Merge remote-tracking branch 'origin/main' into embedded-vnc

# Conflicts:
#	client/cmd/up.go
#	client/internal/engine.go
#	client/internal/profilemanager/config.go
#	client/proto/daemon.pb.go
#	client/server/mdm.go
#	client/server/server.go
#	client/status/status.go
#	client/ui/main.go
#	client/ui/services/windowmanager.go
#	management/internals/shared/grpc/conversion.go
#	management/server/store/sql_store.go
#	management/server/store/sql_store_test.go
#	shared/management/networkmap/envelope.go
#	shared/management/proto/management.pb.go

client/internal/auth/account_match_test.go

@@ -0,0 +1,113 @@
+package auth
+
+import (
+	"encoding/base64"
+	"encoding/json"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestTokenInfoMatchesAccount(t *testing.T) {
+	tests := []struct {
+		name  string
+		token TokenInfo
+		hint  string
+		match bool
+	}{
+		{
+			name:  "same account",
+			token: TokenInfo{EmailClaim: "user@example.com"},
+			hint:  "user@example.com",
+			match: true,
+		},
+		{
+			name:  "different account",
+			token: TokenInfo{EmailClaim: "other@example.com"},
+			hint:  "user@example.com",
+			match: false,
+		},
+		{
+			name:  "case differences are the same account",
+			token: TokenInfo{EmailClaim: "User@Example.com"},
+			hint:  "user@example.com",
+			match: true,
+		},
+		{
+			name:  "no hint leaves the choice to the IdP",
+			token: TokenInfo{EmailClaim: "other@example.com"},
+			hint:  "",
+			match: true,
+		},
+		{
+			name:  "token without an email claim is not judged",
+			token: TokenInfo{EmailClaim: ""},
+			hint:  "user@example.com",
+			match: true,
+		},
+		{
+			name:  "name fallback does not trigger matching",
+			token: TokenInfo{Email: "Some One"},
+			hint:  "user@example.com",
+			match: true,
+		},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			assert.Equal(t, tc.match, tc.token.MatchesAccount(tc.hint))
+		})
+	}
+}
+
+func TestParseEmailFromIDToken(t *testing.T) {
+	tests := []struct {
+		name          string
+		claims        map[string]interface{}
+		wantValue     string
+		wantFromEmail bool
+		wantErr       bool
+	}{
+		{
+			name:          "email claim",
+			claims:        map[string]interface{}{"email": "user@example.com", "name": "Some One"},
+			wantValue:     "user@example.com",
+			wantFromEmail: true,
+		},
+		{
+			name:      "name fallback",
+			claims:    map[string]interface{}{"name": "Some One"},
+			wantValue: "Some One",
+		},
+		{
+			name:    "neither claim present",
+			claims:  map[string]interface{}{"sub": "abc"},
+			wantErr: true,
+		},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			value, fromEmailClaim, err := parseEmailFromIDToken(idTokenWithClaims(t, tc.claims))
+			if tc.wantErr {
+				require.Error(t, err)
+				return
+			}
+			require.NoError(t, err)
+			assert.Equal(t, tc.wantValue, value)
+			assert.Equal(t, tc.wantFromEmail, fromEmailClaim)
+		})
+	}
+}
+
+func TestRetryFlowForAccountUnsupportedFlow(t *testing.T) {
+	assert.Nil(t, RetryFlowForAccount(&DeviceAuthorizationFlow{}))
+}
+
+func idTokenWithClaims(t *testing.T, claims map[string]interface{}) string {
+	t.Helper()
+	payload, err := json.Marshal(claims)
+	require.NoError(t, err)
+	return "header." + base64.RawURLEncoding.EncodeToString(payload) + ".signature"
+}

client/internal/auth/auth.go

@@ -103,7 +103,7 @@ func (a *Auth) IsSSOSupported(ctx context.Context) (bool, error) {
 
 	err := a.withRetry(ctx, func(client *mgm.GrpcClient) error {
 		// Try PKCE flow first
-		_, err := a.getPKCEFlow(client)
+		_, err := a.getPKCEFlow(client, false)
 		if err == nil {
 			supportsSSO = true
 			return nil
@@ -136,9 +136,13 @@ func (a *Auth) IsSSOSupported(ctx context.Context) (bool, error) {
 	return supportsSSO, err
 }
 
-// GetOAuthFlow returns an OAuth flow (PKCE or Device) using the existing management connection
-// This avoids creating a new connection to the management server
-func (a *Auth) GetOAuthFlow(ctx context.Context, forceDeviceAuth bool, hint string) (OAuthFlow, error) {
+// GetOAuthFlow returns an OAuth flow (PKCE or Device) using the existing management connection.
+// This avoids creating a new connection to the management server.
+//
+// sessionExtend marks the flow as renewing an existing peer's session rather than
+// logging one in; the server needs it to rule out a silent authorization that the
+// IdP could answer from another account. See PKCEAuthorizationFlowRequest.
+func (a *Auth) GetOAuthFlow(ctx context.Context, forceDeviceAuth bool, sessionExtend bool, hint string) (OAuthFlow, error) {
 	var flow OAuthFlow
 
 	err := a.withRetry(ctx, func(client *mgm.GrpcClient) error {
@@ -153,7 +157,7 @@ func (a *Auth) GetOAuthFlow(ctx context.Context, forceDeviceAuth bool, hint stri
 		}
 
 		// Try PKCE flow first
-		pkceFlow, err := a.getPKCEFlow(client)
+		pkceFlow, err := a.getPKCEFlow(client, sessionExtend)
 		if err != nil {
 			// If PKCE not supported, try Device flow
 			if s, ok := status.FromError(err); ok && (s.Code() == codes.NotFound || s.Code() == codes.Unimplemented) {
@@ -240,8 +244,8 @@ func (a *Auth) Login(ctx context.Context, setupKey string, jwtToken string) (err
 }
 
 // getPKCEFlow retrieves PKCE authorization flow configuration and creates a flow instance
-func (a *Auth) getPKCEFlow(client *mgm.GrpcClient) (*PKCEAuthorizationFlow, error) {
-	protoFlow, err := client.GetPKCEAuthorizationFlow()
+func (a *Auth) getPKCEFlow(client *mgm.GrpcClient, sessionExtend bool) (*PKCEAuthorizationFlow, error) {
+	protoFlow, err := client.GetPKCEAuthorizationFlow(sessionExtend)
 	if err != nil {
 		if s, ok := status.FromError(err); ok && s.Code() == codes.NotFound {
 			log.Warnf("server couldn't find pkce flow, contact admin: %v", err)

client/internal/auth/device_flow.go

@@ -308,10 +308,13 @@ func (d *DeviceAuthorizationFlow) WaitToken(ctx context.Context, info AuthFlowIn
 			// callers store to send back as the login_hint. Without it a client
 			// driven through the device flow — Android TV and tvOS — never binds
 			// an account to its profile and every later login goes out blind.
-			if email, err := parseEmailFromIDToken(tokenInfo.IDToken); err != nil {
+			if email, fromEmailClaim, err := parseEmailFromIDToken(tokenInfo.IDToken); err != nil {
 				log.Warnf("failed to parse email from ID token: %v", err)
 			} else {
 				tokenInfo.Email = email
+				if fromEmailClaim {
+					tokenInfo.EmailClaim = email
+				}
 			}
 
 			log.Infof("device flow: user authorization confirmed after %d polls in %s", polls, time.Since(start).Round(time.Second))

client/internal/auth/oauth.go

@@ -5,6 +5,7 @@ import (
 	"fmt"
 	"net/http"
 	"runtime"
+	"strings"
 
 	log "github.com/sirupsen/logrus"
 	"google.golang.org/grpc/codes"
@@ -25,6 +26,14 @@ type HTTPClient interface {
 	Do(req *http.Request) (*http.Response, error)
 }
 
+// accountPromptForcer is implemented by the PKCE flow only. The device code
+// flow has no equivalent: RFC 8628 defines no prompt parameter, and the user
+// confirms the code on a page that shows which account signs in, so a silent
+// wrong-account answer is not the failure mode there.
+type accountPromptForcer interface {
+	ForceAccountPrompt()
+}
+
 // AuthFlowInfo holds information for the OAuth 2.0  authorization flow
 type AuthFlowInfo struct { //nolint:revive
 	DeviceCode              string `json:"device_code"`
@@ -49,6 +58,23 @@ type TokenInfo struct {
 	ExpiresIn    int    `json:"expires_in"`
 	UseIDToken   bool   `json:"-"`
 	Email        string `json:"-"`
+	EmailClaim   string `json:"-"`
+}
+
+// MatchesAccount reports whether the token belongs to the account a profile is
+// bound to. A hint the IdP could not have acted on — no hint stored, or a token
+// that carried no email claim — is reported as a match: the check exists to catch a
+// login answered from the wrong account, not to block one it cannot judge.
+//
+// The comparison is case-insensitive. Local-parts are case-sensitive per RFC
+// 5321, but no IdP in practice issues two accounts differing only in case, and
+// an IdP that echoes a differently-cased address would otherwise fail every
+// login.
+func (t TokenInfo) MatchesAccount(hint string) bool {
+	if hint == "" || t.EmailClaim == "" {
+		return true
+	}
+	return strings.EqualFold(t.EmailClaim, hint)
 }
 
 // GetTokenToUse returns either the access or id token based on UseIDToken field
@@ -63,19 +89,22 @@ func shouldUseDeviceFlow(force bool, isUnixDesktopClient bool) bool {
 	return force || (runtime.GOOS == "linux" || runtime.GOOS == "freebsd") && !isUnixDesktopClient
 }
 
-// NewOAuthFlow initializes and returns the appropriate OAuth flow based on the management configuration
+// NewOAuthFlow initializes and returns the appropriate OAuth flow based on the management configuration.
 //
 // It starts by initializing the PKCE.If this process fails, it resorts to the Device Code Flow,
 // and if that also fails, the authentication process is deemed unsuccessful
 //
-// On Linux distros without desktop environment support, it only tries to initialize the Device Code Flow
-// forceDeviceCodeFlow can be used to skip PKCE and go directly to Device Code Flow (e.g., for Android TV)
-func NewOAuthFlow(ctx context.Context, config *profilemanager.Config, isUnixDesktopClient bool, forceDeviceCodeFlow bool, hint string) (OAuthFlow, error) {
+// On Linux distros without desktop environment support, it only tries to initialize the Device Code Flow.
+// forceDeviceCodeFlow can be used to skip PKCE and go directly to Device Code Flow (e.g., for Android TV).
+//
+// sessionExtend marks the flow as renewing an existing peer's session rather than
+// logging one in. See PKCEAuthorizationFlowRequest for what the server makes of it.
+func NewOAuthFlow(ctx context.Context, config *profilemanager.Config, isUnixDesktopClient bool, forceDeviceCodeFlow bool, hint string, sessionExtend bool) (OAuthFlow, error) {
 	if shouldUseDeviceFlow(forceDeviceCodeFlow, isUnixDesktopClient) {
 		return authenticateWithDeviceCodeFlow(ctx, config, hint)
 	}
 
-	pkceFlow, err := authenticateWithPKCEFlow(ctx, config, hint)
+	pkceFlow, err := authenticateWithPKCEFlow(ctx, config, hint, sessionExtend)
 	if err != nil {
 		log.Debugf("failed to initialize pkce authentication with error: %v\n", err)
 		log.Debug("falling back to device code flow")
@@ -85,14 +114,14 @@ func NewOAuthFlow(ctx context.Context, config *profilemanager.Config, isUnixDesk
 }
 
 // authenticateWithPKCEFlow initializes the Proof Key for Code Exchange flow auth flow
-func authenticateWithPKCEFlow(ctx context.Context, config *profilemanager.Config, hint string) (OAuthFlow, error) {
+func authenticateWithPKCEFlow(ctx context.Context, config *profilemanager.Config, hint string, sessionExtend bool) (OAuthFlow, error) {
 	authClient, err := NewAuth(ctx, config.PrivateKey, config.ManagementURL, config)
 	if err != nil {
 		return nil, fmt.Errorf("failed to create auth client: %v", err)
 	}
 	defer authClient.Close()
 
-	pkceFlowInfo, err := authClient.getPKCEFlow(authClient.client)
+	pkceFlowInfo, err := authClient.getPKCEFlow(authClient.client, sessionExtend)
 	if err != nil {
 		return nil, fmt.Errorf("getting pkce authorization flow info failed with error: %v", err)
 	}
@@ -129,3 +158,21 @@ func authenticateWithDeviceCodeFlow(ctx context.Context, config *profilemanager.
 
 	return deviceFlowInfo, nil
 }
+
+// RetryFlowForAccount returns a flow that asks the IdP to re-authenticate, for
+// a login answered with an account other than the one hinted. Returns nil when
+// the flow cannot ask — the caller then proceeds with the token it has.
+//
+// Proceeding rather than failing is deliberate. The hint is an email that may
+// simply have changed since it was stored, and refusing the login would lock a
+// user out of their own profile over a rename. The retry gives the account a
+// chance to be corrected; the server still rejects a token that does not own
+// the peer.
+func RetryFlowForAccount(flow OAuthFlow) OAuthFlow {
+	forcer, ok := flow.(accountPromptForcer)
+	if !ok {
+		return nil
+	}
+	forcer.ForceAccountPrompt()
+	return flow
+}

client/internal/auth/pkce_flow.go

@@ -87,10 +87,11 @@ func validatePKCEConfig(config *PKCEAuthProviderConfig) error {
 // PKCEAuthorizationFlow implements the OAuthFlow interface for
 // the Authorization Code Flow with PKCE.
 type PKCEAuthorizationFlow struct {
-	providerConfig PKCEAuthProviderConfig
-	state          string
-	codeVerifier   string
-	oAuthConfig    *oauth2.Config
+	providerConfig     PKCEAuthProviderConfig
+	state              string
+	codeVerifier       string
+	oAuthConfig        *oauth2.Config
+	forceAccountPrompt bool
 }
 
 // NewPKCEAuthorizationFlow returns new PKCE authorization code flow.
@@ -153,11 +154,16 @@ func (p *PKCEAuthorizationFlow) RequestAuthInfo(ctx context.Context) (AuthFlowIn
 		oauth2.SetAuthURLParam("code_challenge", codeChallenge),
 		oauth2.SetAuthURLParam("audience", p.providerConfig.Audience),
 	}
+	forceAccountPrompt := p.forceAccountPrompt
+	p.forceAccountPrompt = false
+
 	if !p.providerConfig.DisablePromptLogin {
-		switch p.providerConfig.LoginFlag {
-		case common.LoginFlagPromptLogin:
+		switch {
+		case forceAccountPrompt:
 			params = append(params, oauth2.SetAuthURLParam("prompt", "login"))
-		case common.LoginFlagMaxAge0:
+		case p.providerConfig.LoginFlag == common.LoginFlagPromptLogin:
+			params = append(params, oauth2.SetAuthURLParam("prompt", "login"))
+		case p.providerConfig.LoginFlag == common.LoginFlagMaxAge0:
 			params = append(params, oauth2.SetAuthURLParam("max_age", "0"))
 		}
 	}
@@ -178,6 +184,20 @@ func (p *PKCEAuthorizationFlow) SetLoginHint(hint string) {
 	p.providerConfig.LoginHint = hint
 }
 
+// ForceAccountPrompt makes the next authorization request ask the IdP to
+// re-authenticate instead of answering from the session it already holds. Used
+// to retry a login that came back for an account other than the one hinted.
+//
+// The next RequestAuthInfo consumes the flag, so a flow that outlives its retry
+// goes back to the configured behaviour instead of re-authenticating forever.
+//
+// DisablePromptLogin still wins: it is set for IdPs that break on prompt=login,
+// where retrying with it would replace a wrong-account login with one that
+// cannot complete at all.
+func (p *PKCEAuthorizationFlow) ForceAccountPrompt() {
+	p.forceAccountPrompt = true
+}
+
 // WaitToken waits for the OAuth token in the PKCE Authorization Flow.
 // It starts an HTTP server to receive the OAuth token callback and waits for the token or an error.
 // Once the token is received, it is converted to TokenInfo and validated before returning.
@@ -310,49 +330,48 @@ func (p *PKCEAuthorizationFlow) parseOAuthToken(token *oauth2.Token) (TokenInfo,
 		return TokenInfo{}, fmt.Errorf("authentication failed: invalid access token - %w", err)
 	}
 
-	email, err := parseEmailFromIDToken(tokenInfo.IDToken)
+	email, fromEmailClaim, err := parseEmailFromIDToken(tokenInfo.IDToken)
 	if err != nil {
 		log.Warnf("failed to parse email from ID token: %v", err)
 	} else {
 		tokenInfo.Email = email
+		if fromEmailClaim {
+			tokenInfo.EmailClaim = email
+		}
 	}
 
 	return tokenInfo, nil
 }
 
 // parseEmailFromIDToken extracts the email (or name) claim from an ID token
-// without verifying its signature. The value is best-effort and used only as a
-// UX convenience (login hint prefill and display); it never drives an
-// authorization decision. The authoritative identity is established server-side
-// from the signature-verified token.
-func parseEmailFromIDToken(token string) (string, error) {
+// without verifying its signature. The value is best-effort: it prefills the
+// login hint and is displayed. Account matching (see MatchesAccount) only uses
+// it when it came from the email claim, which fromEmailClaim reports. It never
+// grants anything — the authoritative identity is established server-side from
+// the signature-verified token.
+func parseEmailFromIDToken(token string) (value string, fromEmailClaim bool, err error) {
 	parts := strings.Split(token, ".")
 	if len(parts) < 2 {
-		return "", fmt.Errorf("invalid token format")
+		return "", false, fmt.Errorf("invalid token format")
 	}
 
 	data, err := base64.RawURLEncoding.DecodeString(parts[1])
 	if err != nil {
-		return "", fmt.Errorf("failed to decode payload: %w", err)
+		return "", false, fmt.Errorf("failed to decode payload: %w", err)
 	}
 	var claims map[string]interface{}
 	if err := json.Unmarshal(data, &claims); err != nil {
-		return "", fmt.Errorf("json unmarshal error: %w", err)
+		return "", false, fmt.Errorf("json unmarshal error: %w", err)
 	}
 
-	var email string
-	if emailValue, ok := claims["email"].(string); ok {
-		email = emailValue
-	} else {
-		val, ok := claims["name"].(string)
-		if ok {
-			email = val
-		} else {
-			return "", fmt.Errorf("email or name field not found in token payload")
-		}
+	if email, ok := claims["email"].(string); ok {
+		return email, true, nil
+	}
+	if name, ok := claims["name"].(string); ok {
+		return name, false, nil
 	}
 
-	return email, nil
+	return "", false, fmt.Errorf("email or name field not found in token payload")
 }
 
 func createCodeChallenge(codeVerifier string) string {

client/internal/auth/pkce_flow_test.go

@@ -76,6 +76,32 @@ func TestPromptLogin(t *testing.T) {
 	}
 }
 
+func TestForceAccountPromptAppliesOnlyToTheRetry(t *testing.T) {
+	config := PKCEAuthProviderConfig{
+		ClientID:              "test-client-id",
+		Audience:              "test-audience",
+		TokenEndpoint:         "https://test-token-endpoint.com/token",
+		Scope:                 "openid email profile",
+		AuthorizationEndpoint: "https://test-auth-endpoint.com/authorize",
+		RedirectURLs:          []string{"http://127.0.0.1:33992/"},
+		UseIDToken:            true,
+		LoginFlag:             mgm.LoginFlagNone,
+	}
+	pkce, err := NewPKCEAuthorizationFlow(config)
+	require.NoError(t, err)
+
+	pkce.ForceAccountPrompt()
+
+	retry, err := pkce.RequestAuthInfo(context.Background())
+	require.NoError(t, err)
+	require.Contains(t, retry.VerificationURIComplete, "prompt=login")
+
+	next, err := pkce.RequestAuthInfo(context.Background())
+	require.NoError(t, err)
+	require.NotContains(t, next.VerificationURIComplete, "prompt=login",
+		"the forced prompt outlived the retry it was armed for")
+}
+
 func TestIsPortInExcludedRange(t *testing.T) {
 	tests := []struct {
 		name            string

client/internal/auth/sessionwatch/watcher.go

@@ -4,12 +4,20 @@
 // T-WarningLead notification and a dismiss-gated T-FinalWarningLead
 // fallback dialog.
 //
+// The deadline is an absolute wall-clock instant, so the watcher compares
+// it against the wall clock on a ticker rather than arming a relative
+// timer for it. A relative timer runs on the monotonic clock, which does
+// not advance while a device is suspended: it fires once that much awake
+// time has passed, which can be long after the deadline, and nothing
+// re-evaluates when the device wakes up. Polling makes every tick after a
+// resume (or after an NTP correction) see the real remaining time.
+//
 // The watcher is idempotent: Update may be called as often as the network
 // map snapshots arrive. Repeating the same deadline is a no-op; a new
-// deadline reschedules the timers and arms a fresh warning cycle.
+// deadline starts a fresh warning cycle.
 //
-// Warning firing is edge-detected. Each unique deadline value fires each
-// warning callback at most once.
+// Warning firing is edge-detected. Each unique deadline value publishes
+// each warning at most once.
 package sessionwatch
 
 import (
@@ -28,9 +36,15 @@ const (
 
 	// maxDeadlineHorizon caps how far in the future an accepted deadline
 	// can sit. A timestamp beyond this is almost certainly a protocol
-	// glitch, and silently arming a 100-year timer would hide the bug.
+	// glitch, and silently tracking a 100-year deadline would hide the bug.
 	maxDeadlineHorizon = 10 * 365 * 24 * time.Hour
 
+	// defaultEvalInterval is how often the tracked deadline is compared
+	// against the wall clock. The leads are minutes, so a coarse tick costs
+	// nothing in accuracy and keeps the wakeup cheap on battery-powered
+	// devices.
+	defaultEvalInterval = 10 * time.Second
+
 	// WarningLead is how far before expiry the first (interactive)
 	// warning fires. Drives the T-10 OS notification with
 	// Extend/Dismiss actions.
@@ -92,18 +106,23 @@ type StatusRecorder interface {
 // fallback T-FinalWarningLead dialog (suppressed when the user dismissed
 // the first one for the same deadline). Safe for concurrent use.
 type Watcher struct {
-	lead      time.Duration
-	finalLead time.Duration
+	lead         time.Duration
+	finalLead    time.Duration
+	interval     time.Duration
+	deadlineOnly bool // record the deadline, leave the warnings to the caller
 
 	mu           sync.Mutex
 	current      time.Time
-	timer        *time.Timer
-	finalTimer   *time.Timer
-	firedAt      time.Time // deadline value the T-WarningLead callback last fired against
-	finalFiredAt time.Time // deadline value the T-FinalWarningLead callback last fired against
-	dismissedAt  time.Time // deadline value the user dismissed via Dismiss(); gates fireFinal
+	firedAt      time.Time // deadline value the T-WarningLead warning last published for
+	finalFiredAt time.Time // deadline value the T-FinalWarningLead warning last published for
+	dismissedAt  time.Time // deadline value the user dismissed via Dismiss(); gates the final warning
+	announcedAt  time.Time // deadline value the recorder has been told about; gates publishing
 	closed       bool
 	recorder     StatusRecorder
+	nowFn        func() time.Time
+	stop         chan struct{} // closed to stop the evaluation loop; nil while it is not running
+	done         chan struct{} // closed by the loop on its way out
+	wake         chan struct{} // buffered nudge asking the loop to evaluate before its next tick
 }
 
 // New returns a watcher with the package defaults WarningLead and
@@ -115,27 +134,40 @@ func New(recorder StatusRecorder) *Watcher {
 }
 
 // NewWithLeads returns a watcher with custom lead times. Useful for tests.
-// final must be strictly less than lead; otherwise both timers fire in the
-// wrong order or simultaneously and the UI flow breaks. A zero final lead
-// disables the final-warning timer entirely (see armTimerLocked) so a
-// millisecond-scale deadline doesn't flush both timers in one tick.
+// final must be strictly less than lead; otherwise the final warning takes
+// over the whole warning window and the interactive notification never
+// shows. A zero final lead disables the final warning entirely (see
+// evaluate), leaving the interactive one as the only warning.
 func NewWithLeads(lead, final time.Duration, recorder StatusRecorder) *Watcher {
 	return &Watcher{
 		lead:      lead,
 		finalLead: final,
+		interval:  defaultEvalInterval,
 		recorder:  recorder,
+		nowFn:     time.Now,
 	}
 }
 
+// NewDeadlineOnly returns a watcher that validates and records deadlines
+// but publishes no warnings about them, and runs no evaluation loop to
+// decide. Used where the deadline is handed on to something that schedules
+// the warnings itself, such as the Android app.
+func NewDeadlineOnly(recorder StatusRecorder) *Watcher {
+	w := New(recorder)
+	w.deadlineOnly = true
+	return w
+}
+
 // Update sets the latest deadline. Pass the zero time to clear (e.g. when
 // a Sync push from the server omits the field because login expiration
 // was disabled).
 //
-// Same-value updates are no-ops. A different non-zero value cancels any
-// pending timer, resets the "already fired" guards, and — when the
-// deadline lies in the future — arms fresh warning timers. A deadline
-// already in the past (within maxPastHorizon) is recorded as-is with no
-// timers: the session has expired and consumers render it that way.
+// Same-value updates are no-ops. A different non-zero value resets the
+// "already fired" guards and starts a fresh warning cycle, evaluated
+// immediately so a deadline that already sits inside a warning window
+// warns without waiting for the next tick. A deadline already in the past
+// (within maxPastHorizon) is recorded as-is and warns nothing: the session
+// has expired and consumers render it that way.
 //
 // Returns one of the sentinel Err* values when the deadline fails the
 // sanity checks (pre-epoch, far future, or past beyond maxPastHorizon).
@@ -155,7 +187,7 @@ func (w *Watcher) Update(deadline time.Time) error {
 		return nil
 	}
 
-	now := time.Now()
+	now := w.nowFn()
 	switch {
 	case deadline.Before(time.Unix(0, 0)):
 		w.clearLocked()
@@ -173,18 +205,22 @@ func (w *Watcher) Update(deadline time.Time) error {
 		return nil
 	}
 
-	w.stopTimerLocked()
 	w.current = deadline
-	// Reset every per-deadline guard so a refreshed deadline arms a fresh
+	// Reset every per-deadline guard so a refreshed deadline starts a fresh
 	// warning cycle: both edge triggers and the user Dismiss decision
 	// (the user agreed to the old deadline expiring; a new deadline
 	// restarts the contract).
 	w.firedAt = time.Time{}
 	w.finalFiredAt = time.Time{}
 	w.dismissedAt = time.Time{}
+	w.announcedAt = time.Time{}
 
-	if deadline.After(now) {
-		w.armTimerLocked(deadline)
+	// Poll every accepted deadline, including one that reads as already
+	// expired: the clock may be running ahead of real time and get
+	// corrected later, and evaluate ignores a deadline that has genuinely
+	// passed anyway.
+	if !w.deadlineOnly {
+		w.startPollLocked()
 	}
 	recorder := w.recorder
 	w.mu.Unlock()
@@ -192,6 +228,28 @@ func (w *Watcher) Update(deadline time.Time) error {
 		recorder.SetSessionExpiresAt(deadline)
 	}
 	log.Infof("auth session deadline set to: %s (in %s)", deadline.Format(time.RFC3339), time.Until(deadline).Round(time.Second))
+
+	// Open the gate only once the recorder knows the new deadline, so a
+	// warning that refers to it can never reach consumers before the state
+	// change itself. A tick landing in between finds the gate shut.
+	w.mu.Lock()
+	if w.closed || !w.current.Equal(deadline) {
+		w.mu.Unlock()
+		return nil
+	}
+	w.announcedAt = deadline
+	wake := w.wake
+	w.mu.Unlock()
+
+	// Hand the evaluation to the loop rather than running it here, so a
+	// deadline that already sits inside a warning window is published at
+	// once without this goroutine ever touching the recorder.
+	if wake != nil {
+		select {
+		case wake <- struct{}{}:
+		default:
+		}
+	}
 	return nil
 }
 
@@ -204,9 +262,9 @@ func (w *Watcher) Deadline() time.Time {
 }
 
 // Dismiss records the user's "Dismiss" action against the current deadline
-// and suppresses the upcoming final-warning callback for that deadline.
-// Idempotent: repeated calls are no-ops. A subsequent Update with a fresh
-// deadline resets the dismissal so the final-warning cycle re-arms.
+// and suppresses the final warning for that deadline. Idempotent: repeated
+// calls are no-ops. A subsequent Update with a fresh deadline resets the
+// dismissal so the final-warning cycle starts over.
 //
 // No-op when the watcher holds no deadline or has been closed.
 func (w *Watcher) Dismiss() {
@@ -219,35 +277,48 @@ func (w *Watcher) Dismiss() {
 		return
 	}
 	w.dismissedAt = w.current
-	// Cancel the armed final-warning timer eagerly. fireFinal would also
-	// gate on dismissedAt, but stopping the timer avoids a wakeup with
-	// nothing to do and makes the intent visible.
-	if w.finalTimer != nil {
-		w.finalTimer.Stop()
-		w.finalTimer = nil
-	}
 	log.Infof("auth session final-warning dismissed for deadline %s", w.current.Format(time.RFC3339))
 }
 
-// Close stops any pending timer. Update calls after Close are ignored.
-// The recorder keeps its deadline: the watcher is engine-scoped and closes
-// on every engine restart (network change, sleep/wake, stream errors)
-// while the SSO deadline stays valid across those, so clearing here would
-// blank the UI's "expires in" row on every transient reconnect. The
-// client run loop clears the server-scoped recorder when it exits for
-// real (Down, profile switch, permanent login failure).
+// Close stops the evaluation loop and waits for it to exit. Update calls
+// after Close are ignored. The recorder keeps its deadline: the watcher is
+// engine-scoped and closes on every engine restart (network change,
+// sleep/wake, stream errors) while the SSO deadline stays valid across
+// those, so clearing here would blank the UI's "expires in" row on every
+// transient reconnect. The client run loop clears the server-scoped
+// recorder when it exits for real (Down, profile switch, permanent login
+// failure).
 func (w *Watcher) Close() {
 	w.mu.Lock()
-	defer w.mu.Unlock()
 	if w.closed {
+		// A concurrent Close is already tearing down. w.done outlives it,
+		// so this caller waits for the same loop rather than returning
+		// while a warning is still on its way to the recorder.
+		done := w.done
+		w.mu.Unlock()
+		if done != nil {
+			<-done
+		}
 		return
 	}
 	w.closed = true
-	w.stopTimerLocked()
 	w.current = time.Time{}
 	w.firedAt = time.Time{}
 	w.finalFiredAt = time.Time{}
 	w.dismissedAt = time.Time{}
+	w.announcedAt = time.Time{}
+	// Copy the channels out before releasing the lock: the loop takes w.mu
+	// on every tick, so waiting for it while holding the lock would
+	// deadlock. w.done stays on the receiver for the branch above.
+	stop, done := w.stop, w.done
+	w.stop, w.wake = nil, nil
+	w.mu.Unlock()
+
+	if stop == nil {
+		return
+	}
+	close(stop)
+	<-done
 }
 
 // clearLocked drops the tracked deadline and notifies the recorder so
@@ -259,11 +330,11 @@ func (w *Watcher) clearLocked() {
 		w.mu.Unlock()
 		return
 	}
-	w.stopTimerLocked()
 	w.current = time.Time{}
 	w.firedAt = time.Time{}
 	w.finalFiredAt = time.Time{}
 	w.dismissedAt = time.Time{}
+	w.announcedAt = time.Time{}
 	recorder := w.recorder
 	w.mu.Unlock()
 	if recorder != nil {
@@ -272,87 +343,121 @@ func (w *Watcher) clearLocked() {
 	log.Infof("auth session deadline cleared")
 }
 
-func (w *Watcher) stopTimerLocked() {
-	if w.timer != nil {
-		w.timer.Stop()
-		w.timer = nil
+// startPollLocked starts the evaluation loop unless it is already
+// running. The loop starts lazily on the first accepted deadline, so a
+// client whose server never publishes a session expiry never pays for a
+// ticker, and it runs until Close: the watcher is engine-scoped, and a
+// cleared deadline is normally followed by a fresh one on the next sync.
+// Caller must hold w.mu.
+func (w *Watcher) startPollLocked() {
+	if w.stop != nil {
+		return
 	}
-	if w.finalTimer != nil {
-		w.finalTimer.Stop()
-		w.finalTimer = nil
+	stop := make(chan struct{})
+	done := make(chan struct{})
+	wake := make(chan struct{}, 1)
+	w.stop, w.done, w.wake = stop, done, wake
+	go w.poll(stop, done, wake, w.interval)
+}
+
+// poll re-evaluates the tracked deadline every interval, and as soon as a
+// new deadline is announced. It is the only caller of evaluate, so Close
+// waiting for it to exit is enough to know no warning is still on its way
+// to the recorder. Its channels and interval are passed in rather than
+// read off the receiver, so Close can clear them without racing it.
+func (w *Watcher) poll(stop <-chan struct{}, done chan<- struct{}, wake <-chan struct{}, interval time.Duration) {
+	defer close(done)
+
+	ticker := time.NewTicker(interval)
+	defer ticker.Stop()
+
+	for {
+		select {
+		case <-stop:
+			return
+		case <-wake:
+			w.evaluate()
+		case <-ticker.C:
+			w.evaluate()
+		}
 	}
 }
 
-func (w *Watcher) armTimerLocked(deadline time.Time) {
-	w.timer = armOneShotLocked(deadline.Add(-w.lead), func() { w.fire(deadline) })
-	// finalLead <= 0 disables the final-warning timer entirely. Used by
-	// tests that predate the final-warning fallback so a millisecond-scale
-	// deadline does not flush both timers at once.
-	if w.finalLead > 0 {
-		w.finalTimer = armOneShotLocked(deadline.Add(-w.finalLead), func() { w.fireFinal(deadline) })
-	}
-}
-
-func (w *Watcher) fire(armedFor time.Time) {
+// evaluate compares the tracked deadline against the wall clock and
+// publishes whichever warning the remaining time calls for. Inside the
+// final-warning window the interactive warning is stale, so the final one
+// is published in its place: a device that resumes there never saw the
+// T-WarningLead notification.
+func (w *Watcher) evaluate() {
 	w.mu.Lock()
-	if w.closed || !w.current.Equal(armedFor) {
-		// Deadline moved while we were waiting (e.g. a successful extend).
-		// The reschedule path armed a fresh timer; this one is stale.
+	if w.closed || w.deadlineOnly || w.current.IsZero() {
 		w.mu.Unlock()
 		return
 	}
-	if !w.firedAt.IsZero() && w.firedAt.Equal(armedFor) {
+
+	deadline := w.current
+	if !w.announcedAt.Equal(deadline) {
+		// Update is still on its way to the recorder with this deadline.
 		w.mu.Unlock()
 		return
 	}
-	w.firedAt = armedFor
+	// Round(0) strips the monotonic reading so the comparison is wall
+	// clock on both sides, whether the deadline came off the wire or from
+	// a caller that derived it from time.Now.
+	remaining := deadline.Round(0).Sub(w.nowFn().Round(0))
+
+	switch {
+	case remaining <= 0:
+		// Already expired: the post-mortem SessionExpired flow owns it.
+		w.mu.Unlock()
+	case w.finalLead > 0 && remaining <= w.finalLead:
+		w.publishFinalLocked(deadline, remaining)
+	case remaining <= w.lead:
+		w.publishWarningLocked(deadline, remaining)
+	default:
+		w.mu.Unlock()
+	}
+}
+
+// publishWarningLocked emits the interactive T-WarningLead warning, at
+// most once per deadline value. Caller must hold w.mu; this helper
+// releases it.
+func (w *Watcher) publishWarningLocked(deadline time.Time, remaining time.Duration) {
+	if w.firedAt.Equal(deadline) {
+		w.mu.Unlock()
+		return
+	}
+	w.firedAt = deadline
 	recorder := w.recorder
 	w.mu.Unlock()
 	if recorder == nil {
 		return
 	}
-	log.Infof("auth session expiry soon warning fired")
-	publishWarning(recorder, armedFor, false)
+	log.Infof("auth session expiry soon warning fired for deadline %s (in %s)",
+		deadline.Format(time.RFC3339), remaining.Round(time.Second))
+	publishWarning(recorder, deadline, false)
 }
 
-// fireFinal mirrors fire for the T-FinalWarningLead timer with an extra
-// dismiss-gate: if the user dismissed the T-WarningLead notification for
-// this deadline, the final warning is suppressed entirely.
-func (w *Watcher) fireFinal(armedFor time.Time) {
-	w.mu.Lock()
-	if w.closed || !w.current.Equal(armedFor) {
+// publishFinalLocked emits the final warning, at most once per deadline
+// value and never once the user dismissed that deadline. It marks the
+// interactive warning as handled too: the final window is open, so a
+// "expires in WarningLead minutes" notification would be wrong. Caller
+// must hold w.mu; this helper releases it.
+func (w *Watcher) publishFinalLocked(deadline time.Time, remaining time.Duration) {
+	if w.finalFiredAt.Equal(deadline) || w.dismissedAt.Equal(deadline) {
 		w.mu.Unlock()
 		return
 	}
-	if !w.finalFiredAt.IsZero() && w.finalFiredAt.Equal(armedFor) {
-		w.mu.Unlock()
-		return
-	}
-	if w.dismissedAt.Equal(armedFor) {
-		w.mu.Unlock()
-		log.Infof("auth session final-warning skipped (dismissed by user)")
-		return
-	}
-	w.finalFiredAt = armedFor
+	w.firedAt = deadline
+	w.finalFiredAt = deadline
 	recorder := w.recorder
 	w.mu.Unlock()
 	if recorder == nil {
 		return
 	}
-	log.Infof("auth session final-warning fired")
-	publishWarning(recorder, armedFor, true)
-}
-
-// armOneShotLocked schedules cb at fireAt. When fireAt is already in the
-// past it dispatches on the next scheduler tick so a state-change recorder
-// notification (invoked after w.mu is released) lands first. Caller must
-// hold w.mu.
-func armOneShotLocked(fireAt time.Time, cb func()) *time.Timer {
-	delay := time.Until(fireAt)
-	if delay <= 0 {
-		return time.AfterFunc(0, cb)
-	}
-	return time.AfterFunc(delay, cb)
+	log.Infof("auth session final-warning fired for deadline %s (in %s)",
+		deadline.Format(time.RFC3339), remaining.Round(time.Second))
+	publishWarning(recorder, deadline, true)
 }
 
 // publishWarning composes the SystemEvent for a watcher-fired warning and

client/internal/auth/sessionwatch/watcher_test.go

@@ -19,6 +19,9 @@ type fakeRecorder struct {
 	mu           sync.Mutex
 	events       []event
 	lastDeadline time.Time
+	// setDelay stalls SetSessionExpiresAt, widening the window in which a
+	// concurrent evaluation could publish a warning out of order.
+	setDelay time.Duration
 }
 
 type eventKind int
@@ -43,6 +46,12 @@ type event struct {
 // is the zero time, so an initial clear before any deadline is set emits
 // nothing — matching the real recorder.
 func (r *fakeRecorder) SetSessionExpiresAt(deadline time.Time) {
+	r.mu.Lock()
+	delay := r.setDelay
+	r.mu.Unlock()
+	// Stall without the lock held, so a concurrent publish can still record.
+	time.Sleep(delay)
+
 	r.mu.Lock()
 	defer r.mu.Unlock()
 	if r.lastDeadline.Equal(deadline) {
@@ -117,10 +126,54 @@ func waitForEvents(t *testing.T, r *fakeRecorder, want int) []event {
 	return nil
 }
 
-// newWatcher builds a watcher with the final timer disabled (finalLead=0),
+// testInterval keeps the ticker-driven tests fast; the leads they use are
+// in the same millisecond scale.
+const testInterval = 2 * time.Millisecond
+
+// newWatcher builds a watcher with the final warning disabled (finalLead=0),
 // matching the lead-only behaviour the pre-final-warning tests assume.
 func newWatcher(lead time.Duration, r *fakeRecorder) *Watcher {
-	return NewWithLeads(lead, 0, r)
+	return newWatcherWithLeads(lead, 0, r)
+}
+
+// newWatcherWithLeads builds a watcher that evaluates on testInterval. The
+// interval is set before Update, so the evaluation loop does not exist yet
+// and the write cannot race it.
+func newWatcherWithLeads(lead, final time.Duration, r StatusRecorder) *Watcher {
+	w := NewWithLeads(lead, final, r)
+	w.interval = testInterval
+	return w
+}
+
+// fakeClock is the watcher's wall clock under test. Reads come from the
+// evaluation goroutine while the test writes, so both go through the mutex.
+type fakeClock struct {
+	mu sync.Mutex
+	t  time.Time
+}
+
+func newFakeClock(t time.Time) *fakeClock {
+	return &fakeClock{t: t}
+}
+
+func (c *fakeClock) now() time.Time {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	return c.t
+}
+
+// set jumps the clock, standing in for a resume from suspension or for an
+// NTP correction.
+func (c *fakeClock) set(t time.Time) {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	c.t = t
+}
+
+// settle waits out a handful of evaluation ticks so a test can assert that
+// nothing was published.
+func settle() {
+	time.Sleep(20 * testInterval)
 }
 
 func TestUpdateZeroBeforeAnythingIsNoop(t *testing.T) {
@@ -411,7 +464,7 @@ func TestCloseWithoutDeadlineLeavesRecorderUntouched(t *testing.T) {
 func TestFinalWarningFiresAfterRegularWarning(t *testing.T) {
 	r := &fakeRecorder{}
 	// Warning fires at deadline-80ms, final at deadline-30ms.
-	w := NewWithLeads(80*time.Millisecond, 30*time.Millisecond, r)
+	w := newWatcherWithLeads(80*time.Millisecond, 30*time.Millisecond, r)
 	defer w.Close()
 
 	d := time.Now().Add(100 * time.Millisecond)
@@ -447,7 +500,7 @@ func TestFinalWarningFiresAfterRegularWarning(t *testing.T) {
 
 func TestDismissSuppressesFinalWarning(t *testing.T) {
 	r := &fakeRecorder{}
-	w := NewWithLeads(80*time.Millisecond, 30*time.Millisecond, r)
+	w := newWatcherWithLeads(80*time.Millisecond, 30*time.Millisecond, r)
 	defer w.Close()
 
 	d := time.Now().Add(100 * time.Millisecond)
@@ -478,7 +531,7 @@ func TestDismissSuppressesFinalWarning(t *testing.T) {
 
 func TestDismissResetByNewDeadline(t *testing.T) {
 	r := &fakeRecorder{}
-	w := NewWithLeads(80*time.Millisecond, 30*time.Millisecond, r)
+	w := newWatcherWithLeads(80*time.Millisecond, 30*time.Millisecond, r)
 	defer w.Close()
 
 	first := time.Now().Add(100 * time.Millisecond)
@@ -508,7 +561,7 @@ func TestDismissResetByNewDeadline(t *testing.T) {
 
 func TestDismissBeforeUpdateIsNoop(t *testing.T) {
 	r := &fakeRecorder{}
-	w := NewWithLeads(80*time.Millisecond, 30*time.Millisecond, r)
+	w := newWatcherWithLeads(80*time.Millisecond, 30*time.Millisecond, r)
 	defer w.Close()
 
 	// No deadline tracked yet; Dismiss must be a no-op (no panic, no state).
@@ -528,3 +581,340 @@ func TestDismissBeforeUpdateIsNoop(t *testing.T) {
 	}
 	t.Fatalf("final-warning did not publish after no-op pre-Update Dismiss, events=%+v", r.snapshot())
 }
+
+// The tests below drive the watcher's wall clock directly. The deadline sits
+// an hour out in real time and the fake clock jumps, standing in for a device
+// that was suspended and resumed somewhere inside — or past — the warning
+// windows. The evaluation loop is what reacts to the jump, so these exercise
+// the same path production takes on a resume.
+func newResumeWatcher(t *testing.T, r *fakeRecorder) (*Watcher, *fakeClock, time.Time) {
+	t.Helper()
+
+	w := newWatcherWithLeads(WarningLead, FinalWarningLead, r)
+	start := time.Now()
+	clock := newFakeClock(start)
+	// Set before Update: the evaluation loop does not exist yet.
+	w.nowFn = clock.now
+	t.Cleanup(w.Close)
+
+	deadline := start.Add(time.Hour).Round(0)
+	if err := w.Update(deadline); err != nil {
+		t.Fatalf("Update: %v", err)
+	}
+	return w, clock, deadline
+}
+
+func TestResumeInsideWarningWindowWarns(t *testing.T) {
+	r := &fakeRecorder{}
+	_, clock, deadline := newResumeWatcher(t, r)
+
+	clock.set(deadline.Add(-5 * time.Minute))
+
+	events := waitForEvents(t, r, 2)
+	if !events[1].isWarning() {
+		t.Fatalf("expected the interactive warning after the resume, got %+v", events[1])
+	}
+	if n := countWhere(events, event.isFinalWarning); n != 0 {
+		t.Fatalf("final-warning must wait for its own window, got %d: %+v", n, events)
+	}
+}
+
+func TestResumeInsideFinalWindowSendsFinalWarningOnly(t *testing.T) {
+	r := &fakeRecorder{}
+	_, clock, deadline := newResumeWatcher(t, r)
+
+	clock.set(deadline.Add(-time.Minute))
+
+	events := waitForEvents(t, r, 2)
+	if !events[1].isFinalWarning() {
+		t.Fatalf("expected the final warning after the resume, got %+v", events[1])
+	}
+	settle()
+	if n := countWhere(r.snapshot(), event.isWarning); n != 0 {
+		t.Fatalf("the interactive warning is stale inside the final window, got %d: %+v", n, r.snapshot())
+	}
+}
+
+func TestResumePastDeadlinePublishesNothing(t *testing.T) {
+	r := &fakeRecorder{}
+	_, clock, deadline := newResumeWatcher(t, r)
+
+	clock.set(deadline.Add(time.Minute))
+
+	settle()
+	if n := countWhere(r.snapshot(), func(e event) bool { return e.kind == publish }); n != 0 {
+		t.Fatalf("an expired session must not warn, got %d publishes: %+v", n, r.snapshot())
+	}
+}
+
+func TestWarningPublishesOncePerDeadline(t *testing.T) {
+	r := &fakeRecorder{}
+	_, clock, deadline := newResumeWatcher(t, r)
+
+	clock.set(deadline.Add(-5 * time.Minute))
+	waitForEvents(t, r, 2)
+
+	// Many ticks pass inside the same window.
+	settle()
+	if n := countWhere(r.snapshot(), event.isWarning); n != 1 {
+		t.Fatalf("expected exactly 1 warning publish across ticks, got %d: %+v", n, r.snapshot())
+	}
+}
+
+// TestWarningRecoversFromAClockRunningAhead covers the device that boots
+// before NTP has corrected it: the deadline looks long gone, nothing is
+// published, and the warning still arrives once the clock is fixed.
+func TestWarningRecoversFromAClockRunningAhead(t *testing.T) {
+	r := &fakeRecorder{}
+	_, clock, deadline := newResumeWatcher(t, r)
+
+	clock.set(deadline.Add(2 * time.Hour))
+	settle()
+	if n := countWhere(r.snapshot(), func(e event) bool { return e.kind == publish }); n != 0 {
+		t.Fatalf("a deadline that looks expired must not warn, got %d publishes: %+v", n, r.snapshot())
+	}
+
+	clock.set(deadline.Add(-5 * time.Minute))
+
+	events := waitForEvents(t, r, 2)
+	if !events[1].isWarning() {
+		t.Fatalf("expected the warning once the clock was corrected, got %+v", events[1])
+	}
+}
+
+func TestResumeInsideFinalWindowRespectsDismiss(t *testing.T) {
+	r := &fakeRecorder{}
+	w, clock, deadline := newResumeWatcher(t, r)
+
+	// The user dismissed the warning for this deadline before the device
+	// was suspended; resuming inside the final window must not reopen it.
+	w.Dismiss()
+	clock.set(deadline.Add(-time.Minute))
+
+	settle()
+	if n := countWhere(r.snapshot(), func(e event) bool { return e.kind == publish }); n != 0 {
+		t.Fatalf("a dismissed deadline must not warn on resume, got %d: %+v", n, r.snapshot())
+	}
+}
+
+func TestCloseStopsTheEvaluationLoop(t *testing.T) {
+	r := &fakeRecorder{}
+	w := newWatcherWithLeads(WarningLead, FinalWarningLead, r)
+	start := time.Now()
+	clock := newFakeClock(start)
+	w.nowFn = clock.now
+
+	deadline := start.Add(time.Hour).Round(0)
+	if err := w.Update(deadline); err != nil {
+		t.Fatalf("Update: %v", err)
+	}
+	w.Close()
+
+	clock.set(deadline.Add(-5 * time.Minute))
+	settle()
+
+	if n := countWhere(r.snapshot(), func(e event) bool { return e.kind == publish }); n != 0 {
+		t.Fatalf("a closed watcher must not publish, got %d: %+v", n, r.snapshot())
+	}
+}
+
+func TestDeadlineOnlyRecordsDeadlineWithoutWarnings(t *testing.T) {
+	r := &fakeRecorder{}
+	w := NewDeadlineOnly(r)
+	defer w.Close()
+
+	// With the default leads this deadline sits inside the final-warning
+	// window, so a watcher that warns at all would publish on the spot.
+	d := time.Now().Add(50 * time.Millisecond).Round(0)
+	if err := w.Update(d); err != nil {
+		t.Fatalf("Update: %v", err)
+	}
+	if got := r.deadline(); !got.Equal(d) {
+		t.Fatalf("expected recorder deadline %v, got %v", d, got)
+	}
+
+	time.Sleep(100 * time.Millisecond)
+
+	events := r.snapshot()
+	if got := countWhere(events, func(e event) bool { return e.kind == publish }); got != 0 {
+		t.Fatalf("expected no publish in deadline-only mode, got %d: %+v", got, events)
+	}
+	w.mu.Lock()
+	polling := w.stop != nil
+	w.mu.Unlock()
+	if polling {
+		t.Fatal("expected no evaluation loop in deadline-only mode")
+	}
+}
+
+func TestDeadlineOnlyStillRejectsOutOfRangeDeadlines(t *testing.T) {
+	r := &fakeRecorder{}
+	w := NewDeadlineOnly(r)
+	defer w.Close()
+
+	if err := w.Update(time.Now().Add(time.Hour)); err != nil {
+		t.Fatalf("Update: %v", err)
+	}
+
+	err := w.Update(time.Now().Add(-maxPastHorizon - time.Hour))
+	if !errors.Is(err, ErrDeadlineInPast) {
+		t.Fatalf("expected ErrDeadlineInPast, got %v", err)
+	}
+	if got := r.deadline(); !got.IsZero() {
+		t.Fatalf("expected recorder cleared after rejection, got %v", got)
+	}
+}
+
+// TestClockAheadAtUpdateStillWarnsOnceCorrected covers a client that connects
+// while its clock runs ahead of real time: the deadline reads as already
+// expired when it arrives, so nothing publishes, and the warning must still
+// come once NTP pulls the clock back.
+func TestClockAheadAtUpdateStillWarnsOnceCorrected(t *testing.T) {
+	r := &fakeRecorder{}
+	w := newWatcherWithLeads(WarningLead, FinalWarningLead, r)
+	deadline := time.Now().Add(time.Hour).Round(0)
+	// Two hours past the deadline from the device's point of view, well
+	// inside maxPastHorizon, so Update accepts and records it.
+	clock := newFakeClock(deadline.Add(2 * time.Hour))
+	w.nowFn = clock.now
+	t.Cleanup(w.Close)
+
+	if err := w.Update(deadline); err != nil {
+		t.Fatalf("Update: %v", err)
+	}
+
+	settle()
+	if n := countWhere(r.snapshot(), func(e event) bool { return e.kind == publish }); n != 0 {
+		t.Fatalf("a deadline that reads as expired must not warn, got %d: %+v", n, r.snapshot())
+	}
+
+	clock.set(deadline.Add(-5 * time.Minute))
+
+	events := waitForEvents(t, r, 2)
+	if !events[1].isWarning() {
+		t.Fatalf("expected the warning once the clock was corrected, got %+v", events[1])
+	}
+}
+
+// TestWarningNeverPrecedesTheDeadlineStateChange pins the ordering Update
+// documents: consumers learn the new deadline before they see a warning that
+// refers to it. The deadline here already sits inside the warning window, and
+// the recorder stalls, so the evaluation loop gets many chances to publish
+// while Update is still announcing.
+func TestWarningNeverPrecedesTheDeadlineStateChange(t *testing.T) {
+	r := &fakeRecorder{setDelay: 50 * time.Millisecond}
+	w := newWatcherWithLeads(WarningLead, FinalWarningLead, r)
+	deadline := time.Now().Add(time.Hour).Round(0)
+	clock := newFakeClock(deadline.Add(-5 * time.Minute))
+	w.nowFn = clock.now
+	t.Cleanup(w.Close)
+
+	if err := w.Update(deadline); err != nil {
+		t.Fatalf("Update: %v", err)
+	}
+
+	events := waitForEvents(t, r, 2)
+	if events[0].kind != stateChange {
+		t.Fatalf("event[0] should be the deadline state change, got %+v", events)
+	}
+	if !events[1].isWarning() {
+		t.Fatalf("event[1] should be the warning, got %+v", events)
+	}
+}
+
+// TestUpdateWakesTheLoopWithoutWaitingForATick pins the hand-off: Update
+// publishes nothing itself, it nudges the evaluation loop, so a deadline that
+// already sits inside a warning window is warned about at once even though the
+// ticker here would not fire for an hour.
+func TestUpdateWakesTheLoopWithoutWaitingForATick(t *testing.T) {
+	r := &fakeRecorder{}
+	w := newWatcherWithLeads(WarningLead, FinalWarningLead, r)
+	w.interval = time.Hour
+	t.Cleanup(w.Close)
+
+	d := time.Now().Add(5 * time.Minute).Round(0)
+	if err := w.Update(d); err != nil {
+		t.Fatalf("Update: %v", err)
+	}
+
+	events := waitForEvents(t, r, 2)
+	if !events[1].isWarning() {
+		t.Fatalf("expected the warning on the wake-up, got %+v", events[1])
+	}
+}
+
+// blockingRecorder holds a publish open until the test releases it, so the
+// evaluation loop can be parked mid-publish while Close is called.
+type blockingRecorder struct {
+	fakeRecorder
+	entered chan struct{}
+	release chan struct{}
+}
+
+func (r *blockingRecorder) PublishEvent(
+	severity cProto.SystemEvent_Severity,
+	category cProto.SystemEvent_Category,
+	message string,
+	userMessage string,
+	metadata map[string]string,
+) bool {
+	select {
+	case r.entered <- struct{}{}:
+	default:
+	}
+	<-r.release
+	return r.fakeRecorder.PublishEvent(severity, category, message, userMessage, metadata)
+}
+
+// TestConcurrentCloseWaitsForTheLoop pins the contract for the caller that
+// loses the race: Close returns only once the loop is done, whichever of the
+// two calls got there first.
+func TestConcurrentCloseWaitsForTheLoop(t *testing.T) {
+	r := &blockingRecorder{entered: make(chan struct{}, 1), release: make(chan struct{})}
+	var releaseOnce sync.Once
+	release := func() { releaseOnce.Do(func() { close(r.release) }) }
+
+	w := newWatcherWithLeads(WarningLead, FinalWarningLead, r)
+	w.interval = time.Hour
+	// Order matters: Close waits for the loop, which stays parked in the
+	// publish until the release, so a t.Fatal below would hang the cleanup
+	// instead of reporting the failure.
+	t.Cleanup(func() {
+		release()
+		w.Close()
+	})
+
+	d := time.Now().Add(5 * time.Minute).Round(0)
+	if err := w.Update(d); err != nil {
+		t.Fatalf("Update: %v", err)
+	}
+
+	// The loop is now parked inside the warning publish.
+	select {
+	case <-r.entered:
+	case <-time.After(2 * time.Second):
+		t.Fatal("the loop never reached the publish")
+	}
+
+	first := make(chan struct{})
+	second := make(chan struct{})
+	go func() { w.Close(); close(first) }()
+	go func() { w.Close(); close(second) }()
+
+	select {
+	case <-first:
+		t.Fatal("Close returned while the loop was still publishing")
+	case <-second:
+		t.Fatal("Close returned while the loop was still publishing")
+	case <-time.After(100 * time.Millisecond):
+	}
+
+	release()
+	for _, done := range []chan struct{}{first, second} {
+		select {
+		case <-done:
+		case <-time.After(2 * time.Second):
+			t.Fatal("Close did not return after the publish completed")
+		}
+	}
+}