/escalation", c.body)
+		resp.Body.Close()
+		if resp.StatusCode != http.StatusBadRequest {
+			t.Errorf("%s: expected 400, got %d", c.name, resp.StatusCode)
+		}
+	}
+}
+
+// Editing the ladder is an owner's job; reading it is any member's.
+func TestEscalation_OwnerOnlyToEdit(t *testing.T) {
+	s := newTS(t)
+	_, call := member(t, s, "plain")
+
+	resp := call(http.MethodPut, "/api/teams/"+defaultTeam+"/escalation", map[string]any{
+		"levels": []map[string]any{{"timeout_seconds": 300, "targets": []map[string]any{{"kind": "oncall"}}}},
+	})
+	resp.Body.Close()
+	if resp.StatusCode != http.StatusForbidden {
+		t.Errorf("a member editing the ladder: expected 403, got %d", resp.StatusCode)
+	}
+
+	resp = call(http.MethodGet, "/api/teams/"+defaultTeam+"/escalation", nil)
+	resp.Body.Close()
+	if resp.StatusCode != http.StatusOK {
+		t.Errorf("a member reading the ladder: expected 200, got %d", resp.StatusCode)
+	}
+}
+
+// A target who cannot be woken is not a reason to stop: the next level is the
+// answer to an unreachable one.
+func TestEscalation_SkipsUnreachableTargets(t *testing.T) {
+	s, f := notifyTS(t, api.NotifyConfig{PublicURL: "https://terdut.example.com", RepeatEvery: 15 * time.Minute})
+	// Second user has no ntfy topic at all.
+	var user struct {
+		ID int64 `json:"id"`
+	}
+	decode(t, s.req(t, http.MethodPost, "/api/users",
+		map[string]string{"username": "silent", "email": "silent@test.com"}), &user)
+	s.req(t, http.MethodPost, "/api/teams/"+defaultTeam+"/members",
+		map[string]any{"user_id": user.ID, "role": "member"}).Body.Close()
+
+	ladder(t, s, user.ID, 0, "terdut-fallback")
+
+	postWebhook(t, s, []map[string]any{
+		amAlert("fp-silent", "DiskFull", "firing", "2026-05-20T10:00:00Z", zeroTime, nil),
+	})
+	s.sweepNotify(t)
+
+	f.forget()
+	overdue(t, s, 1)
+	s.sweepNotify(t)
+
+	// Level 2 was entered even though it woke nobody, so the ladder keeps
+	// moving toward the fallback rather than stalling on a silent rung.
+	if level, _ := escalationLevel(t, s, 1); level != 2 {
+		t.Errorf("expected the ladder to advance past an unreachable target, level is %d", level)
+	}
+	if got := f.topicsSince(t); len(got) != 0 {
+		t.Errorf("a target with no topic should page nothing, paged %v", got)
+	}
+}

internal/api/incident_store.go

@@ -52,6 +52,13 @@ type querier interface {
 
 const incidentSelectFrom = `
 	SELECT i.id, i.team_id, t.name, i.group_key, i.title, i.group_labels, i.status, i.severity,
+	       i.escalation_level,
+	       -- When this level runs out. Computed here rather than in Go because
+	       -- the timeout lives beside the level in the policy, and one join is
+	       -- cheaper than a second query per incident in a list.
+	       (SELECT i.escalation_level_at + el.timeout_seconds
+	          FROM escalation_levels el
+	         WHERE el.team_id = i.team_id AND el.position = i.escalation_level),
 	       i.triggered_at,
 	       i.acknowledged_by, i.acknowledged_at, ack.username,
 	       i.assigned_to, asg.username, i.snoozed_until,
@@ -65,10 +72,11 @@ func scanIncident(s scanner) (models.Incident, error) {
 	var i models.Incident
 	var groupLabelsJSON string
 	var triggeredAt int64
-	var ackAt, snoozedUntil, resolvedAt, archivedAt *int64
+	var ackAt, snoozedUntil, resolvedAt, archivedAt, escalationDue *int64
 
 	if err := s.Scan(
 		&i.ID, &i.TeamID, &i.TeamName, &i.GroupKey, &i.Title, &groupLabelsJSON, &i.Status, &i.Severity,
+		&i.EscalationLevel, &escalationDue,
 		&triggeredAt,
 		&i.AcknowledgedByID, &ackAt, &i.AcknowledgedByUser,
 		&i.AssignedToID, &i.AssignedToUser, &snoozedUntil,
@@ -83,6 +91,7 @@ func scanIncident(s scanner) (models.Incident, error) {
 	i.SnoozedUntil = unixPtr(snoozedUntil)
 	i.ResolvedAt = unixPtr(resolvedAt)
 	i.ArchivedAt = unixPtr(archivedAt)
+	i.EscalationDueAt = unixPtr(escalationDue)
 	return i, nil
 }
 
@@ -235,6 +244,9 @@ func resolveIfSettled(ctx context.Context, q querier, incidentID int64) (bool, e
 	if n == 0 {
 		return false, nil
 	}
+	if err := stopEscalation(ctx, q, incidentID); err != nil {
+		return false, err
+	}
 	if err := logEvent(ctx, q, incidentID, evResolved, nil, nil, nil); err != nil {
 		return false, err
 	}
@@ -260,6 +272,10 @@ func acknowledgeIncident(ctx context.Context, q querier, incidentID, userID int6
 	if n, _ := res.RowsAffected(); n == 0 {
 		return false, nil
 	}
+	// Somebody has it: stop waking anybody else.
+	if err := stopEscalation(ctx, q, incidentID); err != nil {
+		return false, err
+	}
 	return true, logEvent(ctx, q, incidentID, evAcknowledged, &userID, nil, nil)
 }
 

internal/api/incidents.go

@@ -243,6 +243,11 @@ func handleIncidentResolve(db *sql.DB) http.HandlerFunc {
 			time.Now().Unix(), incidentResolutionManual, id) {
 			return
 		}
+		// A person closing an incident is the clearest possible "I have this".
+		if err := stopEscalation(r.Context(), db, id); err != nil {
+			respond(w, http.StatusInternalServerError, errResp("internal error"))
+			return
+		}
 		if err := logEvent(r.Context(), db, id, evResolved, &user.ID, nil, nil); err != nil {
 			respond(w, http.StatusInternalServerError, errResp("internal error"))
 			return

internal/api/notifier.go

@@ -43,6 +43,10 @@ const (
 	notifyTriggered = "triggered"
 	notifyReminder  = "reminder"
 	notifyResolved  = "resolved"
+
+	// notifyEscalated is a page that went out because nobody answered the last
+	// one. Told apart from a reminder because it goes to somebody else.
+	notifyEscalated = "escalated"
 )
 
 // Timeline event types the notifier writes, so an incident's history says who
@@ -120,6 +124,9 @@ func StartNotifier(ctx context.Context, db *sql.DB, cfg NotifyConfig) {
 // Exported so tests can drive a pass without waiting on the ticker.
 func NotifySweep(ctx context.Context, db *sql.DB, cfg NotifyConfig) {
 	enqueueReminders(ctx, db, cfg)
+	// Escalation before delivery, so a level that comes due on this tick is
+	// paged on this tick rather than waiting for the next one.
+	escalate(ctx, db, cfg)
 	deliverPending(ctx, db, cfg)
 }
 
@@ -159,7 +166,12 @@ func enqueueReminders(ctx context.Context, db *sql.DB, cfg NotifyConfig) {
 		  AND  i.resolved_at IS NULL
 		  AND  i.archived_at IS NULL
 		  AND  i.status = 'triggered'
-		  AND  (i.snoozed_until IS NULL OR i.snoozed_until <= $2)`,
+		  AND  (i.snoozed_until IS NULL OR i.snoozed_until <= $2)
+		  -- A team with an escalation ladder gets escalation instead. Both
+		  -- would mean two pages for one silence, which is how people learn to
+		  -- mute a tool.
+		  AND  NOT EXISTS (
+		           SELECT 1 FROM escalation_levels el WHERE el.team_id = i.team_id)`,
 		now.Add(-repeat).Unix(), now.Unix())
 	if err != nil {
 		log.Printf("notifier: find reminders: %v", err)

internal/api/notify_test.go

@@ -69,6 +69,27 @@ func (f *fakeNtfy) messages() []pushed {
 	return append([]pushed(nil), f.got...)
 }
 
+// topicsSince lists the topics published to since the last forget, which is how
+// the escalation tests ask "who did this tick wake".
+func (f *fakeNtfy) topicsSince(t *testing.T) []string {
+	t.Helper()
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	out := make([]string, 0, len(f.got))
+	for _, m := range f.got {
+		out = append(out, m.Topic)
+	}
+	return out
+}
+
+// forget drops what has been published so far, so the next assertion is about
+// this tick rather than the whole test.
+func (f *fakeNtfy) forget() {
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	f.got = nil
+}
+
 func (f *fakeNtfy) failWith(status int) {
 	f.mu.Lock()
 	defer f.mu.Unlock()

internal/api/router.go

@@ -109,6 +109,10 @@ func NewRouter(db *sql.DB, notify NotifyConfig, cfg config.Config) http.Handler
 		r.Post("/api/teams/{teamID}/members", handleAddTeamMember(db))
 		r.Delete("/api/teams/{teamID}/members/{userID}", handleRemoveTeamMember(db))
 
+		// A team's escalation ladder: who is paged when nobody answers.
+		r.Get("/api/teams/{teamID}/escalation", handleGetEscalation(db))
+		r.Put("/api/teams/{teamID}/escalation", handleSetEscalation(db))
+
 		// A team's own dead man's switches: which of its alerts are heartbeats,
 		// and how long a silence has to last before somebody is paged.
 		r.Get("/api/teams/{teamID}/deadman", handleGetTeamDeadman(db))

internal/db/migrations/006_escalation.sql

@@ -0,0 +1,95 @@
+-- Escalation: page somebody else when the first person does not answer.
+--
+-- This is the gap the whole multi-tenancy line of work was opened to close.
+-- Until now an unacknowledged incident re-paged the same topic every
+-- notify_repeat forever, which is a louder version of the same silence: if the
+-- person on call is asleep, has no signal, or has left, nothing else happens.
+--
+-- Shape: one policy per team, an ordered list of levels, each level with a
+-- timeout and a set of targets. When a level's timeout passes and the incident
+-- is still triggered, the next level is paged. When the last level passes, the
+-- chain repeats repeat_count times, and then the team's fallback topic is paged
+-- once as the end of the line.
+--
+-- A team WITHOUT a policy keeps exactly today's behaviour: page the assignee,
+-- then remind on the same topic. Escalation is opt-in per team, and the two
+-- never both run for one incident -- see enqueueReminders.
+CREATE TABLE escalation_policies (
+    -- One per team for now, hence the team as the key rather than an id with a
+    -- unique index: routing different alerts to different chains needs the
+    -- alert to carry something to route ON, which is a separate question.
+    team_id BIGINT PRIMARY KEY REFERENCES teams(id) ON DELETE CASCADE,
+
+    -- How many extra times to run the whole chain after it has been walked
+    -- once. 0 means walk it once and stop at the fallback.
+    repeat_count BIGINT NOT NULL DEFAULT 0 CHECK (repeat_count >= 0 AND repeat_count <= 10),
+
+    -- Where the last page goes when every level has been tried. Per team now:
+    -- TERDUT_NTFY_FALLBACK_TOPIC was one topic for the whole install, which in
+    -- a multi-team server pages the wrong people. Empty means the chain simply
+    -- ends.
+    fallback_topic TEXT NOT NULL DEFAULT '',
+
+    updated_at BIGINT NOT NULL DEFAULT FLOOR(EXTRACT(EPOCH FROM now()))::bigint
+);
+
+CREATE TABLE escalation_levels (
+    id              BIGINT GENERATED BY DEFAULT AS IDENTITY PRIMARY KEY,
+    team_id         BIGINT NOT NULL REFERENCES escalation_policies(team_id) ON DELETE CASCADE,
+    -- 1-based, dense. The API rewrites the whole ladder on every edit rather
+    -- than patching one rung, so there is no way to leave a gap.
+    position        BIGINT NOT NULL,
+    -- How long this level has to produce an acknowledgement before the next one
+    -- is paged. Seconds, like every other duration in this schema.
+    timeout_seconds BIGINT NOT NULL CHECK (timeout_seconds > 0),
+
+    UNIQUE (team_id, position)
+);
+
+-- Who a level pages. Either a named person, or whoever the team's rota says is
+-- on call today -- which is the target that keeps working when the rota
+-- changes and nobody remembers to edit the policy.
+CREATE TABLE escalation_targets (
+    id       BIGINT GENERATED BY DEFAULT AS IDENTITY PRIMARY KEY,
+    level_id BIGINT NOT NULL REFERENCES escalation_levels(id) ON DELETE CASCADE,
+    kind     TEXT   NOT NULL CHECK (kind IN ('user', 'oncall')),
+    -- Set for kind='user', NULL for kind='oncall'.
+    user_id  BIGINT REFERENCES users(id) ON DELETE CASCADE,
+
+    CHECK ((kind = 'user' AND user_id IS NOT NULL) OR (kind = 'oncall' AND user_id IS NULL))
+);
+
+CREATE INDEX escalation_targets_level_idx ON escalation_targets(level_id);
+
+-- ---------------------------------------------------------------------------
+-- Where an incident is in its chain.
+--
+-- On the incident rather than in a side table: it is read on every notifier
+-- tick alongside the incident's status, and one row per incident is exactly
+-- what the state is.
+-- ---------------------------------------------------------------------------
+
+-- 0 means no level has been paged yet, which is the state of every incident
+-- that existed before escalation and of every incident in a team with no
+-- policy. 1 is the first level.
+ALTER TABLE incidents ADD COLUMN escalation_level BIGINT NOT NULL DEFAULT 0;
+
+-- When the current level was entered, and therefore what its timeout is
+-- measured from. NULL while escalation_level is 0.
+ALTER TABLE incidents ADD COLUMN escalation_level_at BIGINT;
+
+-- How many times the chain has been walked in full. Compared against the
+-- policy's repeat_count.
+ALTER TABLE incidents ADD COLUMN escalation_round BIGINT NOT NULL DEFAULT 0;
+
+-- The notifier's escalation query: incidents still waiting, oldest level first.
+CREATE INDEX incidents_escalation_idx
+    ON incidents(escalation_level_at)
+ WHERE resolved_at IS NULL AND status = 'triggered';
+
+-- 'escalated' joins the outbox kinds: a page that went out because nobody
+-- answered the last one, which is worth telling apart from the first page and
+-- from a reminder when reading the timeline or debugging a delivery.
+ALTER TABLE notifications DROP CONSTRAINT notifications_kind_check;
+ALTER TABLE notifications ADD CONSTRAINT notifications_kind_check
+    CHECK (kind IN ('triggered', 'reminder', 'resolved', 'escalated'));

internal/models/incident.go

@@ -11,6 +11,13 @@ import "time"
 // the webhook and the sweeper may flip to "resolved" once every member alert has
 // stopped firing.
 type Incident struct {
+	// EscalationLevel is which rung of its team's ladder this incident is on,
+	// 0 for none — either the team has no ladder, or somebody has answered.
+	// EscalationDueAt is when the current level runs out, so a client can say
+	// how long is left rather than only what already happened.
+	EscalationLevel int64      `json:"escalation_level"`
+	EscalationDueAt *time.Time `json:"escalation_due_at,omitempty"`
+
 	// TeamID is the team that owns this incident, fixed when it opens: an
 	// incident never moves between teams. TeamName rides along so the combined
 	// queue can badge each row without a second request.

internal/web/static/app.css

@@ -657,3 +657,33 @@ kbd {
 .admin-settings .setting-unit { max-width: 8em; }
 .admin-settings button[type="submit"] { margin-top: 12px; }
 .small { font-size: 13px; }
+
+/* --- team settings -------------------------------------------------------
+   Forms with a label above each control, rather than the queue's rows of
+   links. The escalation ladder is the only nested structure in the app, so it
+   gets a little indentation to make the levels read as an order. */
+.stacked-form { display: flex; flex-direction: column; gap: 10px; margin-top: 12px; align-items: flex-start; }
+.stacked-form label { display: flex; align-items: center; gap: 6px; flex-wrap: wrap; font-size: 14px; }
+.stacked-form label.checkbox { gap: 8px; }
+.stacked-form input.wide { min-width: min(420px, 100%); }
+.team-picker { margin-top: 8px; max-width: 100%; }
+
+.ladder-level {
+  border-left: 3px solid var(--border-strong);
+  padding: 8px 0 8px 12px; margin: 12px 0;
+}
+.ladder-head { display: flex; align-items: center; gap: 10px; margin-bottom: 6px; }
+.ladder-targets { display: flex; flex-direction: column; gap: 6px; margin-top: 8px; }
+.target-row { display: flex; gap: 6px; align-items: center; flex-wrap: wrap; }
+
+/* An integration key is shown exactly once, so it should look like something
+   to act on rather than another row of text. */
+.key-panel {
+  margin-top: 12px; padding: 12px;
+  border: 1px solid var(--accent); border-radius: 8px; background: var(--accent-soft);
+}
+.key-panel pre {
+  overflow-x: auto; background: var(--surface); border: 1px solid var(--border);
+  border-radius: 6px; padding: 8px; font-size: 12px;
+}
+.key-url code { word-break: break-all; }

internal/web/static/index.html

@@ -59,6 +59,10 @@
         <svg viewBox="0 0 24 24" aria-hidden="true"><path d="M6 16V11a6 6 0 0 1 12 0v5l1.5 2h-15z"/><path d="M10 20.5a2 2 0 0 0 4 0"/></svg>
         <span class="nav-label">Alerts</span>
       </a>
+      <a class="nav-link" href="/team" data-section="team">
+        <svg viewBox="0 0 24 24" aria-hidden="true"><circle cx="9" cy="8" r="3"/><circle cx="17" cy="9" r="2.5"/><path d="M3 19a6 6 0 0 1 12 0M15 19a5 5 0 0 1 6-4"/></svg>
+        <span class="nav-label">Team</span>
+      </a>
       <!-- Hidden unless the signed-in user is a system administrator; app.js
            unhides it once /api/me says so. The server refuses every admin
            endpoint regardless, so this is a courtesy and not a gate. -->
@@ -87,6 +91,7 @@
 
     <section id="view-oncall" class="view view-page" data-view="oncall" hidden></section>
     <section id="view-alerts" class="view view-page" data-view="alerts" hidden></section>
+    <section id="view-team" class="view view-page" data-view="team" hidden></section>
     <section id="view-admin" class="view view-page" data-view="admin" hidden></section>
     <section id="view-more" class="view view-page" data-view="more" hidden></section>
   </div>

internal/web/static/js/api.js

@@ -92,6 +92,30 @@ export const createTeam = (name) => call('POST', '/teams', { body: { name } });
 export const renameTeam = (id, name) => call('PUT', `/teams/${id}`, { body: { name } });
 export const deleteTeam = (id) => call('DELETE', `/teams/${id}`);
 
+// A team's own settings. Every write is owner-only and every read is
+// member-only; the server answers 403 and 404 respectively, so the UI shows
+// what the role allows rather than guarding it.
+export const teamMembers = (id) => call('GET', `/teams/${id}/members`);
+export const addTeamMember = (id, userID, role) =>
+  call('POST', `/teams/${id}/members`, { body: { user_id: userID, role } });
+export const removeTeamMember = (id, userID) => call('DELETE', `/teams/${id}/members/${userID}`);
+
+export const integrations = (id) => call('GET', `/teams/${id}/integrations`);
+export const createIntegration = (id, name) =>
+  call('POST', `/teams/${id}/integrations`, { body: { name } });
+export const deleteIntegration = (id, integrationID) =>
+  call('DELETE', `/teams/${id}/integrations/${integrationID}`);
+
+export const deadman = (id) => call('GET', `/teams/${id}/deadman`);
+export const setDeadman = (id, body) => call('PUT', `/teams/${id}/deadman`, { body });
+
+export const escalation = (id) => call('GET', `/teams/${id}/escalation`);
+export const setEscalation = (id, body) => call('PUT', `/teams/${id}/escalation`, { body });
+
+export const assignSchedule = (id, userID, dates, replace = false) =>
+  call('POST', `/teams/${id}/schedule`, { body: { user_id: userID, dates, replace } });
+export const unassignSchedule = (id, entryID) => call('DELETE', `/teams/${id}/schedule/${entryID}`);
+
 // Administration. Every one of these is refused with 403 for anybody without
 // the flag, so the UI hides the section rather than guarding it.
 export const adminTeams = () => call('GET', '/admin/teams');

internal/web/static/js/app.js

@@ -9,6 +9,7 @@ import * as incident from './incident.js';
 import * as oncall from './oncall.js';
 import * as alerts from './alerts.js';
 import * as account from './account.js';
+import * as team from './team.js';
 import * as admin from './admin.js';
 
 const $ = (id) => document.getElementById(id);
@@ -18,6 +19,7 @@ const SECTIONS = {
   queue: { title: 'Queue', view: queue },
   oncall: { title: 'On-call', view: oncall },
   alerts: { title: 'Alerts', view: alerts },
+  team: { title: 'Team', view: team },
   admin: { title: 'Admin', view: admin },
   more: { title: 'Account', view: account },
 };
@@ -26,7 +28,7 @@ function parseRoute(pathname) {
   const m = pathname.match(/^\/incidents\/(\d+)\/?$/);
   if (m) return { section: 'queue', incident: Number(m[1]) };
   const name = pathname.replace(/^\/|\/$/g, '');
-  if (name === 'oncall' || name === 'alerts' || name === 'admin' || name === 'more') return { section: name };
+  if (name === 'oncall' || name === 'alerts' || name === 'team' || name === 'admin' || name === 'more') return { section: name };
   return { section: 'queue', incident: null };
 }
 

internal/web/static/js/incident.js

@@ -95,6 +95,15 @@ function statusBadges() {
   if (inc.status !== 'resolved' && isFuture(inc.snoozed_until)) {
     out.push(badge(`Snoozed · ${until(inc.snoozed_until)} left`, 'st-snoozed'));
   }
+  // Where it is on the ladder, while it is still climbing. The queue shows
+  // what happened; this says what happens next, which is the question somebody
+  // looking at an unacknowledged incident actually has.
+  if (inc.escalation_level > 0) {
+    const left = inc.escalation_due_at && isFuture(inc.escalation_due_at)
+      ? ` · next in ${until(inc.escalation_due_at)}`
+      : ' · next page due';
+    out.push(badge(`Escalating · level ${inc.escalation_level}${left}`, 'st-triggered'));
+  }
   if (inc.archived_at) out.push(badge('Archived', 'plain'));
   return out;
 }
@@ -115,6 +124,10 @@ function facts() {
   if (inc.status !== 'resolved' && isFuture(inc.snoozed_until)) {
     add('Snoozed until', when(inc.snoozed_until));
   }
+  if (inc.escalation_level > 0 && inc.escalation_due_at) {
+    add('Escalates next', when(inc.escalation_due_at),
+      h('span', { class: 'sub', text: ` · level ${inc.escalation_level}` }));
+  }
   if (inc.resolved_at) {
     const how = inc.resolution_source === 'manual' ? 'by hand' : 'alerts stopped firing';
     add('Resolved', when(inc.resolved_at), h('span', { class: 'sub', text: ` · ${how}` }));

internal/web/static/js/team.js

@@ -0,0 +1,453 @@
+// Team settings: the rota, who is in the team, where its alerts come from,
+// what it escalates through, and which of its alerts are heartbeats.
+//
+// Everything here was API-only until now, which meant a team owner had to use
+// curl to set up escalation — the feature this whole line of work exists for.
+//
+// The server decides what a role may do: an owner's edits succeed, a member's
+// are refused with 403, and a non-member gets 404 for the lot. This view hides
+// the controls a member cannot use, because a form that always fails is worse
+// than no form, but it is not the thing enforcing anything.
+
+import * as api from './api.js';
+import { h, clear, spinner, confirm } from './ui.js';
+import { state, currentTeam, users as allUsers } from './state.js';
+import { isoDate, addDays } from './format.js';
+
+const view = () => document.getElementById('view-team');
+
+let teamID = null;
+let data = null; // { team, members, integrations, escalation, deadman, schedule, users }
+let error = null;
+let freshKey = null; // an integration key, shown once, until the view is left
+
+export function show() {
+  if (!data) clear(view(), spinner());
+  refresh();
+}
+
+function selectedTeam() {
+  const teams = state.teams || [];
+  return teams.find((t) => t.id === teamID) || currentTeam();
+}
+
+export async function refresh() {
+  const team = selectedTeam();
+  if (!team) {
+    data = null;
+    render();
+    return;
+  }
+  teamID = team.id;
+  try {
+    // A member may read all of this; only the writes are owner-only.
+    const [members, integrations, escalation, deadman, schedule, users] = await Promise.all([
+      api.teamMembers(team.id),
+      api.integrations(team.id),
+      api.escalation(team.id),
+      api.deadman(team.id),
+      api.schedule(team.id, isoDate(new Date()), isoDate(addDays(new Date(), 30))),
+      allUsers(),
+    ]);
+    data = { team, members, integrations, escalation, deadman, schedule, users };
+    error = null;
+  } catch (err) {
+    error = err.message;
+  }
+  render();
+}
+
+function isOwner() {
+  return data?.team?.role === 'owner' || state.me?.user?.is_admin;
+}
+
+function render() {
+  if (!data) {
+    clear(view(), error
+      ? h('div', { class: 'load-error', text: error })
+      : h('div', { class: 'card' }, h('p', { class: 'muted', text: 'You are not in a team yet.' })));
+    return;
+  }
+  clear(view(),
+    error && h('div', { class: 'load-error', text: `Showing older data: ${error}` }),
+    teamPicker(),
+    !isOwner() && h('div', { class: 'card' },
+      h('p', { class: 'muted small', text: 'You are a member of this team. Only an owner can change its settings.' })),
+    scheduleCard(),
+    escalationCard(),
+    integrationsCard(),
+    deadmanCard(),
+    membersCard(),
+  );
+}
+
+// Only shown to somebody in more than one team, like the queue's filter chips.
+function teamPicker() {
+  if ((state.teams || []).length < 2) {
+    return h('div', { class: 'card' }, h('h2', { text: data.team.name }));
+  }
+  const select = h('select', { class: 'team-picker' },
+    ...state.teams.map((t) => h('option', {
+      value: String(t.id), text: t.name, selected: t.id === teamID,
+    })));
+  select.addEventListener('change', () => {
+    teamID = Number(select.value);
+    data = null;
+    freshKey = null;
+    show();
+  });
+  return h('div', { class: 'card' }, h('h2', { text: 'Team' }), select);
+}
+
+// --- schedule --------------------------------------------------------------
+
+// The rota is one person per UTC day. The on-call page shows it; this is where
+// it is set, which until now was the TUI's job and the TUI cannot do it any
+// more.
+function scheduleCard() {
+  const rows = (data.schedule || []).map((e) =>
+    h('tr', {},
+      h('td', { text: e.date }),
+      h('td', {}, h('strong', { text: e.username })),
+      h('td', {}, isOwner() && h('button', {
+        class: 'btn-sm danger', type: 'button', text: 'Clear',
+        onclick: () => act(() => api.unassignSchedule(teamID, e.id)),
+      })),
+    ));
+
+  return h('div', { class: 'card' },
+    h('h2', { text: 'On-call rota' }),
+    h('p', { class: 'muted small', text: 'One person per UTC day, for the next 30 days.' }),
+    rows.length
+      ? h('table', { class: 'admin-table' }, h('tbody', {}, rows))
+      : h('p', { class: 'muted', text: 'Nobody is scheduled.' }),
+    isOwner() && assignForm(),
+  );
+}
+
+function assignForm() {
+  const who = memberSelect();
+  const from = h('input', { type: 'date', required: true, value: isoDate(new Date()) });
+  const days = h('input', { type: 'number', min: '1', max: '31', value: '1', class: 'setting-value' });
+  const replace = h('input', { type: 'checkbox' });
+
+  const form = h('form', { class: 'stacked-form' },
+    h('label', {}, 'Who ', who),
+    h('label', {}, 'From ', from),
+    h('label', {}, 'Days ', days),
+    // Taking a day somebody else holds has to be asked for, the same rule the
+    // API enforces: a plain assignment that silently moved a shift would move
+    // who gets paged without telling either of them.
+    h('label', { class: 'checkbox' }, replace, ' Take days somebody else holds'),
+    h('button', { class: 'btn', type: 'submit', text: 'Assign' }));
+
+  form.addEventListener('submit', (e) => {
+    e.preventDefault();
+    const start = new Date(from.value + 'T00:00:00Z');
+    const dates = [];
+    for (let i = 0; i < Number(days.value || 1); i++) dates.push(isoDate(addDays(start, i)));
+    act(() => api.assignSchedule(teamID, Number(who.value), dates, replace.checked));
+  });
+  return form;
+}
+
+function memberSelect(selected) {
+  return h('select', {},
+    ...(data.members || []).map((m) => h('option', {
+      value: String(m.user_id), text: m.username, selected: m.user_id === selected,
+    })));
+}
+
+// --- escalation ------------------------------------------------------------
+
+// The ladder is edited as a whole and sent as a whole, because the API replaces
+// it wholesale: the levels are an order, and patching one rung would leave the
+// numbering of the others undecided.
+let draft = null;
+
+function escalationCard() {
+  const esc = data.escalation;
+  if (!draft) {
+    draft = {
+      repeat_count: esc.repeat_count || 0,
+      fallback_topic: esc.fallback_topic || '',
+      levels: (esc.levels || []).map((l) => ({
+        timeout_seconds: l.timeout_seconds,
+        targets: (l.targets || []).map((t) => ({ kind: t.kind, user_id: t.user_id })),
+      })),
+    };
+  }
+
+  const body = [];
+  if (!draft.levels.length) {
+    body.push(h('p', { class: 'muted' },
+      'No ladder. An unacknowledged incident re-pages the same person every ',
+      'reminder interval and nobody else is woken.'));
+  }
+
+  draft.levels.forEach((level, i) => {
+    body.push(h('div', { class: 'ladder-level' },
+      h('div', { class: 'ladder-head' },
+        h('strong', { text: `Level ${i + 1}` }),
+        isOwner() && h('button', {
+          class: 'btn-sm danger', type: 'button', text: 'Remove',
+          onclick: () => { draft.levels.splice(i, 1); render(); },
+        })),
+      h('label', {}, 'Wait ', minutesInput(level.timeout_seconds, (secs) => {
+        level.timeout_seconds = secs;
+      }), ' before the next level'),
+      h('div', { class: 'ladder-targets' },
+        ...level.targets.map((t, ti) => targetRow(level, t, ti)),
+        isOwner() && h('button', {
+          class: 'btn-sm', type: 'button', text: '+ target',
+          onclick: () => { level.targets.push({ kind: 'oncall' }); render(); },
+        })),
+    ));
+  });
+
+  if (isOwner()) {
+    body.push(h('button', {
+      class: 'btn-sm', type: 'button', text: '+ level',
+      onclick: () => {
+        draft.levels.push({ timeout_seconds: 300, targets: [{ kind: 'oncall' }] });
+        render();
+      },
+    }));
+
+    const repeat = h('input', {
+      type: 'number', min: '0', max: '10', class: 'setting-value',
+      value: String(draft.repeat_count),
+      oninput: (e) => { draft.repeat_count = Number(e.target.value); },
+    });
+    const fallback = h('input', {
+      type: 'text', value: draft.fallback_topic, placeholder: 'terdut-oncall-all',
+      oninput: (e) => { draft.fallback_topic = e.target.value; },
+    });
+    body.push(h('label', {}, 'Repeat the whole ladder ', repeat, ' more times'));
+    body.push(h('label', {}, 'Then page this ntfy topic once ', fallback));
+    body.push(h('button', {
+      class: 'btn', type: 'button', text: 'Save ladder',
+      onclick: () => act(() => api.setEscalation(teamID, draft), { resetDraft: true }),
+    }));
+  }
+
+  return h('div', { class: 'card' },
+    h('h2', { text: 'Escalation' }),
+    h('p', { class: 'muted small' },
+      'When a level’s wait passes and nobody has acknowledged, the next level is ',
+      'paged. Acknowledging or resolving stops it; snoozing pauses it.'),
+    ...body,
+  );
+}
+
+function targetRow(level, target, index) {
+  const kind = h('select', {},
+    h('option', { value: 'oncall', text: 'Whoever is on call', selected: target.kind === 'oncall' }),
+    h('option', { value: 'user', text: 'A specific person', selected: target.kind === 'user' }));
+  kind.addEventListener('change', () => {
+    target.kind = kind.value;
+    target.user_id = kind.value === 'user' ? (data.members[0] || {}).user_id : undefined;
+    render();
+  });
+
+  const who = target.kind === 'user'
+    ? memberSelect(target.user_id)
+    : null;
+  if (who) {
+    who.addEventListener('change', () => { target.user_id = Number(who.value); });
+  }
+
+  return h('div', { class: 'target-row' }, kind, who,
+    isOwner() && h('button', {
+      class: 'btn-sm danger', type: 'button', text: '×',
+      title: 'Remove this target',
+      onclick: () => { level.targets.splice(index, 1); render(); },
+    }));
+}
+
+function minutesInput(seconds, onChange) {
+  const input = h('input', {
+    type: 'number', min: '1', class: 'setting-value',
+    value: String(Math.max(1, Math.round(seconds / 60))),
+    oninput: (e) => onChange(Number(e.target.value) * 60),
+  });
+  return h('span', {}, input, ' minutes');
+}
+
+// --- integrations ----------------------------------------------------------
+
+function integrationsCard() {
+  const rows = (data.integrations || []).map((i) =>
+    h('tr', {},
+      h('td', {}, h('strong', { text: i.name })),
+      h('td', { class: 'muted small', text: i.kind }),
+      h('td', { class: 'muted small', text: i.last_used_at ? 'in use' : 'never used' }),
+      h('td', {}, isOwner() && h('button', {
+        class: 'btn-sm danger', type: 'button', text: 'Revoke',
+        onclick: async () => {
+          if (!(await confirm({
+            title: `Revoke ${i.name}?`,
+            text: 'Anything posting with this key stops delivering immediately.',
+            confirmLabel: 'Revoke',
+            danger: true,
+          }))) return;
+          act(() => api.deleteIntegration(teamID, i.id));
+        },
+      })),
+    ));
+
+  return h('div', { class: 'card' },
+    h('h2', { text: 'Alert sources' }),
+    h('p', { class: 'muted small' },
+      'Alerts arrive on an integration key, which says both that the sender may ',
+      'post and which team the alerts belong to.'),
+    rows.length
+      ? h('table', { class: 'admin-table' }, h('tbody', {}, rows))
+      : h('p', { class: 'muted', text: 'No alert source yet, so nothing can reach this team.' }),
+    freshKey && newKeyPanel(),
+    isOwner() && !freshKey && newIntegrationForm(),
+  );
+}
+
+// The key is returned exactly once. Say so, show it large, and give the
+// Alertmanager snippet with it already in place — the next thing anybody does
+// with it is paste it into a config.
+function newKeyPanel() {
+  const url = freshKey.url || `${location.origin}/api/integrations/${freshKey.key}/alertmanager`;
+  const snippet = `receivers:
+  - name: terdut
+    webhook_configs:
+      - url: ${url}
+        send_resolved: true`;
+
+  return h('div', { class: 'key-panel' },
+    h('strong', { text: 'Copy this now — it is not shown again.' }),
+    h('pre', { class: 'key-url' }, h('code', { text: url })),
+    h('button', {
+      class: 'btn-sm', type: 'button', text: 'Copy URL',
+      onclick: () => navigator.clipboard?.writeText(url),
+    }),
+    h('p', { class: 'muted small', text: 'Alertmanager receiver:' }),
+    h('pre', {}, h('code', { text: snippet })),
+    h('button', {
+      class: 'btn-sm', type: 'button', text: 'Done',
+      onclick: () => { freshKey = null; render(); },
+    }),
+  );
+}
+
+function newIntegrationForm() {
+  const name = h('input', { type: 'text', placeholder: 'prod alertmanager', required: true });
+  const form = h('form', { class: 'inline-form' }, name,
+    h('button', { class: 'btn', type: 'submit', text: 'Add' }));
+  form.addEventListener('submit', async (e) => {
+    e.preventDefault();
+    try {
+      freshKey = await api.createIntegration(teamID, name.value.trim());
+      await refresh();
+    } catch (err) {
+      error = err.message;
+      render();
+    }
+  });
+  return form;
+}
+
+// --- dead man's switches ---------------------------------------------------
+
+function deadmanCard() {
+  const d = data.deadman || {};
+  const matchers = h('input', {
+    type: 'text', value: d.matchers || '', placeholder: 'alertname=Watchdog',
+    class: 'wide',
+  });
+  const timeout = h('input', {
+    type: 'number', min: '0', class: 'setting-value',
+    value: String(Math.round((d.timeout_seconds || 0) / 60)),
+  });
+  const severity = h('select', {},
+    ...['critical', 'error', 'warning', 'info'].map((s) =>
+      h('option', { value: s, text: s, selected: (d.severity || 'critical') === s })));
+
+  const form = h('form', { class: 'stacked-form' },
+    h('label', {}, 'Heartbeat alerts ', matchers),
+    h('label', {}, 'Declare dead after ', timeout, ' minutes of silence'),
+    h('label', {}, 'Open the incident at severity ', severity),
+    h('button', { class: 'btn', type: 'submit', text: 'Save switches' }));
+
+  form.addEventListener('submit', (e) => {
+    e.preventDefault();
+    act(() => api.setDeadman(teamID, {
+      matchers: matchers.value.trim(),
+      timeout_seconds: Number(timeout.value) * 60,
+      severity: severity.value,
+    }));
+  });
+
+  return h('div', { class: 'card' },
+    h('h2', { text: 'Dead man’s switches' }),
+    h('p', { class: 'muted small' },
+      'Alerts whose ABSENCE is the signal. Receiving one opens nothing; going ',
+      'quiet for longer than the timeout opens an incident. ',
+      h('code', { text: 'alertname=Watchdog,cluster=prod; alertname=EdgeHeartbeat' }),
+      ' — semicolons separate switches, commas separate conditions, and every ',
+      'switch must name an alertname. Leave empty to watch nothing.'),
+    isOwner() ? form : h('p', { class: 'muted', text: d.matchers || 'Nothing watched.' }),
+  );
+}
+
+// --- members ---------------------------------------------------------------
+
+function membersCard() {
+  const rows = (data.members || []).map((m) =>
+    h('tr', {},
+      h('td', {}, h('strong', { text: m.username })),
+      h('td', { class: 'muted small', text: m.role }),
+      h('td', {}, isOwner() && h('button', {
+        class: 'btn-sm', type: 'button',
+        text: m.role === 'owner' ? 'Make member' : 'Make owner',
+        onclick: () => act(() =>
+          api.addTeamMember(teamID, m.user_id, m.role === 'owner' ? 'member' : 'owner')),
+      }), isOwner() && h('button', {
+        class: 'btn-sm danger', type: 'button', text: 'Remove',
+        onclick: () => act(() => api.removeTeamMember(teamID, m.user_id)),
+      })),
+    ));
+
+  const inTeam = new Set((data.members || []).map((m) => m.user_id));
+  const candidates = (data.users || []).filter((u) => !inTeam.has(u.id) && !u.disabled_at);
+  const pick = h('select', {},
+    ...candidates.map((u) => h('option', { value: String(u.id), text: u.username })));
+  const role = h('select', {},
+    h('option', { value: 'member', text: 'member' }),
+    h('option', { value: 'owner', text: 'owner' }));
+  const form = h('form', { class: 'inline-form' }, pick, role,
+    h('button', { class: 'btn', type: 'submit', text: 'Add' }));
+  form.addEventListener('submit', (e) => {
+    e.preventDefault();
+    act(() => api.addTeamMember(teamID, Number(pick.value), role.value));
+  });
+
+  return h('div', { class: 'card' },
+    h('h2', { text: 'Members' }),
+    h('table', { class: 'admin-table' }, h('tbody', {}, rows)),
+    isOwner() && candidates.length > 0 && form,
+  );
+}
+
+// --- plumbing --------------------------------------------------------------
+
+// act runs a write and reloads. Errors are shown rather than thrown away: a
+// 409 from the last-owner guard or the schedule's conflict rule is the server
+// explaining itself, and the reader needs to see it.
+async function act(fn, { resetDraft = false } = {}) {
+  try {
+    await fn();
+    error = null;
+    if (resetDraft) draft = null;
+  } catch (err) {
+    error = err.message;
+  }
+  if (!resetDraft) draft = null;
+  await refresh();
+}