CB-578 stage A: classify a usage-limit refusal instead of a completed reply

A backend that refuses on a subscription usage limit leaves the pane healthy but
the turn ends with no bridge_reply; the completion fallback used to scrape and
hand that refusal back as if it were a real answer. CompletionResolver now
matches the scrape against a per-profile exhaustedPattern (config, never a
vendor string) and resolves the send as Rendezvous.Kind/Outcome.BACKEND_EXHAUSTED
with a reason carrying the matched line, kept distinct from GONE/WORKER_FAILED.
A profile with no pattern configured is unaffected. Coverage is logged at
startup via CompletionResolver.coverage(...), naming which profiles have a
pattern and which don't, following FleetHealthMonitor.coverage's pattern.

bridged/bridged.example.yaml

@@ -47,15 +47,18 @@ bind:
 # (say a Claude lead and an opencode lead) work as peers: the second is silently demoted and refused
 # every orchestration call. List each lead's pane here and all of them resolve as leads.
 #
-#   terminal → the ONLY field identity depends on; get it from that session's bridge_whoami
+#   tab → the ONLY field identity depends on (CB-579); the exact label of the tab hosting the lead.
+#         Label the tab yourself, or let bridged label one it launches — see `fleet.leaders:` below.
 #   kind/model → descriptive; they document what runs in the pane and are echoed by bridge_whoami
 #
-# A lead is never spawned — it pre-exists, which is exactly why it must be named rather than created.
+# A lead's tab must already carry its label (or be launched by bridged, which labels it) — there is
+# no terminal id to paste in and nothing to re-pin when the session restarts: the tab survives, so
+# the same label resolves the same lead again on the next scan.
 # `bridge_whoami` reports `{"role":"primary","leader":"<name>"}`; role stays "primary" because a lead
 # IS a primary for authorization, so nothing that keys on the role breaks.
 #
 # KEEP `primary:` when adding leads: it still addresses the CB-307 push loop, which needs a single
-# destination for its nudges. If both name the same terminal, the `fleet.leaders:` entry wins.
+# destination for its nudges, and is a separate mechanism from lead identity — see `fleet.leaders:`.
 #
 # Leads are configured under `fleet.leaders:` — see THE FLEET further down.
 #
@@ -85,6 +88,16 @@ bind:
 #   backoffMs: 60000
 #   quietNudgeCap: 3
 
+# Fleet health detection is dormant unless enabled. It reads one whole-fleet agent list per tick.
+# It can run without a webhook; bridge_list then reports healthCoverage: detection-only.
+# health:
+#   enabled: true
+#   intervalSeconds: 30              # minimum 15
+#   workingSuspectAfterSeconds: 600  # minimum 300
+#   paneProbeIntervalSeconds: 60     # minimum 60
+#   notifications:
+#     mode: disabled                 # disabled (default) or webhook
+
 # herdr Unix socket. Omit to use the client default
 # (${HERDR_SOCKET_PATH:-~/.config/herdr/herdr.sock}).
 herdrSocket: ~/.config/herdr/herdr.sock
@@ -125,6 +138,12 @@ herdrSocket: ~/.config/herdr/herdr.sock
 #                      SSH is unaffected). The token value itself is never stored in this file.
 #   gitHostEnv       → host env var holding the forge host (default GITEA_HOST). Injected as
 #                      GITEA_HOST *only* alongside a resolved gitTokenEnv.
+#   exhaustedPattern → regex matched against a completion-fallback scrape (CB-578 stage A) to
+#                      classify a turn that ended with no bridge_reply as the backend having
+#                      refused on a subscription usage limit, rather than a real answer. Opt-in —
+#                      omit and this profile's completion fallback behaves exactly as before.
+#                      Every backend words its refusal differently, so this is config, never a
+#                      vendor string baked into bridged itself.
 #   env              → extra environment for this profile's workers, as a literal key/value map
 #                      (CB-511). Use it to give workers a toolchain.
 #
@@ -158,6 +177,7 @@ profiles:
     maxLoad: 2                        # max live workers on this profile (omit for unlimited)
     # gitTokenEnv: GITEA_TOKEN        # opt-in: let this profile's workers open their own PR (CB-302)
     # gitHostEnv: GITEA_HOST          # defaults to GITEA_HOST; injected only with gitTokenEnv
+    # exhaustedPattern: "usage limit has been reached"   # opt-in: classify a usage-limit refusal (CB-578)
     # configDir: /Users/me/.ccs/instances/gx10   # CLAUDE_CONFIG_DIR — inherit that profile's skills/MCP
     # cwd: /Users/me/src/myrepo       # pin the working dir; omit to inherit the primary's
     # parityOverlay: [".claude/settings.local.json", ".env", ".envrc"]   # never add .mcp.json — see above
@@ -299,15 +319,18 @@ fleet:
 
   # Panes that orchestrate rather than are orchestrated. A lead may now be CREATED as well as
   # recognised: give it a `profile:` and the daemon launches the shortfall when fewer than
-  # `instances` are live. Give it only a `terminal:` and it is recognise-only, as before.
+  # `instances` are live. Omit `profile:` and it is recognise-only, as before.
   #
-  # `tabPrefix` is the naming convention that finds a lead without pasting a terminal id: label the
-  # tab `lead: <name>` when you open it and the pane is recognised on the next rescan. Reopen the
-  # tab later and the id changes; the label does not.
+  # `tab:` (CB-579) is REQUIRED and is the only field identity depends on — the exact label of the
+  # tab hosting the lead, matched case-insensitively. Label the tab yourself and put that same
+  # string here, and the pane is recognised on the next rescan. Reopen the tab later, or the session
+  # inside it restarts — the terminal id changes; the tab, and its label, do not, so no config edit
+  # follows a restart.
   #
-  # A lead the daemon launches is labelled BY the daemon, using the same convention, so it is found
+  # A lead the daemon launches is labelled BY the daemon with this same `tab:` value, so it is found
   # by the same scan. A lead counts as live only when herdr also reports a running agent in that
-  # tab — a label left behind by a session that died does not block the relaunch.
+  # tab — a label left behind by a session that died does not block the relaunch, and a tab that is
+  # gone entirely drops out of the next scan rather than being remembered forever.
   #
   # An auto-launched lead is NOT a member: it gets no worker reply charter, is never registered with
   # the session lifecycle (the idle reaper would kill your orchestrator), and stays on the
@@ -316,10 +339,9 @@ fleet:
   #   opus-5.0:
   #     profile: opus                # omit to never create this lead, only recognise it
   #     instances: 1                 # desired live count; only the shortfall is launched. 0 = off
-  #     terminal: term_0123456789abcd  # optional hand-pin; usually found by tabPrefix instead.
-  #                                  # A running agent on this terminal also counts as live, so a
-  #                                  # lead you opened by hand is not relaunched under you.
-  #     tabPrefix: "lead:"           # `lead: opus-5.0` ⇒ a lead named opus-5.0 (case-insensitive)
+  #     tab: "lead: opus-5.0"        # REQUIRED — the exact tab label this lead lives in
+  #     tabPrefix: "lead:"           # only used to guard against a worker tabLabel colliding with
+  #                                  # this convention at startup; plays no part in matching a lead
   #     scanIntervalSeconds: 10      # rescan cadence, and the worst case before a new tab is seen
   #     workspace: leads             # where a launched lead's tab is created (default "leads").
   #                                  # MUST NOT be a member workspace — those are excluded from the
@@ -327,7 +349,7 @@ fleet:
   #     cwd: /path/to/repo           # the launched lead's working directory (default: bridged's own)
   #     kind: claude                 # descriptive; reported by bridge_whoami
   #   gpt-sol-5.6:
-  #     terminal: term_fedcba9876543
+  #     tab: "lead: gpt-sol-5.6"
   #     kind: opencode
   #     model: openai/gpt-5.6-terra
 

bridged/src/main/java/dev/ltms/bridged/Bridged.java

@@ -13,6 +13,7 @@ import dev.ltms.bridged.herdr.PaneLocator;
 import dev.ltms.bridged.herdr.UnixSocketHerdrClient;
 import dev.ltms.bridged.herdr.WorkspaceControl;
 import dev.ltms.bridged.inject.CompletionResolver;
+import dev.ltms.bridged.inject.ExhaustedPatternLookup;
 import dev.ltms.bridged.inject.Injector;
 import dev.ltms.bridged.inject.StatusPoller;
 import dev.ltms.bridged.inject.TurnListener;
@@ -23,6 +24,7 @@ import dev.ltms.bridged.mcp.BridgeMcp;
 import dev.ltms.bridged.mcp.ConnectionIdentity;
 import dev.ltms.bridged.metrics.BridgedMetrics;
 import dev.ltms.bridged.metrics.Metrics;
+import dev.ltms.bridged.health.FleetHealthMonitor;
 import dev.ltms.bridged.mcp.PrimaryRegistry;
 import dev.ltms.bridged.mcp.LsofPeerPidLookup;
 import dev.ltms.bridged.mcp.LsofProcessCwdLookup;
@@ -59,6 +61,7 @@ import java.util.concurrent.atomic.AtomicReference;
 import java.util.function.Function;
 import java.util.function.Predicate;
 import java.util.function.Supplier;
+import java.util.regex.Pattern;
 import java.util.stream.Collectors;
 
 /**
@@ -192,34 +195,34 @@ public final class Bridged {
         if (leadTerminals.size() > 1) {
             log.info("leads: {} panes recognised {}", leadTerminals.size(), leadTerminals.values());
         }
-        // CB-531: on top of the static registry, discover leads by the tab labels the operator
-        // writes. CB-557 moved the settings onto the lead they describe, so scanning is on whenever
-        // a `fleet.leaders:` entry exists — with no leads configured the supplier is a constant and
-        // never touches herdr, exactly as a missing `leadScan:` block used to behave.
+        // CB-531: on top of the legacy primary.terminal pin, discover leads by the tab labels the
+        // operator writes. CB-557 moved the settings onto the lead they describe, so scanning is on
+        // whenever a `fleet.leaders:` entry exists — with no leads configured the supplier is a
+        // constant and never touches herdr, exactly as a missing `leadScan:` block used to behave.
+        // CB-579: each lead now names its own exact `tab:` label, so one scanner discovers every
+        // configured lead regardless of how differently their tabs are labelled — the old
+        // single-shared-tabPrefix limitation (and its warning) is gone.
         final Supplier<Map<String, String>> leads;
         var leaders = cfg.fleet().leaders();
         if (!leaders.isEmpty()) {
-            // One scanner, so one prefix and one interval. Distinct per-lead prefixes would need a
-            // scanner each; until a config actually wants that, take the first entry's settings and
-            // say so, rather than silently honouring one lead's prefix and dropping another's.
-            var scan = leaders.values().iterator().next();
             Set<String> memberSpaces = cfg.profiles().values().stream()
                     .map(BridgedConfig.Profile::workspace)
                     .filter(Objects::nonNull)
                     .collect(Collectors.toSet());
-            leads = new LeadTabScanner(herdr, scan.tabPrefix(), memberSpaces, leadTerminals,
-                    TimeUnit.SECONDS.toNanos(scan.scanIntervalSeconds()), System::nanoTime);
-            log.info("lead scan: tabs labelled '{}…' host a lead (rescan every {}s, member spaces {} "
-                            + "excluded)",
-                    scan.tabPrefix(), scan.scanIntervalSeconds(), memberSpaces);
-            long distinctPrefixes = leaders.values().stream()
-                    .map(BridgedConfig.Leader::tabPrefix).distinct().count();
-            if (distinctPrefixes > 1) {
-                log.warn("fleet.leaders declares {} different tabPrefix values; only '{}' is scanned "
-                                + "for. Give every lead the same tabPrefix, or leads under the others "
-                                + "will not be discovered.",
-                        distinctPrefixes, scan.tabPrefix());
-            }
+            Map<String, String> tabToName = new LinkedHashMap<>();
+            leaders.forEach((name, leader) -> {
+                if (leader != null && leader.tab() != null && !leader.tab().isBlank()) {
+                    tabToName.put(leader.tab(), name);
+                }
+            });
+            // One shared rescan cadence: still taken from the first entry, as before — it is an
+            // operational cadence, not identity, so there is no correctness reason to give every
+            // lead its own scanner.
+            int scanIntervalSeconds = leaders.values().iterator().next().scanIntervalSeconds();
+            leads = new LeadTabScanner(herdr, tabToName, memberSpaces,
+                    TimeUnit.SECONDS.toNanos(scanIntervalSeconds), System::nanoTime);
+            log.info("lead scan: tabs {} host a lead (rescan every {}s, member spaces {} excluded)",
+                    tabToName.keySet(), scanIntervalSeconds, memberSpaces);
         } else {
             leads = () -> leadTerminals;
         }
@@ -252,7 +255,24 @@ public final class Bridged {
         // The blocking message endpoint (CB-104) is the producer; the poller is inert until then.
         // CB-106: a confirmed turn completion resolves a blocked send whose worker never replied.
         Rendezvous rendezvous = new Rendezvous();
-        CompletionResolver completion = new CompletionResolver(agents, rendezvous);
+        // CB-578 stage A: classify a completion-fallback scrape that matches a profile's configured
+        // usage-limit refusal as BACKEND_EXHAUSTED rather than handing it back as a real answer.
+        // Compiled once at startup, keyed by profile name; a profile with no exhaustedPattern is
+        // simply absent here, so its workers keep today's completion-fallback behaviour unchanged.
+        Map<String, Pattern> exhaustedPatternsByProfile = new LinkedHashMap<>();
+        cfg.profiles().forEach((name, profile) -> {
+            if (profile.hasExhaustedPattern()) {
+                exhaustedPatternsByProfile.put(name, Pattern.compile(profile.exhaustedPattern()));
+            }
+        });
+        ExhaustedPatternLookup exhaustedPatterns = target -> sessions.roster().stream()
+                .filter(session -> target.equals(session.terminalId()))
+                .findFirst()
+                .map(session -> exhaustedPatternsByProfile.get(session.profile()))
+                .orElse(null);
+        log.info("backend-exhausted classification (CB-578 stage A): {}",
+                CompletionResolver.coverage(cfg.profiles().keySet(), exhaustedPatternsByProfile.keySet()));
+        CompletionResolver completion = new CompletionResolver(agents, rendezvous, exhaustedPatterns);
         // CB-113: deliver only to an available worker (its MCP is connected), never its boot window.
         // CB-301: the manager's presence bridge records availability and drives SPAWNING → READY.
         MemberPresence presence = sessions.asPresence();
@@ -275,9 +295,9 @@ public final class Bridged {
             }
 
             @Override
-            public void onDelivered(String target) {
-                completion.onDelivered(target);
-                sessions.onDelivered(target);
+            public void onDelivered(String target, dev.ltms.bridged.msg.TurnToken token) {
+                completion.onDelivered(target, token);
+                sessions.onDelivered(target, token);
             }
 
             @Override
@@ -285,6 +305,12 @@ public final class Bridged {
                 completion.onTurnFailed(target);
                 sessions.onTurnFailed(target);
             }
+
+            @Override
+            public void onTurnFailed(String target, String reason) {
+                completion.onTurnFailed(target, reason);
+                sessions.onTurnFailed(target);
+            }
         };
         Injector injector = new Injector(agents, turnListener, deliverableTo(presence, leads),
                 presence::forget);
@@ -348,6 +374,28 @@ public final class Bridged {
         MessageService messages = new MessageService(agents, injector, rendezvous, replyInbox,
                 pushLoop, metrics);
 
+        // Health is a slow whole-fleet observer. Keep it separate from the 250ms delivery poller.
+        final FleetHealthMonitor healthMonitor;
+        var healthScheduler = Executors.newSingleThreadScheduledExecutor(r ->
+                Thread.ofVirtual().name("bridge-health-").unstarted(r));
+        if (cfg.health() != null && cfg.health().isEnabled()) {
+            // CB-580: a member found GONE/NEVER_READY must fail whatever ticket is waiting on it,
+            // through the same idempotent target-wide operation CB-516 already uses on release.
+            healthMonitor = new FleetHealthMonitor(agents, sessions::roster, messages, healthScheduler,
+                    System::nanoTime, cfg.health().intervalOrDefault(), messages::abandon);
+            String coverage = FleetHealthMonitor.coverage(true,
+                    cfg.health().notifications() != null && cfg.health().notifications().configured());
+            if ("detection-only".equals(coverage)) {
+                log.warn("fleet health: {} (no notification sink configured)", coverage);
+            } else {
+                log.info("fleet health: {}", coverage);
+            }
+            healthMonitor.start();
+        } else {
+            healthMonitor = null;
+            healthScheduler.shutdownNow();
+        }
+
         // CB-520: the reply inbox only consumes for agents this gateway owns. own on acquire,
         // release on teardown. Do this before CB-516 so the inbox is owned before any reply can land.
         sessions.onAcquire(replyInbox::own);
@@ -383,7 +431,16 @@ public final class Bridged {
         }
 
         BridgeMcp mcp = new BridgeMcp(messages, workers, sessions, identity, presence,
-                primaryRegistry, callers, metrics);
+                primaryRegistry, callers, metrics, new BridgeMcp.CapacitySource(profile -> liveCountRef.get().apply(profile),
+                profile -> {
+                    var configured = config.get().profiles().get(profile);
+                    return configured == null ? null : configured.maxLoad();
+                }, () -> config.get().profiles().keySet(), System::nanoTime),
+                new BridgeMcp.HealthCoverageSource(() -> {
+                    var health = config.get().health();
+                    return FleetHealthMonitor.coverage(health != null && health.isEnabled(),
+                            health != null && health.notifications() != null && health.notifications().configured());
+                }));
 
         // CB-559: opt-in config reload. With no `configReload:` block nothing is constructed, so an
         // upgraded daemon behaves exactly as before — the file is read once at boot and never again.
@@ -404,6 +461,7 @@ public final class Bridged {
             messages.close();
             pushLoop.close();
             if (heartbeat != null) heartbeat.close(); // CB-551: stop the idle-lead heartbeat scheduler
+            if (healthMonitor != null) healthMonitor.stop();
             if (configWatcher != null) configWatcher.stop(); // CB-559: stop polling the config file
             mcp.close();
             if (reaper != null) reaper.stop();

bridged/src/main/java/dev/ltms/bridged/config/BridgedConfig.java

@@ -73,6 +73,7 @@ public record BridgedConfig(
         Primary primary,
         Fleet fleet,
         LeadHeartbeat leadHeartbeat,
+        Health health,
         String placement,
         Auth auth,
         ConfigReload configReload) {
@@ -83,7 +84,16 @@ public record BridgedConfig(
                          Integer spawnReadyPollMs, Broker broker, Primary primary, Fleet fleet,
                          LeadHeartbeat leadHeartbeat, String placement, Auth auth) {
         this(bind, herdrSocket, profiles, guard, worktreeRoot, lifecycle, spawnReadyTimeoutMs,
-                spawnReadyPollMs, broker, primary, fleet, leadHeartbeat, placement, auth, null);
+                spawnReadyPollMs, broker, primary, fleet, leadHeartbeat, null, placement, auth, null);
+    }
+
+    /** Back-compat form before the optional {@code health:} block was added. */
+    public BridgedConfig(Bind bind, String herdrSocket, Map<String, Profile> profiles, Guard guard,
+                         String worktreeRoot, Lifecycle lifecycle, Integer spawnReadyTimeoutMs,
+                         Integer spawnReadyPollMs, Broker broker, Primary primary, Fleet fleet,
+                         LeadHeartbeat leadHeartbeat, String placement, Auth auth, ConfigReload configReload) {
+        this(bind, herdrSocket, profiles, guard, worktreeRoot, lifecycle, spawnReadyTimeoutMs,
+                spawnReadyPollMs, broker, primary, fleet, leadHeartbeat, null, placement, auth, configReload);
     }
 
     /**
@@ -171,6 +181,12 @@ public record BridgedConfig(
      *                       opposite intents). For the same reason, an {@code env:} entry naming
      *                       {@code ANTHROPIC_BASE_URL} or {@code ANTHROPIC_AUTH_TOKEN} is refused at
      *                       config load (CB-542): on the subscription path no guard would vet it.
+     * @param exhaustedPattern  regex matched against a completion-fallback scrape (CB-578 stage A) to
+     *                       classify a turn that ended with no {@code bridge_reply} as the backend
+     *                       having refused on a subscription usage limit, rather than a real answer.
+     *                       {@code null}/blank ⇒ the classification never fires for this profile and
+     *                       today's completion-fallback behaviour is unchanged. Every backend words
+     *                       its refusal differently, so this is config, never a vendor string in code.
      */
     @JsonIgnoreProperties(ignoreUnknown = true)
     public record Profile(String profile, String baseUrl, String model,
@@ -183,7 +199,8 @@ public record BridgedConfig(
                          Map<String, String> env,
                          Float weight,
                          Integer maxLoad,
-                         Boolean subscription) {
+                         Boolean subscription,
+                         String exhaustedPattern) {
 
         /** Peer kind spawned by {@link dev.ltms.bridged.member.ClaudeCodeLauncher} (the default). */
         public static final String KIND_CLAUDE_CODE = "claude-code";
@@ -226,6 +243,9 @@ public record BridgedConfig(
             weight = (weight == null || weight <= 0.0f) ? 1.0f : weight;
             maxLoad = (maxLoad == null || maxLoad <= 0) ? null : maxLoad;
             subscription = (subscription != null && subscription) ? Boolean.TRUE : Boolean.FALSE;
+            // exhaustedPattern stays null when unset/blank (opt-in) — no defaulting, no vendor
+            // wording: an unconfigured profile keeps today's completion-fallback behaviour exactly.
+            exhaustedPattern = (exhaustedPattern == null || exhaustedPattern.isBlank()) ? null : exhaustedPattern;
         }
 
         /**
@@ -238,7 +258,7 @@ public record BridgedConfig(
                       String placement, String workspace, String tabLabel, String mcpUrl,
                       String cwd, List<String> parityOverlay) {
             this(profile, baseUrl, model, configDir, tokenEnv, argv, placement, workspace, tabLabel,
-                    mcpUrl, cwd, parityOverlay, null, null, null, null, null, null, null);
+                    mcpUrl, cwd, parityOverlay, null, null, null, null, null, null, null, null);
         }
 
         /**
@@ -250,7 +270,7 @@ public record BridgedConfig(
                       String placement, String workspace, String tabLabel, String mcpUrl,
                       String cwd, List<String> parityOverlay, String gitTokenEnv, String gitHostEnv) {
             this(profile, baseUrl, model, configDir, tokenEnv, argv, placement, workspace, tabLabel,
-                    mcpUrl, cwd, parityOverlay, gitTokenEnv, gitHostEnv, null, null, null, null, null);
+                    mcpUrl, cwd, parityOverlay, gitTokenEnv, gitHostEnv, null, null, null, null, null, null);
         }
 
         /**
@@ -263,13 +283,14 @@ public record BridgedConfig(
                       String cwd, List<String> parityOverlay, String gitTokenEnv, String gitHostEnv,
                       String kind) {
             this(profile, baseUrl, model, configDir, tokenEnv, argv, placement, workspace, tabLabel,
-                    mcpUrl, cwd, parityOverlay, gitTokenEnv, gitHostEnv, kind, null, null, null, null);
+                    mcpUrl, cwd, parityOverlay, gitTokenEnv, gitHostEnv, kind, null, null, null, null, null);
         }
 
         /** A copy with {@code profile} set — used to default a profile to its {@code workers} key. */
         public Profile withProfile(String p) {
             return new Profile(p, baseUrl, model, configDir, tokenEnv, argv, placement, workspace, tabLabel,
-                    mcpUrl, cwd, parityOverlay, gitTokenEnv, gitHostEnv, kind, env, weight, maxLoad, subscription);
+                    mcpUrl, cwd, parityOverlay, gitTokenEnv, gitHostEnv, kind, env, weight, maxLoad, subscription,
+                    exhaustedPattern);
         }
 
         /** True when this profile is served by the Claude Code adapter (the default kind). */
@@ -302,7 +323,12 @@ public record BridgedConfig(
                       String cwd, List<String> parityOverlay, String gitTokenEnv, String gitHostEnv,
                       String kind, Map<String, String> env, Float weight, Integer maxLoad) {
             this(profile, baseUrl, model, configDir, tokenEnv, argv, placement, workspace, tabLabel,
-                    mcpUrl, cwd, parityOverlay, gitTokenEnv, gitHostEnv, kind, env, weight, maxLoad, null);
+                    mcpUrl, cwd, parityOverlay, gitTokenEnv, gitHostEnv, kind, env, weight, maxLoad, null, null);
+        }
+
+        /** True when this profile's CB-578 stage A backend-exhausted classification is configured. */
+        public boolean hasExhaustedPattern() {
+            return exhaustedPattern != null;
         }
 
         /** True when this profile's workers are granted a forge token to open their own PR (CB-302). */
@@ -377,6 +403,17 @@ public record BridgedConfig(
                             boolean clearAfterTurn) {
     }
 
+    /** Optional fleet detection. A missing block stays dormant. */
+    @JsonIgnoreProperties(ignoreUnknown = true)
+    public record Health(Boolean enabled, Integer intervalSeconds, Integer workingSuspectAfterSeconds,
+                         Integer paneProbeIntervalSeconds, Notifications notifications) {
+        public boolean isEnabled() { return Boolean.TRUE.equals(enabled); }
+        public int intervalOrDefault() { return Math.max(15, intervalSeconds == null ? 30 : intervalSeconds); }
+        public record Notifications(String mode) {
+            public boolean configured() { return "webhook".equalsIgnoreCase(mode); }
+        }
+    }
+
     /**
      * External AMQP broker for durable, cross-restart reply delivery (CB-307 Stage 2). Its mere
      * presence swaps the in-memory {@code ReplyInbox} for the AMQP-backed adapter; absent, bridged
@@ -437,24 +474,32 @@ public record BridgedConfig(
      * pre-existed, which is why it had to be recognised by configuration rather than created. With
      * {@code profile} and {@code instances} the daemon may stand one up when none is live, so the
      * pane no longer has to exist before the daemon does. Recognition still comes first: a lead
-     * already running under {@code tabPrefix} is adopted, and only the shortfall is launched.
+     * already running in its configured {@code tab} is adopted, and only the shortfall is launched.
+     *
+     * <p><b>{@code tab} replaced {@code terminal} (CB-579).</b> A herdr {@code terminal_id} changes
+     * every time the lead's session restarts, so pinning one cost a config edit and a daemon restart
+     * per restart. A tab is stable: a human opens it once, it holds exactly one pane, and its label
+     * survives restarts of the agent inside it — so identity is now the tab label alone.
      *
      * @param profile             the {@code profiles:} entry to launch this lead on when one must
      *                            be created; {@code null} ⇒ recognise-only, never create
-     * @param terminal            the lead's herdr {@code terminal_id} when pinned by hand; the only
-     *                            field identity depends on. {@code null} ⇒ found by {@code tabPrefix}
+     * @param tab                 the exact tab label hosting this lead, matched case-insensitively;
+     *                            the only field identity depends on. Required — a lead with no
+     *                            {@code tab} can never be discovered, launched or not
      * @param instances           how many of this lead should be live (default 1). The daemon
      *                            launches only the shortfall, so a restart adopts rather than doubles
-     * @param tabPrefix           label prefix marking this lead's tab, matched case-insensitively;
-     *                            the remainder is the lead's name ({@code "lead: opus"} →
-     *                            {@code opus}). Default {@code "lead:"}
+     * @param tabPrefix           no longer used to find a lead's tab — {@code tab} is matched
+     *                            exactly. Its only remaining job is the startup collision guard
+     *                            ({@link #validateLeadTabPrefixes()}), which still uses it to refuse
+     *                            a worker {@code tabLabel} template that could be misread as a lead.
+     *                            Default {@code "lead:"}
      * @param scanIntervalSeconds how long a tab scan is cached before herdr is asked again; also the
      *                            worst case before a newly-labelled tab is recognised. Default 10
      * @param kind                which agent runs there ({@code claude}, {@code opencode}, …)
      * @param model               the model or selector it runs, for operators reading the roster
      */
     @JsonIgnoreProperties(ignoreUnknown = true)
-    public record Leader(String profile, String terminal, Integer instances, String tabPrefix,
+    public record Leader(String profile, String tab, Integer instances, String tabPrefix,
                          Integer scanIntervalSeconds, String kind, String model,
                          String workspace, String cwd) {
 
@@ -472,12 +517,13 @@ public record BridgedConfig(
                     (scanIntervalSeconds == null || scanIntervalSeconds <= 0) ? 10 : scanIntervalSeconds;
             workspace = (workspace == null || workspace.isBlank())
                     ? DEFAULT_WORKSPACE : workspace.strip();
+            tab = (tab == null || tab.isBlank()) ? null : tab.strip();
         }
 
         /** Back-compat 7-arg form — no workspace or cwd, so both take their defaults. */
-        public Leader(String profile, String terminal, Integer instances, String tabPrefix,
+        public Leader(String profile, String tab, Integer instances, String tabPrefix,
                       Integer scanIntervalSeconds, String kind, String model) {
-            this(profile, terminal, instances, tabPrefix, scanIntervalSeconds, kind, model, null, null);
+            this(profile, tab, instances, tabPrefix, scanIntervalSeconds, kind, model, null, null);
         }
 
         /** True when this lead may be launched by the daemon rather than only recognised. */
@@ -485,9 +531,9 @@ public record BridgedConfig(
             return profile != null && !profile.isBlank() && instances > 0;
         }
 
-        /** The tab label an auto-launched instance of this lead gets — what the scanner reads back. */
-        public String tabLabel(String name) {
-            return tabPrefix + " " + name;
+        /** The tab label an auto-launched instance of this lead gets — its configured {@code tab}. */
+        public String tabLabel() {
+            return tab;
         }
     }
 
@@ -686,28 +732,20 @@ public record BridgedConfig(
     }
 
     /**
-     * The terminal → lead-name map that {@link dev.ltms.bridged.auth.CallerResolver} resolves
-     * against, merging the {@code leaders:} registry with the legacy singular {@code primary:} pin.
+     * The terminal → lead-name map seeded from the legacy singular {@code primary:} pin (CB-530).
      *
-     * <p>Precedence: an explicit {@code leaders:} entry wins over the {@code primary:} pin for the
-     * same terminal. The pin is the older, less expressive spelling of the same fact, so when both
-     * name a pane the named entry is the one an operator meant. The pin is still honoured on its
-     * own — a config carrying only {@code primary:} behaves exactly as it did before CB-530.
+     * <p>{@code fleet.leaders} no longer carries a per-entry terminal pin (CB-579): a lead's identity
+     * comes from its {@code tab} alone, resolved live by {@code LeadTabScanner}. This method now
+     * exists only for the {@code primary.terminal} fallback — a config that never migrated off it
+     * still resolves that one pane as a lead named {@code "primary"}, exactly as before CB-530.
      *
-     * @return an unmodifiable map, empty when neither block is configured (nothing is pinned, and
-     *         every pane therefore resolves as a worker — the pre-CB-307 behaviour)
+     * @return an unmodifiable map, empty when {@code primary.terminal} is not configured (nothing is
+     *         pinned, and every pane therefore resolves as a worker — the pre-CB-307 behaviour)
      */
     public Map<String, String> leaderTerminals() {
         Map<String, String> byTerminal = new LinkedHashMap<>();
-        if (fleet != null) {
-            fleet.leaders().forEach((name, leader) -> {
-                if (leader != null && leader.terminal() != null && !leader.terminal().isBlank()) {
-                    byTerminal.put(leader.terminal(), name);
-                }
-            });
-        }
         if (primary != null && primary.terminal() != null && !primary.terminal().isBlank()) {
-            byTerminal.putIfAbsent(primary.terminal(), "primary");
+            byTerminal.put(primary.terminal(), "primary");
         }
         return Collections.unmodifiableMap(byTerminal);
     }
@@ -812,13 +850,14 @@ public record BridgedConfig(
     private static final Set<String> KNOWN_TOP_LEVEL_KEYS = Set.of(
             "bind", "herdrSocket", "profiles", "guard", "worktreeRoot",
             "lifecycle", "spawnReadyTimeoutMs", "spawnReadyPollMs", "broker", "primary", "fleet",
-            "leadHeartbeat", "placement", "auth", "configReload");
+            "leadHeartbeat", "health", "placement", "auth", "configReload");
 
     /** Load and validate config from {@code path}. */
     public static BridgedConfig load(Path path) {
         try {
             String yaml = Files.readString(path);
             rejectRenamedTopLevelKeys(yaml);
+            rejectLeaderTerminalKey(yaml);
             warnUnknownTopLevelKeys(yaml, path);
             rejectDuplicateMemberSlots(yaml);
             BridgedConfig cfg = YAML.readValue(yaml, BridgedConfig.class);
@@ -1043,6 +1082,43 @@ public record BridgedConfig(
         }
     }
 
+    /**
+     * Reject a config whose {@code fleet.leaders.<name>} still carries the retired {@code terminal:}
+     * pin (CB-579), naming {@code tab:} as its replacement.
+     *
+     * <p>{@code Leader} is {@code @JsonIgnoreProperties(ignoreUnknown = true)}, so simply dropping
+     * the record component would make a leftover {@code terminal:} key silently no-op — the daemon
+     * would start, the pin would never take effect, and nothing would say why. Fatal and specific
+     * instead, exactly like {@link #rejectRenamedTopLevelKeys}, which this mirrors for a key one
+     * level deeper than the ones that method covers.
+     *
+     * @param yaml the raw config text
+     * @throws IllegalStateException when any {@code fleet.leaders.<name>.terminal} key is present
+     */
+    static void rejectLeaderTerminalKey(String yaml) {
+        Map<?, ?> raw;
+        try {
+            raw = YAML.readValue(yaml, Map.class);
+        } catch (IOException | IllegalArgumentException e) {
+            return; // a malformed file is reported by the real parse, not here
+        }
+        if (raw == null || !(raw.get("fleet") instanceof Map<?, ?> fleet)
+                || !(fleet.get("leaders") instanceof Map<?, ?> leaders)) {
+            return;
+        }
+        List<String> bad = leaders.entrySet().stream()
+                .filter(e -> e.getValue() instanceof Map<?, ?> leader && leader.containsKey("terminal"))
+                .map(e -> String.valueOf(e.getKey()))
+                .sorted()
+                .toList();
+        if (!bad.isEmpty()) {
+            throw new IllegalStateException("refusing to start: fleet.leaders entries ["
+                    + String.join(", ", bad) + "] still use the retired 'terminal:' key — replace it "
+                    + "with 'tab:', the exact tab label hosting the lead. A terminal_id changes on "
+                    + "every restart of the lead's session; a tab label does not.");
+        }
+    }
+
     static List<String> unknownTopLevelKeys(String yaml) {
         Map<?, ?> raw;
         try {
@@ -1082,7 +1158,7 @@ public record BridgedConfig(
         // defaults the fields of a block that IS present. Defaulting it here would start watching
         // the file for every config that never asked to be watched.
         return new BridgedConfig(b, herdrSocket, profiles, g, worktreeRoot, l, timeout, pollMs,
-                broker, primary, f, leadHeartbeat, placementOrDefault, a, configReload);
+                broker, primary, f, leadHeartbeat, health, placementOrDefault, a, configReload);
     }
 
     /**
@@ -1292,10 +1368,9 @@ public record BridgedConfig(
                         + "', which is not a configured profiles: entry (have: " + profiles.keySet()
                         + ").");
             }
-            if (!leader.isCreatable() && (leader.terminal() == null || leader.terminal().isBlank())) {
-                bad.add("fleet.leaders." + name + " can neither be found nor created — it pins no "
-                        + "terminal: and names no profile: to launch one on. Give it one or the "
-                        + "other, or drop the entry.");
+            if (leader.tab() == null || leader.tab().isBlank()) {
+                bad.add("fleet.leaders." + name + " has no tab: — a lead is now found (and, if "
+                        + "auto-launched, labelled) purely by its tab, so every entry must name one.");
             }
         });
         if (!bad.isEmpty()) {

bridged/src/main/java/dev/ltms/bridged/config/ConfigRef.java

@@ -30,7 +30,11 @@ import java.util.function.Supplier;
  *       {@code fleet:} (every role pool, {@code charters}, and {@code tabLabel}),
  *       {@code placement:}, and an existing profile's {@code weight} / {@code maxLoad}. Those
  *       three are read through a supplier on {@code CompositePeerLauncher}, which is what makes
- *       them hot — not the fact that they are config.</li>
+ *       them hot — not the fact that they are config. <strong>This does NOT include
+ *       {@code fleet.leaders}</strong>: {@code Bridged.main} reads {@code cfg.fleet().leaders()}
+ *       once at startup to build the {@code LeadTabScanner} and the {@code LeadLauncher}, and
+ *       neither is reconstructed on reload — so a lead added, removed, or re-{@code tab}'d under
+ *       {@code fleet.leaders} needs a restart, the same as any deferred key below.</li>
  *   <li><strong>Deferred</strong> — accepted into the new snapshot, but the wiring built at startup
  *       keeps the old value until a restart: {@code lifecycle:}, {@code leadHeartbeat:},
  *       {@code spawnReadyTimeoutMs} / {@code spawnReadyPollMs}, {@code guard:},

bridged/src/main/java/dev/ltms/bridged/health/FleetHealth.java

@@ -0,0 +1,40 @@
+package dev.ltms.bridged.health;
+
+import dev.ltms.bridged.herdr.AgentStatus;
+import dev.ltms.bridged.session.MemberSession;
+
+/**
+ * Pure classifier. Collection and repair are deliberately outside this package.
+ * {@link HealthState#ERROR_ON_SCREEN} is not decided yet because it needs a bounded pane detection
+ * read and an adapter-specific fatal signature; status facts alone must not guess it.
+ */
+public final class FleetHealth {
+    private FleetHealth() { }
+
+    public static HealthDecision decide(HealthSnapshot s, HealthPrior prior, long nowNanos) {
+        if (s.controlLinkDown()) return result(HealthState.CONTROL_LINK_DOWN, false);
+        if (s.targetNotFound()) return result(HealthState.GONE, false);
+        if (s.sessionState() == MemberSession.State.SPAWNING && !s.present() && s.readinessGraceElapsed()) {
+            return result(HealthState.NEVER_READY, false);
+        }
+        if (s.orphanedDelegation()) return result(HealthState.DELEGATION_ORPHANED, false);
+        boolean disagreement = s.sessionState() == MemberSession.State.BUSY && s.acceptedDelivery()
+                && (s.liveStatus() == AgentStatus.IDLE || s.liveStatus() == AgentStatus.DONE);
+        if (disagreement && prior.busyButDone()) return result(HealthState.TURN_BOUNDARY_LOST, true);
+        if (s.stalled()) return result(HealthState.STALL_SUSPECTED, disagreement);
+        if (s.replyStranded()) return result(HealthState.REPLY_STRANDED, disagreement);
+        if (s.queuedDelivery() || s.inboxMessage()) return result(HealthState.WORK_PENDING, disagreement);
+        if (s.sessionState() == MemberSession.State.SPAWNING) return result(HealthState.STARTING, disagreement);
+        if (s.acceptedDelivery() && s.liveStatus() == AgentStatus.BLOCKED) {
+            return result(HealthState.BLOCKED_AMBIGUOUS, disagreement);
+        }
+        // An accepted delivery remains bridge work even when herdr is late, unknown, or has already
+        // reported DONE once. It cannot be IDLE until the delegation has resolved.
+        if (s.acceptedDelivery()) return result(HealthState.WORKING, disagreement);
+        return result(HealthState.IDLE, disagreement);
+    }
+
+    private static HealthDecision result(HealthState state, boolean disagreement) {
+        return new HealthDecision(state, new HealthPrior(disagreement));
+    }
+}

bridged/src/main/java/dev/ltms/bridged/health/FleetHealthMonitor.java

@@ -0,0 +1,151 @@
+package dev.ltms.bridged.health;
+
+import dev.ltms.bridged.herdr.Agent;
+import dev.ltms.bridged.herdr.AgentControl;
+import dev.ltms.bridged.herdr.AgentStatus;
+import dev.ltms.bridged.msg.MessageService;
+import dev.ltms.bridged.session.MemberSession;
+import org.slf4j.Logger;
+import org.slf4j.LoggerFactory;
+
+import java.util.HashMap;
+import java.util.HashSet;
+import java.util.List;
+import java.util.Map;
+import java.util.Objects;
+import java.util.concurrent.ScheduledExecutorService;
+import java.util.concurrent.TimeUnit;
+import java.util.function.BiConsumer;
+import java.util.function.LongSupplier;
+import java.util.function.Supplier;
+
+/** Slow whole-fleet evidence collection. It is deliberately separate from the delivery poller. */
+public final class FleetHealthMonitor {
+    private static final Logger log = LoggerFactory.getLogger(FleetHealthMonitor.class);
+
+    /** Bounded attempts to run {@link #failTarget} for one transition. Never retried tick-to-tick (CB-580). */
+    static final int MAX_FAIL_TARGET_ATTEMPTS = 3;
+
+    private final AgentControl agents;
+    private final Supplier<List<MemberSession>> roster;
+    private final MessageService messages;
+    private final ScheduledExecutorService scheduler;
+    private final LongSupplier clock;
+    private final long intervalSeconds;
+    private final BiConsumer<String, String> failTarget;
+    private final Map<String, HealthPrior> priors = new HashMap<>();
+    private final Map<String, HealthState> states = new HashMap<>();
+
+    // These facts need the evidence publishers introduced by later M4 units. They are not negatives.
+    private static final boolean NOT_YET_OBSERVED = false;
+
+    /**
+     * @param failTarget CB-568's idempotent target-wide failure operation (e.g. {@code messages::abandon}),
+     *                   invoked once when a member transitions into a terminal health state. Required —
+     *                   there is deliberately no defaulting overload; a caller that does not want the
+     *                   fail-tickets-on-terminal-health behavior must pass an explicit inert value (see
+     *                   {@code TestTurnTokens.inert} / {@code BridgeMcp.CapacitySource.none()} for the pattern).
+     */
+    public FleetHealthMonitor(AgentControl agents, Supplier<List<MemberSession>> roster, MessageService messages,
+                              ScheduledExecutorService scheduler, LongSupplier clock, long intervalSeconds,
+                              BiConsumer<String, String> failTarget) {
+        this.agents = agents;
+        this.roster = roster;
+        this.messages = messages;
+        this.scheduler = scheduler;
+        this.clock = clock;
+        this.intervalSeconds = intervalSeconds;
+        this.failTarget = Objects.requireNonNull(failTarget, "failTarget");
+    }
+
+    /** Pure per-member decision seam. */
+    static HealthDecision decide(HealthSnapshot snapshot, HealthPrior prior, long nowNanos) {
+        return FleetHealth.decide(snapshot, prior, nowNanos);
+    }
+
+    public void start() { scheduler.schedule(this::tick, intervalSeconds, TimeUnit.SECONDS); }
+    public void stop() { scheduler.shutdownNow(); }
+
+    // Package-private so tests can run one tick without waiting.
+    void tick() {
+        try {
+            List<Agent> agentsNow = agents.list(); // Exactly one list call for this complete observation.
+            List<MemberSession> rosterNow = roster.get(); // One in-memory roster snapshot for this tick.
+            Map<String, Agent> live = new HashMap<>();
+            for (Agent agent : agentsNow) live.put(agent.terminalId(), agent);
+            HashSet<String> current = new HashSet<>();
+            for (MemberSession session : rosterNow) {
+                current.add(session.terminalId());
+                Agent agent = live.get(session.terminalId());
+                AgentStatus status = agent == null ? AgentStatus.UNKNOWN : agent.status();
+                boolean accepted = messages.hasAcceptedDelivery(session.terminalId());
+                HealthSnapshot snapshot = new HealthSnapshot(session.state(), status, accepted, NOT_YET_OBSERVED,
+                        messages.hasInboxMessage(session.terminalId()), agent != null, NOT_YET_OBSERVED,
+                        NOT_YET_OBSERVED, NOT_YET_OBSERVED, NOT_YET_OBSERVED, NOT_YET_OBSERVED, NOT_YET_OBSERVED);
+                HealthDecision decision = decide(snapshot, priors.getOrDefault(session.terminalId(), HealthPrior.NONE),
+                        clock.getAsLong());
+                priors.put(session.terminalId(), decision.prior());
+                reportTransition(session.terminalId(), decision.state());
+            }
+            priors.keySet().retainAll(current);
+            states.keySet().retainAll(current);
+        } catch (Throwable error) {
+            // A list failure is health evidence, and must never kill the monitor's only scheduler task.
+            log.warn("fleet health collection failed; will retry next tick", error);
+        } finally {
+            if (!scheduler.isShutdown()) {
+                scheduler.schedule(this::tick, intervalSeconds, TimeUnit.SECONDS);
+            }
+        }
+    }
+
+    void reportTransition(String target, HealthState next) {
+        HealthState previous = states.put(target, next);
+        if (previous == next) return;
+        if (fault(next)) {
+            log.warn("fleet health member={} state={} previous={}", target, next, previous);
+        } else if (previous != null && fault(previous)) {
+            log.info("fleet health member={} recovered state={} previous={}", target, next, previous);
+        }
+        // CB-580: a member entering GONE/NEVER_READY must not leave its waiting tickets pending
+        // forever. Fire exactly once per transition — never on a tick where the state is unchanged,
+        // which is what made the rejected commit call abandon() once per tick for as long as a
+        // member stayed terminal.
+        if (terminal(next)) {
+            failTerminalTarget(target, next);
+        }
+    }
+
+    private void failTerminalTarget(String target, HealthState state) {
+        String reason = "fleet health: member reached terminal state " + state.name();
+        RuntimeException last = null;
+        for (int attempt = 1; attempt <= MAX_FAIL_TARGET_ATTEMPTS; attempt++) {
+            try {
+                failTarget.accept(target, reason);
+                return;
+            } catch (RuntimeException error) {
+                last = error;
+                log.warn("fleet health: failTarget attempt {}/{} failed for member={} state={}",
+                        attempt, MAX_FAIL_TARGET_ATTEMPTS, target, state, error);
+            }
+        }
+        log.warn("fleet health: giving up on failTarget for member={} state={} after {} attempts",
+                target, state, MAX_FAIL_TARGET_ATTEMPTS, last);
+    }
+
+    private static boolean terminal(HealthState state) {
+        return state == HealthState.GONE || state == HealthState.NEVER_READY;
+    }
+
+    private static boolean fault(HealthState state) {
+        return switch (state) {
+            case NEVER_READY, GONE, TURN_BOUNDARY_LOST, ERROR_ON_SCREEN, STALL_SUSPECTED,
+                    MUTE, REPLY_STRANDED, DELEGATION_ORPHANED, CONTROL_LINK_DOWN -> true;
+            default -> false;
+        };
+    }
+
+    public static String coverage(boolean enabled, boolean notificationConfigured) {
+        return !enabled ? "off" : notificationConfigured ? "full" : "detection-only";
+    }
+}

bridged/src/main/java/dev/ltms/bridged/health/HealthDecision.java

@@ -0,0 +1,4 @@
+package dev.ltms.bridged.health;
+
+/** Classification plus the private fact that the next pure decision needs. */
+public record HealthDecision(HealthState state, HealthPrior prior) { }

bridged/src/main/java/dev/ltms/bridged/health/HealthPrior.java

@@ -0,0 +1,6 @@
+package dev.ltms.bridged.health;
+
+/** Private cross-tick observation. It is deliberately not a reported health value. */
+public record HealthPrior(boolean busyButDone) {
+    public static final HealthPrior NONE = new HealthPrior(false);
+}

bridged/src/main/java/dev/ltms/bridged/health/HealthSnapshot.java

@@ -0,0 +1,11 @@
+package dev.ltms.bridged.health;
+
+import dev.ltms.bridged.herdr.AgentStatus;
+import dev.ltms.bridged.session.MemberSession;
+
+/** Read-only facts from one fleet collection tick. */
+public record HealthSnapshot(MemberSession.State sessionState, AgentStatus liveStatus,
+                             boolean acceptedDelivery, boolean queuedDelivery, boolean inboxMessage,
+                             boolean present, boolean targetNotFound, boolean controlLinkDown,
+                             boolean readinessGraceElapsed, boolean orphanedDelegation,
+                             boolean replyStranded, boolean stalled) { }

bridged/src/main/java/dev/ltms/bridged/health/HealthState.java

@@ -0,0 +1,8 @@
+package dev.ltms.bridged.health;
+
+/** Health classifications reported for a member. */
+public enum HealthState {
+    STARTING, IDLE, WORKING, WORK_PENDING, BLOCKED_AMBIGUOUS,
+    NEVER_READY, GONE, TURN_BOUNDARY_LOST, ERROR_ON_SCREEN, STALL_SUSPECTED,
+    MUTE, REPLY_STRANDED, DELEGATION_ORPHANED, CONTROL_LINK_DOWN
+}

bridged/src/main/java/dev/ltms/bridged/health/MuteCounter.java

@@ -0,0 +1,25 @@
+package dev.ltms.bridged.health;
+
+import dev.ltms.bridged.msg.Rendezvous;
+
+import java.util.Map;
+import java.util.concurrent.ConcurrentHashMap;
+
+/**
+ * Counts turns that ended via the completion fallback instead of {@code bridge_reply}.
+ * MUTE is an observation by target and profile, not a classifier state and never suppresses faults.
+ */
+public final class MuteCounter {
+    private final Map<String, Integer> byTarget = new ConcurrentHashMap<>();
+    private final Map<String, Integer> byProfile = new ConcurrentHashMap<>();
+
+    /** Record only fallback completion; a structured reply does not make a member mute. */
+    public void observe(String target, String profile, Rendezvous.Kind kind) {
+        if (kind != Rendezvous.Kind.COMPLETION) return;
+        byTarget.merge(target, 1, Integer::sum);
+        byProfile.merge(profile, 1, Integer::sum);
+    }
+
+    public int forTarget(String target) { return byTarget.getOrDefault(target, 0); }
+    public int forProfile(String profile) { return byProfile.getOrDefault(profile, 0); }
+}

bridged/src/main/java/dev/ltms/bridged/health/PaneBudget.java

@@ -0,0 +1,26 @@
+package dev.ltms.bridged.health;
+
+import java.util.ArrayList;
+import java.util.HashMap;
+import java.util.List;
+import java.util.Map;
+
+/** Fixed pane-probe limits. Pane content is never retained here. */
+public final class PaneBudget {
+    public static final long COOLDOWN_NANOS = 60_000_000_000L;
+    public static final int MAX_PER_TICK = 2;
+    private final Map<String, Long> lastProbe = new HashMap<>();
+    private int cursor;
+
+    public List<String> choose(List<String> candidates, long nowNanos, long configuredCooldownNanos) {
+        long cooldown = Math.max(COOLDOWN_NANOS, configuredCooldownNanos);
+        List<String> out = new ArrayList<>();
+        for (int n = 0; n < candidates.size() && out.size() < MAX_PER_TICK; n++) {
+            String target = candidates.get((cursor + n) % candidates.size());
+            Long last = lastProbe.get(target);
+            if (last == null || nowNanos - last >= cooldown) { out.add(target); lastProbe.put(target, nowNanos); }
+        }
+        if (!candidates.isEmpty()) cursor = (cursor + 1) % candidates.size();
+        return List.copyOf(out);
+    }
+}

bridged/src/main/java/dev/ltms/bridged/herdr/LeadTabScanner.java

@@ -6,6 +6,7 @@ import org.slf4j.LoggerFactory;
 
 import java.util.Collections;
 import java.util.LinkedHashMap;
+import java.util.Locale;
 import java.util.Map;
 import java.util.Set;
 import java.util.function.LongSupplier;
@@ -23,6 +24,15 @@ import java.util.function.Supplier;
  * by first starting the session and asking it. Scanning closes that loop: label the tab, and the
  * pane is recognised on the next resolve.
  *
+ * <p><strong>CB-579 — matched by name, not prefix.</strong> This used to strip one shared
+ * {@code tabPrefix} off a label to derive the lead's name, and merged a config-supplied
+ * {@code terminal_id} pin over every scan result so the pin could never expire. Both are gone: each
+ * lead now configures its own exact {@code tab} label ({@code fleet.leaders.<name>.tab}), so this
+ * class is handed a {@code tab → name} map up front and matches labels against it exactly
+ * (case-insensitively). There is no merge step — a scan result is the whole answer. That is the
+ * fix for the bug this replaces: a {@code terminal_id} pin surviving in config after the pane it
+ * named was gone, so the daemon kept treating a dead session as a live lead forever.
+ *
  * <p><strong>Direction of trust.</strong> The label names the lead; it never <em>grants</em>
  * anything a pane could take for itself. Three properties keep that honest:
  * <ol>
@@ -48,7 +58,7 @@ import java.util.function.Supplier;
  * ever make a decision that <em>removes</em> something based on this map, add the same check.
  * The remaining hazard is an <em>operator</em> one — a worker {@code tabLabel} template that
  * happens to start with the same prefix would promote the whole fleet — and that is refused at
- * startup by {@code BridgedConfig.validateLeadScan} rather than documented here.
+ * startup by {@code BridgedConfig.validateLeadTabPrefixes} rather than documented here.
  *
  * <p><strong>Caching.</strong> {@link #get()} is on the request path (every resolve), so the scan
  * is TTL-cached and a stale-but-valid map is preferred to a herdr round-trip. A failed scan keeps
@@ -60,38 +70,46 @@ public final class LeadTabScanner implements Supplier<Map<String, String>> {
     private static final Logger log = LoggerFactory.getLogger(LeadTabScanner.class);
 
     private final HerdrClient herdr;
-    private final String tabPrefix;
+    private final Map<String, String> tabToName;
     private final Set<String> excludedWorkspaceLabels;
-    private final Map<String, String> configuredLeads;
     private final long ttlNanos;
     private final LongSupplier clock;
 
-    private Map<String, String> cached;
+    private Map<String, String> cached = Map.of();
     private long scannedAtNanos;
     private boolean everScanned;
 
     /**
      * @param herdr                   the herdr client to query ({@code workspace.list},
      *                                {@code tab.list}, {@code pane.list} — all read-only)
-     * @param tabPrefix               a tab whose label starts with this (case-insensitively) hosts a
-     *                                lead; the rest of the label, trimmed, is the lead's name
+     * @param tabToName               every configured lead's exact tab label → its name
+     *                                ({@code fleet.leaders.<name>.tab}), matched case-insensitively
      * @param excludedWorkspaceLabels workspaces never scanned — the configured worker spaces
-     * @param configuredLeads         the static {@code leaders:}/{@code primary:} registry, merged
-     *                                over every scan result. Explicit config outranks the
-     *                                convention, and survives a scan that cannot run at all
      * @param ttlNanos                how long a scan result is reused before the next one
      * @param clock                   nanosecond time source ({@code System::nanoTime} in production)
      */
-    public LeadTabScanner(HerdrClient herdr, String tabPrefix, Set<String> excludedWorkspaceLabels,
-                          Map<String, String> configuredLeads, long ttlNanos, LongSupplier clock) {
+    public LeadTabScanner(HerdrClient herdr, Map<String, String> tabToName,
+                          Set<String> excludedWorkspaceLabels, long ttlNanos, LongSupplier clock) {
         this.herdr = herdr;
-        this.tabPrefix = tabPrefix == null || tabPrefix.isBlank() ? "lead:" : tabPrefix.strip();
+        this.tabToName = normalize(tabToName);
         this.excludedWorkspaceLabels = excludedWorkspaceLabels == null
                 ? Set.of() : Set.copyOf(excludedWorkspaceLabels);
-        this.configuredLeads = configuredLeads == null ? Map.of() : Map.copyOf(configuredLeads);
         this.ttlNanos = ttlNanos;
         this.clock = clock;
-        this.cached = this.configuredLeads;
+    }
+
+    /** Keys stripped and lower-cased once, so every lookup is a plain map hit. */
+    private static Map<String, String> normalize(Map<String, String> tabToName) {
+        if (tabToName == null || tabToName.isEmpty()) {
+            return Map.of();
+        }
+        Map<String, String> out = new LinkedHashMap<>();
+        tabToName.forEach((tab, name) -> {
+            if (tab != null && !tab.isBlank() && name != null && !name.isBlank()) {
+                out.put(tab.strip().toLowerCase(Locale.ROOT), name);
+            }
+        });
+        return Collections.unmodifiableMap(out);
     }
 
     /**
@@ -151,25 +169,20 @@ public final class LeadTabScanner implements Supplier<Map<String, String>> {
                 }
             }
         }
-        byTerminal.putAll(configuredLeads); // an explicit pin outranks a label
         return Collections.unmodifiableMap(byTerminal);
     }
 
     /**
-     * The lead name a tab label declares, or {@code null} if it declares none.
+     * The lead name a tab label declares, or {@code null} if it names none of the configured leads.
      *
-     * <p>{@code "lead: opus-5.0"} → {@code "opus-5.0"}. A bare {@code "lead:"} names nobody and is
-     * rejected: an unnamed lead would resolve as {@code PRIMARY} with nothing to attribute it to.
+     * <p>Exact match (case-insensitive, ends stripped) against {@link #tabToName} — no prefix
+     * stripping, so an operator's {@code "lead: something-else"} tab is never mistaken for a
+     * configured lead just because it shares a prefix.
      */
     private String leadNameOf(String label) {
         if (label == null) {
             return null;
         }
-        String l = label.strip();
-        if (!l.regionMatches(true, 0, tabPrefix, 0, tabPrefix.length())) {
-            return null;
-        }
-        String name = l.substring(tabPrefix.length()).strip();
-        return name.isEmpty() ? null : name;
+        return tabToName.get(label.strip().toLowerCase(Locale.ROOT));
     }
 }

bridged/src/main/java/dev/ltms/bridged/inject/CompletionResolver.java

@@ -2,11 +2,17 @@ package dev.ltms.bridged.inject;
 
 import dev.ltms.bridged.herdr.AgentControl;
 import dev.ltms.bridged.msg.Rendezvous;
+import dev.ltms.bridged.msg.TurnToken;
 import org.slf4j.Logger;
 import org.slf4j.LoggerFactory;
 
+import java.util.List;
+import java.util.Objects;
+import java.util.Set;
+import java.util.TreeSet;
 import java.util.concurrent.CompletableFuture;
 import java.util.concurrent.ConcurrentHashMap;
+import java.util.regex.Pattern;
 
 /**
  * The CB-106 completion fallback: bridges the {@link Injector}'s turn-completion signal to the
@@ -60,6 +66,7 @@ public final class CompletionResolver implements TurnListener {
 
     private final AgentControl agents;
     private final Rendezvous rendezvous;
+    private final ExhaustedPatternLookup exhaustedPatterns;
 
     /**
      * Per-target record of the turn currently in flight: the exact {@link Rendezvous} waiter its
@@ -77,23 +84,30 @@ public final class CompletionResolver implements TurnListener {
 
     private final ConcurrentHashMap<String, InFlight> inFlight = new ConcurrentHashMap<>();
 
-    public CompletionResolver(AgentControl agents, Rendezvous rendezvous) {
+    /**
+     * @param exhaustedPatterns CB-578 stage A: per-target lookup for a profile's configured
+     *                          usage-limit refusal pattern. Required — there is deliberately no
+     *                          defaulting overload; a caller that does not want the classification
+     *                          must pass an explicit inert value ({@link ExhaustedPatternLookup#none()}).
+     */
+    public CompletionResolver(AgentControl agents, Rendezvous rendezvous, ExhaustedPatternLookup exhaustedPatterns) {
         this.agents = agents;
         this.rendezvous = rendezvous;
+        this.exhaustedPatterns = Objects.requireNonNull(exhaustedPatterns, "exhaustedPatterns");
     }
 
     @Override
-    public void onDelivered(String target) {
+    public void onDelivered(String target, TurnToken token) {
         // Capture the exact waiter this turn belongs to (CB-116) and snapshot the pane's pre-turn
         // content — what it shows *before* the just-delivered turn produces output — as the staleness
         // reference (CB-115). Done synchronously (like the delivering send itself) so both are in
         // place before this turn's completion can fire.
-        captureBaseline(target);
+        captureBaseline(target, token);
     }
 
     /** Capture the in-flight turn: its waiter and pre-turn baseline (the testable core of {@link #onDelivered}). */
-    void captureBaseline(String target) {
-        CompletableFuture<Rendezvous.Resolution> waiter = rendezvous.currentWaiter(target);
+    void captureBaseline(String target, TurnToken token) {
+        CompletableFuture<Rendezvous.Resolution> waiter = token.waiter();
         if (waiter == null) {
             inFlight.remove(target); // no send is waiting on this delivery — nothing to resolve later
             return;
@@ -137,7 +151,13 @@ public final class CompletionResolver implements TurnListener {
     @Override
     public void onTurnFailed(String target) {
         InFlight turn = inFlight.get(target);
-        Thread.ofVirtual().name("turn-failed-" + target).start(() -> fail(target, turn));
+        Thread.ofVirtual().name("turn-failed-" + target).start(() -> fail(target, turn, null));
+    }
+
+    @Override
+    public void onTurnFailed(String target, String reason) {
+        InFlight turn = inFlight.get(target);
+        Thread.ofVirtual().name("turn-failed-" + target).start(() -> fail(target, turn, reason));
     }
 
     /** Synchronous resolve (the unit-testable core of {@link #onTurnComplete}). */
@@ -150,11 +170,12 @@ public final class CompletionResolver implements TurnListener {
             return;
         }
         String tail;
+        String assistantBlock = null;
         int originalLength = 0;
         boolean clipped = false;
         boolean scrapeFailed = false;
         try {
-            String assistantBlock = lastAssistantBlock(agents.read(target, SCRAPE_SOURCE));
+            assistantBlock = lastAssistantBlock(agents.read(target, SCRAPE_SOURCE));
             originalLength = assistantBlock.strip().length();
             clipped = originalLength > MAX_SCRAPE_CHARS;
             tail = clip(assistantBlock);
@@ -177,6 +198,22 @@ public final class CompletionResolver implements TurnListener {
                     target);
             return; // keep the in-flight record: a later genuine completion still needs it
         }
+        // CB-578 stage A: a turn that ended with no bridge_reply AND whose scrape matches the
+        // backend's configured usage-limit pattern is a refusal, not an answer. Classify it as
+        // BACKEND_EXHAUSTED rather than handing the caller a scrape that reads like a real reply.
+        if (!scrapeFailed) {
+            Pattern exhausted = exhaustedPatterns.patternFor(target);
+            String matchedLine = exhausted == null ? null : firstMatchingLine(assistantBlock, exhausted);
+            if (matchedLine != null) {
+                String reason = "backend exhausted (usage limit): " + matchedLine;
+                if (rendezvous.resolveExhausted(waiter, reason)) {
+                    inFlight.remove(target, turn);
+                    log.warn("completion for {} classified BACKEND_EXHAUSTED (no bridge_reply; scrape "
+                            + "matched the profile's exhausted pattern): {}", target, reason);
+                }
+                return;
+            }
+        }
         String completion = clipped ? tail + "\n" + CLIPPED_PANE_TAIL_MARKER : tail;
         if (rendezvous.resolveCompletion(waiter, completion)) {
             inFlight.remove(target, turn);
@@ -192,6 +229,11 @@ public final class CompletionResolver implements TurnListener {
 
     /** Synchronous fail (the unit-testable core of {@link #onTurnFailed}). */
     void fail(String target, InFlight turn) {
+        fail(target, turn, null);
+    }
+
+    /** Synchronous fail with an optional reason supplied by a dropped worker queue. */
+    void fail(String target, InFlight turn, String explicitReason) {
         // A never-delivered readiness failure has no in-flight record but still has a blocked send;
         // fall back to the currently-registered waiter (unambiguous — that send never completed, so
         // no next turn exists to confuse it with).
@@ -201,16 +243,18 @@ public final class CompletionResolver implements TurnListener {
             inFlight.remove(target, turn); // nobody blocked on this worker — nothing to fail
             return;
         }
-        String reason;
-        try {
-            reason = clip(agents.read(target, SCRAPE_SOURCE));
-        } catch (RuntimeException e) {
-            reason = "";
-        }
-        if (reason.isBlank()) {
-            // No screen to scrape — either the worker is stuck (CB-109) or gone (CB-110).
-            reason = "worker did not reply; its turn ended in an unrecoverable state "
-                    + "(worker unreachable or stuck)";
+        String reason = explicitReason;
+        if (reason == null || reason.isBlank()) {
+            try {
+                reason = clip(agents.read(target, SCRAPE_SOURCE));
+            } catch (RuntimeException e) {
+                reason = "";
+            }
+            if (reason.isBlank()) {
+                // No screen to scrape — either the worker is stuck (CB-109) or gone (CB-110).
+                reason = "worker did not reply; its turn ended in an unrecoverable state "
+                        + "(worker unreachable or stuck)";
+            }
         }
         if (rendezvous.resolveFailure(waiter, reason)) {
             inFlight.remove(target, turn);
@@ -218,6 +262,45 @@ public final class CompletionResolver implements TurnListener {
         }
     }
 
+    /**
+     * The first line of {@code text} matching {@code pattern}, stripped — the CB-578 stage A
+     * evidence carried in a {@code BACKEND_EXHAUSTED} reason so the operator sees the real refusal
+     * text, never a generic label. {@code null} if no line matches.
+     */
+    static String firstMatchingLine(String text, Pattern pattern) {
+        if (text == null || text.isEmpty()) return null;
+        for (String line : text.split("\n", -1)) {
+            if (pattern.matcher(line).find()) {
+                return line.strip();
+            }
+        }
+        return null;
+    }
+
+    /**
+     * Coverage summary for the CB-578 stage A exhausted-pattern classification, logged at startup
+     * the way {@link dev.ltms.bridged.health.FleetHealthMonitor#coverage} is — so an operator can
+     * see whether the classification is on, and for which profiles, without reading every
+     * profile's config by hand.
+     *
+     * @param allProfiles        every configured profile name
+     * @param configuredProfiles the subset of {@code allProfiles} that carry an exhausted pattern
+     */
+    public static String coverage(Set<String> allProfiles, Set<String> configuredProfiles) {
+        if (configuredProfiles.isEmpty()) {
+            return "off (no profile has an exhaustedPattern configured; profiles: " + sorted(allProfiles) + ")";
+        }
+        Set<String> unconfigured = new TreeSet<>(allProfiles);
+        unconfigured.removeAll(configuredProfiles);
+        return unconfigured.isEmpty()
+                ? "full (all profiles configured: " + sorted(allProfiles) + ")"
+                : "partial (configured: " + sorted(configuredProfiles) + "; not configured: " + sorted(unconfigured) + ")";
+    }
+
+    private static List<String> sorted(Set<String> names) {
+        return names.stream().sorted().toList();
+    }
+
     private static String clip(String s) {
         if (s == null) return "";
         String trimmed = s.strip();

bridged/src/main/java/dev/ltms/bridged/inject/ExhaustedPatternLookup.java

@@ -0,0 +1,27 @@
+package dev.ltms.bridged.inject;
+
+import java.util.regex.Pattern;
+
+/**
+ * Per-target lookup for a profile's configured usage-limit refusal pattern (CB-578 stage A): how
+ * {@link CompletionResolver} tells a backend that refused on a subscription usage limit — the
+ * worker's pane stays healthy, but the account is exhausted — apart from a genuine completion.
+ *
+ * <p>The pattern is always profile config, never a vendor string in Java source: every backend
+ * words its refusal differently, so a hardcoded sentence would only ever match one of them.
+ */
+@FunctionalInterface
+public interface ExhaustedPatternLookup {
+
+    /** The compiled pattern configured for {@code target}'s profile, or {@code null} if none. */
+    Pattern patternFor(String target);
+
+    /**
+     * Inert lookup — no profile has a pattern configured, so the classification never fires and
+     * the completion fallback behaves exactly as before CB-578 stage A. The explicit stand-in a
+     * caller (or a test not exercising this feature) passes instead of a defaulting overload.
+     */
+    static ExhaustedPatternLookup none() {
+        return target -> null;
+    }
+}

bridged/src/main/java/dev/ltms/bridged/inject/Injector.java

@@ -2,6 +2,7 @@ package dev.ltms.bridged.inject;
 
 import dev.ltms.bridged.herdr.AgentControl;
 import dev.ltms.bridged.herdr.AgentStatus;
+import dev.ltms.bridged.msg.TurnToken;
 import org.slf4j.Logger;
 import org.slf4j.LoggerFactory;
 
@@ -129,7 +130,7 @@ public final class Injector {
     }
 
     /** A pending message and the future that completes when it has been delivered. */
-    private record Pending(String text, CompletableFuture<Void> delivered) {
+    private record Pending(String text, TurnToken token, CompletableFuture<Void> delivered) {
     }
 
     /** Per-worker delivery state, guarded by its own monitor (single writer per worker). */
@@ -159,9 +160,9 @@ public final class Injector {
      * <p>Uses an atomic map update so a concurrent {@link #drop} cannot slip between "find the
      * target" and "queue the message" and orphan it in a target it just removed.
      */
-    public CompletableFuture<Void> enqueue(String target, String text) {
+    public CompletableFuture<Void> enqueue(String target, String text, TurnToken token) {
         CompletableFuture<Void> delivered = new CompletableFuture<>();
-        Pending p = new Pending(text, delivered);
+        Pending p = new Pending(text, token, delivered);
         targets.compute(target, (_, existing) -> {
             Target t = (existing != null) ? existing : new Target();
             t.add(p); // synchronized on the Target monitor — atomic with a concurrent drop
@@ -356,7 +357,7 @@ public final class Injector {
             } else {
                 // Baseline the pane's pre-turn content so a misattributed completion (no new output)
                 // can't resolve this send with the previous turn's stale answer (CB-115).
-                turnListener.onDelivered(target);
+                turnListener.onDelivered(target, sent.token());
                 sent.delivered().complete(null);
             }
         }
@@ -410,8 +411,8 @@ public final class Injector {
         for (Pending p : pending) {
             p.delivered().completeExceptionally(cause);
         }
-        if (hadDeliveredTurn) {
-            turnListener.onTurnFailed(target);
-        }
+        // A queued send has no in-flight record, while a delivered turn does. CompletionResolver
+        // handles both forms and resolves its waiter at most once.
+        turnListener.onTurnFailed(target, cause.getMessage());
     }
 }

bridged/src/main/java/dev/ltms/bridged/inject/TurnListener.java

@@ -1,5 +1,7 @@
 package dev.ltms.bridged.inject;
 
+import dev.ltms.bridged.msg.TurnToken;
+
 /**
  * Notified when a worker's delegated turn is observed to complete — a confirmed
  * {@code WORKING → IDLE} transition after a delivery. This is the CB-106 completion signal the
@@ -41,6 +43,14 @@ public interface TurnListener {
     default void onTurnFailed(String target) {
     }
 
+    /**
+     * As {@link #onTurnFailed(String)}, carrying the reason a worker became unreachable. The default
+     * keeps existing listeners working while allowing the completion resolver to report a useful cause.
+     */
+    default void onTurnFailed(String target, String reason) {
+        onTurnFailed(target);
+    }
+
     /**
      * A message was just delivered into {@code target}'s pane (CB-115). Fired so the completion
      * resolver can snapshot the pane's pre-turn content: a later {@link #onTurnComplete} whose
@@ -49,7 +59,7 @@ public interface TurnListener {
      * resolve the send with the previous turn's stale answer. A default no-op keeps the interface
      * functional for callers that don't scrape.
      */
-    default void onDelivered(String target) {
+    default void onDelivered(String target, TurnToken token) {
     }
 
     /** No-op default for callers that only need delivery, not completion signalling. */

bridged/src/main/java/dev/ltms/bridged/lead/LeadLauncher.java

@@ -59,7 +59,7 @@ public final class LeadLauncher {
     /**
      * @param agents herdr agent control (start, list)
      * @param spaces workspace / tab control (ensure, create, label, list)
-     * @param cfg    the loaded config — {@code fleet.leaders}, {@code profiles} and the lead pins
+     * @param cfg    the loaded config — {@code fleet.leaders}, {@code profiles} and each lead's tab
      */
     public LeadLauncher(AgentControl agents, WorkspaceControl spaces, BridgedConfig cfg) {
         this.agents = agents;
@@ -102,8 +102,8 @@ public final class LeadLauncher {
                 continue;
             }
             if (!lead.isCreatable()) {
-                // A lead with a `terminal:` pin and no `profile:` is recognise-only by design: the
-                // operator opens it by hand. Say so once rather than looking like a silent failure.
+                // A lead with a `tab:` but no `profile:` is recognise-only by design: the operator
+                // opens it by hand. Say so once rather than looking like a silent failure.
                 log.info("lead '{}' is not live, and names no profile — it can be recognised but not "
                         + "launched. Add `profile:` under fleet.leaders.{} to have bridged start it.",
                         name, name);
@@ -127,18 +127,16 @@ public final class LeadLauncher {
     }
 
     /**
-     * How many live leads exist per configured name.
+     * How many live leads exist per configured name: a running agent in a tab labelled with that
+     * lead's exact {@code tab} (CB-579). Member workspaces are excluded, exactly as the scanner
+     * excludes them: a member must not be counted as a lead because it happens to sit in a matching
+     * tab.
      *
-     * <p>Two independent pieces of evidence, because either alone double-spawns:
-     * <ul>
-     *   <li>a running agent in a tab labelled {@code "<tabPrefix> <name>"} — how an auto-launched
-     *       lead, or an operator following the labelling convention, is found;</li>
-     *   <li>a running agent on a terminal the config pins in {@code fleet.leaders.<name>.terminal} —
-     *       how a lead the operator opened and pinned by hand is found. Without this, a pinned lead
-     *       whose tab carries no matching label would be relaunched on every boot.</li>
-     * </ul>
-     * Member workspaces are excluded, exactly as the scanner excludes them: a member must not be
-     * counted as a lead because it happens to sit in a matching tab.
+     * <p>There used to be a second path here — a running agent on the terminal a
+     * {@code fleet.leaders.<name>.terminal} pin named, for a lead opened and pinned by hand. That
+     * pin is retired: {@code tab} is now the only field identity depends on, and {@link Agent}
+     * already carries {@link Agent#tabId()} directly, so a hand-opened lead is found the same way an
+     * auto-launched one is — by labelling its tab to match.
      */
     private Map<String, Integer> liveLeads(Map<String, BridgedConfig.Leader> leaders) {
         Set<String> memberSpaces = cfg.profiles().values().stream()
@@ -160,20 +158,9 @@ public final class LeadLauncher {
             }
         }
 
-        // terminalId → the lead name the config pins it to.
-        Map<String, String> nameByPinnedTerminal = new LinkedHashMap<>();
-        leaders.forEach((name, lead) -> {
-            if (lead.terminal() != null && !lead.terminal().isBlank()) {
-                nameByPinnedTerminal.put(lead.terminal().strip(), name);
-            }
-        });
-
         Map<String, Integer> counts = new LinkedHashMap<>();
         for (Agent a : agents.list()) {
             String name = nameByTab.get(a.tabId());
-            if (name == null) {
-                name = nameByPinnedTerminal.get(a.terminalId());
-            }
             if (name != null) {
                 counts.merge(name, 1, Integer::sum);
             }
@@ -184,7 +171,7 @@ public final class LeadLauncher {
     /**
      * The configured lead a tab label names, or {@code null} for a label that names none.
      *
-     * <p>Matched against the declared lead names rather than by splitting on the prefix, so an
+     * <p>Matched exactly (case-insensitively) against each lead's configured {@code tab}, so an
      * operator's {@code "lead: something-else"} tab is not mistaken for a configured lead.
      */
     private String leadNameOf(String label, Map<String, BridgedConfig.Leader> leaders) {
@@ -193,7 +180,8 @@ public final class LeadLauncher {
         }
         String l = label.strip();
         for (Map.Entry<String, BridgedConfig.Leader> e : leaders.entrySet()) {
-            if (l.equalsIgnoreCase(e.getValue().tabLabel(e.getKey()).strip())) {
+            String tab = e.getValue().tabLabel();
+            if (tab != null && l.equalsIgnoreCase(tab.strip())) {
                 return e.getKey();
             }
         }
@@ -202,7 +190,7 @@ public final class LeadLauncher {
 
     /** Start one lead. Returns false (having logged) rather than throwing on any failure. */
     private boolean launch(String name, BridgedConfig.Leader lead, BridgedConfig.Profile profile) {
-        String label = lead.tabLabel(name);
+        String label = lead.tabLabel();
         String cwd = (lead.cwd() == null || lead.cwd().isBlank())
                 ? System.getProperty("user.dir") : lead.cwd();
 

bridged/src/main/java/dev/ltms/bridged/logging/McpCancelledNotificationFilter.java

@@ -0,0 +1,30 @@
+package dev.ltms.bridged.logging;
+
+import ch.qos.logback.classic.Level;
+import ch.qos.logback.classic.Logger;
+import ch.qos.logback.classic.turbo.TurboFilter;
+import ch.qos.logback.core.spi.FilterReply;
+import io.modelcontextprotocol.spec.McpSchema;
+import org.slf4j.Marker;
+
+/** Suppresses only the SDK warning for the normal MCP cancellation notification. */
+public final class McpCancelledNotificationFilter extends TurboFilter {
+
+    static final String LOGGER = "io.modelcontextprotocol.spec.McpStreamableServerSession";
+    static final String UNHANDLED_NOTIFICATION = "No handler registered for notification method: {}";
+
+    @Override
+    public FilterReply decide(Marker marker, Logger logger, Level level, String format, Object[] params,
+                              Throwable throwable) {
+        if (level == Level.WARN
+                && LOGGER.equals(logger.getName())
+                && UNHANDLED_NOTIFICATION.equals(format)
+                && params != null
+                && params.length == 1
+                && params[0] instanceof McpSchema.JSONRPCNotification notification
+                && "notifications/cancelled".equals(notification.method())) {
+            return FilterReply.DENY;
+        }
+        return FilterReply.NEUTRAL;
+    }
+}

bridged/src/main/java/dev/ltms/bridged/mcp/BridgeMcp.java

@@ -33,7 +33,10 @@ import jakarta.servlet.http.HttpServlet;
 import java.util.LinkedHashMap;
 import java.util.List;
 import java.util.Map;
+import java.util.Set;
 import java.util.function.Function;
+import java.util.function.LongSupplier;
+import java.util.function.Supplier;
 import java.util.stream.Collectors;
 
 /**
@@ -73,17 +76,20 @@ public final class BridgeMcp {
     private final McpSyncServer server;
     private final CallerResolver authz; // CB-501: null → authorization not enforced (legacy)
     private final Metrics metrics;      // CB-502: null → auth failures not counted
+    private final CapacitySource capacity;
+    private final HealthCoverageSource healthCoverage;
 
-    /**
-     * Legacy constructor — no authorization. Retained so existing tests exercise tool behaviour
-     * without an auth fixture.
-     */
-    public BridgeMcp(MessageService messages, PeerLauncher workers,
-                     SessionManager sessions, ConnectionIdentity identity, MemberPresence presence,
-                     PrimaryRegistry primaryRegistry) {
-        this(messages, workers, sessions, identity, presence, primaryRegistry, null, null);
+    /** Capacity facts used by {@code bridge_list}; production must supply the placement live count. */
+    public record CapacitySource(Function<String, Integer> liveCount, Function<String, Integer> maxLoad,
+                                 Supplier<Set<String>> configuredProfiles, LongSupplier clock) {
+        /** Inert test-only source. It omits capacity rather than inventing zero live counts. */
+        public static CapacitySource none() { return new CapacitySource(_ -> 0, _ -> null, Set::of, System::nanoTime); }
+        boolean available() { return !configuredProfiles.get().isEmpty(); }
     }
 
+    /** Coverage is supplied by the health wiring, not inferred from a missing dependency. */
+    public record HealthCoverageSource(Supplier<String> value) { }
+
     /**
      * @param callers resolves each call's {@link Principal}; {@code null} disables authorization.
      *                This surface needs its own enforcement: {@code /mcp} is a raw servlet on
@@ -91,9 +97,11 @@ public final class BridgeMcp {
      *                filter, so the REST guard does not cover it.
      * @param metrics registry for auth-failure counting; may be {@code null}
      */
-    public BridgeMcp(MessageService messages, PeerLauncher workers,
-                     SessionManager sessions, ConnectionIdentity identity, MemberPresence presence,
-                     PrimaryRegistry primaryRegistry, CallerResolver callers, Metrics metrics) {
+    public BridgeMcp(MessageService messages, PeerLauncher workers, SessionManager sessions,
+                     ConnectionIdentity identity, MemberPresence presence, PrimaryRegistry primaryRegistry,
+                      CallerResolver callers, Metrics metrics, CapacitySource capacity, HealthCoverageSource healthCoverage) {
+        this.capacity = capacity;
+        this.healthCoverage = healthCoverage;
         McpJsonMapper json = new JacksonMcpJsonMapperSupplier().get();
         this.transport = HttpServletStreamableServerTransportProvider.builder()
                 .jsonMapper(json)
@@ -151,8 +159,8 @@ public final class BridgeMcp {
                     Runnable onAccepted = () -> primaryRegistry.recordDelegation(target, caller);
                     // wait defaults to true (block for the reply); wait:false is fire-and-poll.
                     return Boolean.FALSE.equals(a.get("wait"))
-                            ? sendAsync(messages, target, content, onAccepted)
-                            : send(messages, target, content, timeoutMs(a), onAccepted);
+                            ? sendAsync(messages, target, content, onAccepted, workers.profiles())
+                            : send(messages, target, content, timeoutMs(a), onAccepted, workers.profiles());
                 })
                 // bridge_reply's identity is the CONNECTION, never an argument — so the authz check
                 // is "is this caller a worker at all", and it can only ever reply as itself.
@@ -207,7 +215,7 @@ public final class BridgeMcp {
                 .toolCall(listTool(), (exchange, _) -> {
                     McpSchema.CallToolResult denied = deny(exchange, Authz.Action.READ, null);
                     if (denied != null) return denied;
-                    return listFleet(workers, sessions,
+                    return listFleet(workers, sessions, messages, capacity, healthCoverage,
                             callers == null ? Map.of() : callers.leads(),
                             callerTerminal(exchange));
                 })
@@ -379,21 +387,22 @@ public final class BridgeMcp {
 
     // --- tool logic (thin adapters over the services; unit-testable) ---------------------------
 
-    /** {@code bridge_send}: delegate {@code content} to a worker session and block for its reply. */
-    static McpSchema.CallToolResult send(MessageService messages, String sessionId, String content, Long timeoutMs) {
-        return send(messages, sessionId, content, timeoutMs, null);
-    }
-
     /**
-     * As {@link #send(MessageService, String, String, Long)}, wiring an accepted-delivery hook
+     * {@code bridge_send}: delegate {@code content} to a worker session and block for its reply.
+     * The configured profiles are required so a profile name can never bypass target validation.
+     *
      * (CB-548): {@code onAccepted} records delegator ownership the instant the send is accepted, so
      * a BUSY interloper never claims a turn it did not win. {@code null} disables recording.
      */
     static McpSchema.CallToolResult send(MessageService messages, String sessionId, String content,
-                                         Long timeoutMs, Runnable onAccepted) {
+                                         Long timeoutMs, Runnable onAccepted, Set<String> profiles) {
         if (isBlank(sessionId) || isBlank(content)) {
             return error("sessionId and content are required");
         }
+        McpSchema.CallToolResult targetError = profileTargetError(sessionId, profiles);
+        if (targetError != null) {
+            return targetError;
+        }
         long timeout = clamp(timeoutMs == null ? DEFAULT_TIMEOUT_MS : timeoutMs);
         try {
             return formatReply(messages.send(sessionId, content, timeout, onAccepted), timeout);
@@ -449,6 +458,11 @@ public final class BridgeMcp {
                     "[worker finished without a structured bridge_reply — transcript tail follows]\n" + r.text());
             // The worker ran the turn then wedged (CB-109) — surface the error context.
             case WORKER_FAILED -> text("[worker failed — turn ended in an unrecoverable state]\n" + r.text());
+            // The backend refused on a subscription usage limit (CB-578 stage A) — the worker's
+            // pane stayed healthy, but its account is exhausted. Distinct from WORKER_FAILED so the
+            // primary gets the real cause, not a generic wedge.
+            case BACKEND_EXHAUSTED -> text("[backend exhausted — the worker's account refused on a "
+                    + "usage limit]\n" + r.text());
             // The worker paused mid-turn to ask (CB-205) — tell the primary how to answer in-turn.
             case QUESTION -> text("[question] the worker paused to ask before it can finish:\n" + r.text()
                     + "\n\nAnswer it by calling bridge_send again with turnId=\"" + r.turnId()
@@ -463,24 +477,33 @@ public final class BridgeMcp {
     /**
      * {@code bridge_send} with {@code wait:false}: delegate {@code content} and return a ticket
      * immediately (fire-and-poll), so a long task isn't cut off by the caller's MCP call timeout.
-     */
-    static McpSchema.CallToolResult sendAsync(MessageService messages, String sessionId, String content) {
-        return sendAsync(messages, sessionId, content, null);
-    }
-
-    /**
-     * As {@link #sendAsync(MessageService, String, String)}, wiring the accepted-delivery hook
+     * The configured profiles are required so a profile name can never bypass target validation.
+     *
+     * This wires the accepted-delivery hook
      * (CB-548) so an async flooding send records delegator ownership exactly once it is accepted.
      */
     static McpSchema.CallToolResult sendAsync(MessageService messages, String sessionId, String content,
-                                              Runnable onAccepted) {
+                                              Runnable onAccepted, Set<String> profiles) {
         if (isBlank(sessionId) || isBlank(content)) {
             return error("sessionId and content are required");
         }
+        McpSchema.CallToolResult targetError = profileTargetError(sessionId, profiles);
+        if (targetError != null) {
+            return targetError;
+        }
         String ticket = messages.sendAsync(sessionId, content, onAccepted);
         return text("accepted — task delegated. Poll bridge_poll with ticket=" + ticket);
     }
 
+    /** A configured profile is never a send target; other unknown values may be herdr-owned panes. */
+    private static McpSchema.CallToolResult profileTargetError(String sessionId, Set<String> profiles) {
+        if (profiles.contains(sessionId)) {
+            return error("unknown send target \"" + sessionId + "\": it is a configured profile name, not a "
+                    + "session id. Call bridge_list to find a member or lead sessionId.");
+        }
+        return null;
+    }
+
     /** {@code bridge_poll}: check an async delegation by ticket, or drain a worker's inbox by target. */
     static McpSchema.CallToolResult poll(MessageService messages, String ticket, String target) {
         if (!isBlank(target)) {
@@ -502,6 +525,9 @@ public final class BridgeMcp {
                     ? "[done — worker finished without a structured bridge_reply; transcript tail follows]\n" + v.reply()
                     : v.reply());
             case PENDING -> text("[pending — " + v.detail() + "]");
+            case ASKING -> text("[question — worker is waiting for your answer]\n" + v.reply()
+                    + "\n\nAnswer it by calling bridge_send again with turnId=\"" + v.turnId()
+                    + "\" and content set to your answer; the worker resumes the same turn.");
             case FAILED -> text("[failed — " + v.detail() + "]");
         };
     }
@@ -708,6 +734,12 @@ public final class BridgeMcp {
      */
     static McpSchema.CallToolResult listFleet(PeerLauncher workers, SessionManager sessions,
                                               Map<String, String> leads, String selfTerm) {
+        return listFleet(workers, sessions, null, CapacitySource.none(), new HealthCoverageSource(() -> "off"), leads, selfTerm);
+    }
+
+    static McpSchema.CallToolResult listFleet(PeerLauncher workers, SessionManager sessions, MessageService messages,
+                                               CapacitySource capacity, HealthCoverageSource healthCoverage,
+                                              Map<String, String> leads, String selfTerm) {
         try {
             Map<String, Agent> live = workers.list().stream()
                     .map(Agent.class::cast)
@@ -717,15 +749,57 @@ public final class BridgeMcp {
                     .sorted(Map.Entry.comparingByValue())
                     .map(e -> leadView(e.getKey(), e.getValue(), live.get(e.getKey()), selfTerm))
                     .toList();
-            List<Map<String, Object>> out = sessions.roster().stream()
-                    .map(s -> SessionManager.rosterView(s, live.get(s.terminalId())))
+            List<MemberSession> roster = sessions.roster();
+            List<Map<String, Object>> out = roster.stream()
+                    .map(s -> memberCapacityView(s, live.get(s.terminalId()), messages, capacity.clock().getAsLong()))
                     .toList();
-            return text(json(Map.of("leads", leadRows, "members", out)));
+            Set<String> profiles = new java.util.TreeSet<>(capacity.configuredProfiles().get());
+            roster.stream().map(MemberSession::profile).forEach(profiles::add);
+            Map<String, Object> result = new LinkedHashMap<>();
+            result.put("leads", leadRows); result.put("members", out);
+            result.put("healthCoverage", healthCoverage.value().get());
+            if (capacity.available()) result.put("capacity", profiles.stream()
+                    .map(profile -> capacityView(profile, capacity.liveCount(), capacity.maxLoad(), roster, messages,
+                            capacity.clock().getAsLong())).toList());
+            return text(json(result));
         } catch (HerdrException e) {
             return error("herdr error listing the fleet: " + e.getMessage());
         }
     }
 
+    /**
+     * Capacity is advisory only. {@code reclaimable} says there is no bridge work, not that bridged
+     * may stop the member: the bridge has capacity facts but no work list, and choosing work needs
+     * authority it does not have. {@code idleForSeconds} is derived from monotonic nanoTime and has
+     * no wall-clock meaning across a daemon restart.
+     */
+    private static Map<String, Object> memberCapacityView(MemberSession session, Agent live,
+                                                           MessageService messages, long nowNanos) {
+        Map<String, Object> row = SessionManager.rosterView(session, live);
+        boolean open = messages != null && messages.hasAcceptedDelivery(session.terminalId());
+        boolean inbox = messages != null && messages.hasInboxMessage(session.terminalId());
+        boolean reclaimable = (session.state() == MemberSession.State.READY || session.state() == MemberSession.State.DONE)
+                && !open && !inbox;
+        row.put("reclaimable", reclaimable);
+        row.put("idleForSeconds", reclaimable ? Math.max(0, (nowNanos - session.lastActivityAtNanos()) / 1_000_000_000L) : null);
+        return row;
+    }
+
+    private static Map<String, Object> capacityView(String profile, Function<String, Integer> liveCount,
+                                                     Function<String, Integer> maxLoad, List<MemberSession> roster,
+                                                     MessageService messages, long nowNanos) {
+        Integer cap = maxLoad.apply(profile);
+        int live = liveCount.apply(profile);
+        int reclaimable = (int) roster.stream().filter(s -> profile.equals(s.profile()))
+                .filter(s -> (s.state() == MemberSession.State.READY || s.state() == MemberSession.State.DONE))
+                .filter(s -> messages == null || (!messages.hasAcceptedDelivery(s.terminalId()) && !messages.hasInboxMessage(s.terminalId())))
+                .count();
+        Map<String, Object> row = new LinkedHashMap<>();
+        row.put("profile", profile); row.put("maxLoad", cap); row.put("live", live);
+        row.put("free", cap == null ? null : Math.max(0, cap - live)); row.put("reclaimable", reclaimable);
+        return row;
+    }
+
     /**
      * One lead's row: its address, its name, and whether it can be reached right now.
      *

bridged/src/main/java/dev/ltms/bridged/member/ClaudeCodeLauncher.java

@@ -76,7 +76,7 @@ public final class ClaudeCodeLauncher extends HerdrPeerLauncher {
                               Map<String, BridgedConfig.Profile> profiles, String defaultProfile,
                               Function<String, String> env,
                               long spawnReadyTimeoutMs, long spawnReadyPollMs,
-                               Supplier<BridgedConfig.Fleet> fleet) {
+                              Supplier<BridgedConfig.Fleet> fleet) {
         this(agents, spaces, guard, profiles, defaultProfile, env,
                 spawnReadyTimeoutMs,
                 System::currentTimeMillis, () -> sleepUninterruptibly(spawnReadyPollMs),
@@ -118,8 +118,8 @@ public final class ClaudeCodeLauncher extends HerdrPeerLauncher {
                               Map<String, BridgedConfig.Profile> profiles, String defaultProfile,
                               Function<String, String> env,
                               long spawnReadyTimeoutMs,
-                               LongSupplier nowMillis, Runnable sleeper,
-                               Supplier<BridgedConfig.Fleet> fleet) {
+                              LongSupplier nowMillis, Runnable sleeper,
+                              Supplier<BridgedConfig.Fleet> fleet) {
         super(NAME_PREFIX, agents, spaces, profiles, defaultProfile, env,
                 spawnReadyTimeoutMs, nowMillis, sleeper, fleet);
         this.guard = guard;
@@ -181,8 +181,8 @@ public final class ClaudeCodeLauncher extends HerdrPeerLauncher {
         // id via -r and passes no --session-id (the two conflict). Both are injected before the
         // model flag so --model keeps outranking the operator's own argv.
         // mutableArgv: argvWithBridge may hand back the profile's own (immutable) List.of when it
-        // has no MCP — session flags must be added into a list we own.
-        List<String> argv = mutableArgv(argvWithBridge(cfg));
+        // has neither MCP nor a charter — session flags must be added into a list we own.
+        List<String> argv = mutableArgv(argvWithBridge(cfg, spec.charter()));
         String agentSessionId = applySessionIdentity(argv, spec.sessionName(), spec.resumeSessionId());
         return new Launch(workerEnv, argvWithModel(argv, cfg), agentSessionId);
     }
@@ -217,22 +217,26 @@ public final class ClaudeCodeLauncher extends HerdrPeerLauncher {
     }
 
     /**
-     * The launch argv, plus — when {@code worker.mcpUrl} is set — inline {@code --mcp-config} for
-     * the bridge server and {@code --append-system-prompt} for the {@link #REPLY_CHARTER}. Neither
-     * touches the profile's config; both are pure command-line flags. This inline-flag mount is
-     * Claude Code specific — other adapters mount MCP and instructions their own way.
+     * The launch argv, plus an inline {@code --mcp-config} when {@code worker.mcpUrl} is set and
+     * {@code --append-system-prompt} when the base composed a charter. Neither touches the profile's
+     * config; both are pure command-line flags. This inline-flag mount is Claude Code specific —
+     * other adapters mount MCP and instructions their own way.
      */
-    private List<String> argvWithBridge(BridgedConfig.Profile cfg) {
-        if (!cfg.hasMcp()) {
+    private List<String> argvWithBridge(BridgedConfig.Profile cfg, String charter) {
+        if (!cfg.hasMcp() && charter == null) {
             return cfg.argv();
         }
-        String mcpJson = "{\"mcpServers\":{\"bridge\":{\"type\":\"http\",\"url\":\""
-                + cfg.mcpUrl() + "\"}}}";
         List<String> argv = mutableArgv(cfg.argv());
-        argv.add("--mcp-config");
-        argv.add(mcpJson);
-        argv.add("--append-system-prompt");
-        argv.add(REPLY_CHARTER);
+        if (cfg.hasMcp()) {
+            String mcpJson = "{\"mcpServers\":{\"bridge\":{\"type\":\"http\",\"url\":\""
+                    + cfg.mcpUrl() + "\"}}}";
+            argv.add("--mcp-config");
+            argv.add(mcpJson);
+        }
+        if (charter != null) {
+            argv.add("--append-system-prompt");
+            argv.add(charter);
+        }
         return argv;
     }
 

bridged/src/main/java/dev/ltms/bridged/member/HerdrPeerLauncher.java

@@ -155,16 +155,15 @@ public abstract class HerdrPeerLauncher implements PeerLauncher {
      * As above, plus the live {@code fleet} config (CB-557).
      *
      * @param fleet live fleet config, read once per spawn; {@code null} ⇒ default tab label and no
-     *              role charter. A separate constructor rather than a new parameter on the one above,
-     *              so every existing call
-     *                         site keeps the default without an edit.
+     *              role charter. A separate constructor rather than a new parameter on the one
+     *              above, so every existing call site keeps the default without an edit.
      */
     protected HerdrPeerLauncher(String namePrefix, AgentControl agents, WorkspaceControl spaces,
                                 Map<String, BridgedConfig.Profile> profiles, String defaultProfile,
                                 Function<String, String> env,
                                 long spawnReadyTimeoutMs,
                                 LongSupplier nowMillis, Runnable sleeper,
-                                 Supplier<BridgedConfig.Fleet> fleet) {
+                                Supplier<BridgedConfig.Fleet> fleet) {
         this.fleet = fleet;
         this.namePrefix = namePrefix;
         this.agents = agents;

bridged/src/main/java/dev/ltms/bridged/member/OpenCodeLauncher.java

@@ -38,7 +38,7 @@ import java.util.function.Supplier;
  *   <li><strong>File-based MCP mount + instructions.</strong> opencode has no inline
  *       {@code --mcp-config}/{@code --append-system-prompt}. Instead the bridge writes an ephemeral
  *       {@code opencode.json} that declares the bridge as a {@code remote} MCP server and lists a
- *       reply-charter file under {@code instructions}, then points the worker at it with
+ *       member-charter file under {@code instructions}, then points the worker at it with
  *       {@code OPENCODE_CONFIG}. This is the one place the launcher touches disk — Claude never did.</li>
  *   <li><strong>Model as a flag.</strong> the {@code provider/model} selector is passed as
  *       {@code -m}, not an env var.</li>
@@ -97,7 +97,7 @@ public final class OpenCodeLauncher extends HerdrPeerLauncher {
                             Map<String, BridgedConfig.Profile> profiles, String defaultProfile,
                             Function<String, String> env,
                             long spawnReadyTimeoutMs, long spawnReadyPollMs,
-                             Supplier<BridgedConfig.Fleet> fleet) {
+                            Supplier<BridgedConfig.Fleet> fleet) {
         this(agents, spaces, profiles, defaultProfile, env, spawnReadyTimeoutMs,
                 System::currentTimeMillis, () -> sleepUninterruptibly(spawnReadyPollMs),
                 defaultConfigRoot(), defaultDiscoveryRoot(), fleet);
@@ -143,7 +143,7 @@ public final class OpenCodeLauncher extends HerdrPeerLauncher {
                             long spawnReadyTimeoutMs,
                             LongSupplier nowMillis, Runnable sleeper,
                             Path configRoot, Path discoveryRoot,
-                             Supplier<BridgedConfig.Fleet> fleet) {
+                            Supplier<BridgedConfig.Fleet> fleet) {
         super(NAME_PREFIX, agents, spaces, profiles, defaultProfile, env,
                 spawnReadyTimeoutMs, nowMillis, sleeper, fleet);
         this.configRoot = configRoot;
@@ -163,17 +163,17 @@ public final class OpenCodeLauncher extends HerdrPeerLauncher {
      * {@inheritDoc}
      *
      * <p>Builds the opencode launch: no {@code ANTHROPIC_*} and no guard (opencode reads its own
-     * provider credentials); when the profile mounts the bridge MCP, generate an ephemeral
-     * {@code opencode.json} (remote MCP server + reply-charter instructions) and point the worker at
-     * it via {@code OPENCODE_CONFIG}; carry the parity-neutral git-forge grant; and select the model
-     * with {@code -m}.
+     * provider credentials); when the profile mounts the bridge MCP or has a member charter,
+     * generate an ephemeral {@code opencode.json} (remote MCP server + member-charter instructions)
+     * and point the worker at it via {@code OPENCODE_CONFIG}; carry the parity-neutral git-forge
+     * grant; and select the model with {@code -m}.
      */
     @Override
     protected Launch buildLaunch(BridgedConfig.Profile cfg, LaunchSpec spec) {
         Map<String, String> workerEnv = baseEnv(cfg);
-        // A config file is needed for the bridge MCP mount, for a pinned endpoint (CB-508), or both.
-        if (cfg.hasMcp() || hasCustomProvider(cfg)) {
-            workerEnv.put("OPENCODE_CONFIG", writeConfig(cfg).toString());
+        // A config file is needed for the bridge MCP mount, a member charter, or a pinned endpoint (CB-508).
+        if (cfg.hasMcp() || spec.charter() != null || hasCustomProvider(cfg)) {
+            workerEnv.put("OPENCODE_CONFIG", writeConfig(cfg, spec.charter()).toString());
         }
         applyGitToken(workerEnv, cfg);
         return new Launch(workerEnv,
@@ -236,12 +236,12 @@ public final class OpenCodeLauncher extends HerdrPeerLauncher {
     }
 
     /**
-     * Write an ephemeral {@code opencode.json} (and the reply-charter file it references) into a
+     * Write an ephemeral {@code opencode.json} (and the member-charter file it references) into a
      * fresh per-spawn directory under {@link #configRoot}, and return the config file's path for
      * {@code OPENCODE_CONFIG}. The dir is unique per spawn so concurrent workers never race on it;
      * it is best-effort cleaned on JVM exit (worker config is disposable — regenerated every spawn).
      */
-    private Path writeConfig(BridgedConfig.Profile cfg) {
+    private Path writeConfig(BridgedConfig.Profile cfg, String charterText) {
         try {
             Path dir = Files.createTempDirectory(configRoot, "bridged-opencode-");
             dir.toFile().deleteOnExit();
@@ -264,16 +264,19 @@ public final class OpenCodeLauncher extends HerdrPeerLauncher {
             // if per-profile control is ever wanted, add a profile knob rather than dropping this.
             root.putObject("compaction").put("auto", true);
 
-            if (cfg.hasMcp()) {
-                Path charter = dir.resolve("reply-charter.md");
-                Files.writeString(charter, REPLY_CHARTER);
+            if (charterText != null) {
+                Path charter = dir.resolve("member-charter.md");
+                Files.writeString(charter, charterText);
                 charter.toFile().deleteOnExit();
 
+                root.putArray("instructions").add(charter.toAbsolutePath().toString());
+            }
+
+            if (cfg.hasMcp()) {
                 ObjectNode bridge = root.putObject("mcp").putObject("bridge");
                 bridge.put("type", "remote");
                 bridge.put("url", cfg.mcpUrl());
                 bridge.put("enabled", true);
-                root.putArray("instructions").add(charter.toAbsolutePath().toString());
             }
             if (hasCustomProvider(cfg)) {
                 addCustomProvider(root, cfg);

bridged/src/main/java/dev/ltms/bridged/msg/MessageService.java

@@ -69,6 +69,14 @@ public final class MessageService {
          * failure context (e.g. the error screen). Terminal, but not a successful completion.
          */
         WORKER_FAILED,
+        /**
+         * The turn finished without a {@code bridge_reply} and the scrape matched the backend's
+         * configured usage-limit refusal pattern (CB-578 stage A); {@code text} is the reason,
+         * carrying the matched line. The worker's pane is healthy — only its account is refusing —
+         * so this is never reported as a completed reply, and is kept distinct from
+         * {@link #WORKER_FAILED} (a wedged worker) and a session simply going {@code GONE}.
+         */
+        BACKEND_EXHAUSTED,
         /**
          * The worker paused mid-turn to ask the primary a question (CB-205); {@code text} is the
          * question and {@code turnId} correlates the answer. Not terminal — the primary answers with
@@ -127,6 +135,8 @@ public final class MessageService {
     public enum Phase {
         /** Delegated and in flight — queued for the worker or being worked. */
         PENDING,
+        /** The worker is paused in {@code bridge_ask}; {@link TaskView#reply} and {@link TaskView#turnId} identify it. */
+        ASKING,
         /** The worker's turn finished; {@link TaskView#reply} holds the answer. */
         DONE,
         /** The delegation could not complete (timed out, worker gone, or busy). */
@@ -136,16 +146,28 @@ public final class MessageService {
     /**
      * A poll snapshot of an async delegation.
      *
-     * @param reply       the answer when {@link #phase} is {@link Phase#DONE}, else {@code null}
+     * @param reply       the answer when {@link #phase} is {@link Phase#DONE}, or the question when
+     *                    {@link #phase} is {@link Phase#ASKING}; otherwise {@code null}
      * @param replySource {@code "reply"} (structured {@code bridge_reply}) or {@code "transcript"}
      *                    (completion scrape) when {@link Phase#DONE}, else {@code null}
-     * @param detail      a human note (live worker status while pending, or the failure reason)
+     * @param detail      a human note (live worker status while pending, ask state, or failure reason)
+     * @param turnId      correlation id for an {@link Phase#ASKING} ticket, else {@code null}
      */
-    public record TaskView(String ticket, Phase phase, String reply, String replySource, String detail) {
+    public record TaskView(String ticket, Phase phase, String reply, String replySource, String detail,
+                           String turnId) {
     }
 
     /** An in-flight or finished async delegation, keyed by its ticket. */
-    private record Task(String target, CompletableFuture<Reply> future, long createdNanos) {
+    private static final class Task {
+        private final String target;
+        private final CompletableFuture<Reply> future = new CompletableFuture<>();
+        private final long createdNanos = System.nanoTime();
+        private volatile Reply question;
+        private volatile String turnId;
+
+        private Task(String target) {
+            this.target = target;
+        }
     }
 
     private final AgentControl agents;
@@ -156,6 +178,11 @@ public final class MessageService {
     private final Metrics metrics; // CB-502: nullable — no registry in unit tests
     private final ConcurrentHashMap<String, ReentrantLock> sessionLocks = new ConcurrentHashMap<>();
     private final ConcurrentHashMap<String, Task> tasks = new ConcurrentHashMap<>();
+    /** Async task that owns each exact forward rendezvous waiter. */
+    private final ConcurrentHashMap<CompletableFuture<Rendezvous.Resolution>, Task> asyncTasksByWaiter =
+            new ConcurrentHashMap<>();
+    /** Async tickets paused on a specific {@code bridge_ask} turn. */
+    private final ConcurrentHashMap<String, Task> asyncTasksByTurn = new ConcurrentHashMap<>();
     private final AtomicLong ticketSeq = new AtomicLong();
     private final ExecutorService asyncExecutor = Executors.newThreadPerTaskExecutor(
             Thread.ofVirtual().name("bridge-async-", 0).factory());
@@ -202,6 +229,16 @@ public final class MessageService {
         return agents.status(target);
     }
 
+    /** Read-only delegation fact for fleet views. */
+    public boolean hasAcceptedDelivery(String target) {
+        return rendezvous.isWaiting(target);
+    }
+
+    /** Read-only inbox fact for fleet views. */
+    public boolean hasInboxMessage(String target) {
+        return !inbox.peek(target).isEmpty();
+    }
+
     /**
      * Route a worker's explicit {@code bridge_reply}: resolve an open send, or queue it in the
      * inbox if no send is currently open. Unlike the bare {@link Rendezvous#resolve}, a no-waiter
@@ -251,6 +288,7 @@ public final class MessageService {
             case COMPLETED_UNREPLIED -> "completion_fallback";
             case TIMED_OUT_WORKING, TIMED_OUT_QUEUED, BUSY -> "timeout";
             case WORKER_FAILED -> "failed";
+            case BACKEND_EXHAUSTED -> "backend_exhausted";
             case STALE_TURN, QUESTION -> null; // not a completed delegation
         };
     }
@@ -272,14 +310,18 @@ public final class MessageService {
      */
     public boolean abandon(String target, String reason) {
         CompletableFuture<Rendezvous.Resolution> waiter = rendezvous.currentWaiter(target);
-        if (waiter == null || waiter.isDone()) {
-            return false; // nobody is blocked on this worker — nothing to abandon
+        boolean failed = waiter != null && !waiter.isDone() && rendezvous.resolveFailure(waiter, reason);
+        boolean asyncFailed = false;
+        for (Task task : tasks.values()) {
+            if (target.equals(task.target) && task.question == null
+                    && task.future.complete(new Reply(Outcome.WORKER_FAILED, reason))) {
+                asyncFailed = true;
+            }
         }
-        boolean failed = rendezvous.resolveFailure(waiter, reason);
         if (failed) {
             log.warn("abandoning the blocked send to {}: {}", target, reason);
         }
-        return failed;
+        return failed || asyncFailed;
     }
 
     /**
@@ -326,6 +368,11 @@ public final class MessageService {
      * never earned. {@code null} disables the hook.
      */
     public Reply send(String target, String content, long timeoutMillis, Runnable onAccepted) {
+        return send(target, content, timeoutMillis, onAccepted, null);
+    }
+
+    /** Run a send, optionally stopping an async task that teardown already failed before acceptance. */
+    private Reply send(String target, String content, long timeoutMillis, Runnable onAccepted, Task task) {
         long deadlineNanos = System.nanoTime() + timeoutMillis * 1_000_000L;
         ReentrantLock lock = sessionLocks.computeIfAbsent(target, _ -> new ReentrantLock());
 
@@ -333,6 +380,12 @@ public final class MessageService {
             return new Reply(Outcome.BUSY, null); // another send held the session the whole window
         }
         try {
+            if (task != null && task.future.isDone()) {
+                return task.future.getNow(null);
+            }
+            if (hasAsyncQuestion(target)) {
+                return new Reply(Outcome.BUSY, null); // the worker's current turn is paused for its lead
+            }
             // Open the waiter BEFORE queueing delivery (CB-548). A fast reply — the worker already
             // injectable the instant we enqueue — otherwise arrives before the waiter is registered
             // and orphans into the inbox while this send blocks to the timeout (the enqueue-before-
@@ -341,13 +394,17 @@ public final class MessageService {
             // failed send leaves no stale waiter behind.
             CompletableFuture<Rendezvous.Resolution> reply = rendezvous.open(target);
             try {
+                if (task != null) {
+                    asyncTasksByWaiter.put(reply, task);
+                }
+                TurnToken token = new TurnToken(target, reply);
                 // The send has won the lock; the accepted-delivery hook records delegator ownership
                 // here (CB-548). It runs BEFORE enqueue so a throwing hook — onAccepted is now a
                 // public callback — fails the send without queuing a message that would orphan.
                 if (onAccepted != null) {
                     onAccepted.run();
                 }
-                CompletableFuture<Void> delivered = injector.enqueue(target, content);
+                CompletableFuture<Void> delivered = injector.enqueue(target, content, token);
                 try {
                     Rendezvous.Resolution r = reply.get(remainingMillis(deadlineNanos), TimeUnit.MILLISECONDS);
                     return recorded(new Reply(outcomeOf(r.kind()), r.text(), r.turnId()));
@@ -364,6 +421,7 @@ public final class MessageService {
                     throw new IllegalStateException("interrupted awaiting reply from " + target, e);
                 }
             } finally {
+                asyncTasksByWaiter.remove(reply);
                 rendezvous.close(target, reply);
             }
         } finally {
@@ -389,7 +447,12 @@ public final class MessageService {
         if (ticket.fresh()) {
             // Register the reverse waiter first, then surface the question — so the answer, which can
             // arrive the instant the primary reacts, always finds an open waiter to resolve.
+            CompletableFuture<Rendezvous.Resolution> waiter = rendezvous.currentWaiter(workerSession);
+            Task task = markAsyncQuestion(waiter, question, ticket.turnId());
             if (!rendezvous.resolveQuestion(workerSession, question, ticket.turnId())) {
+                if (task != null) {
+                    clearAsyncQuestion(ticket.turnId(), true);
+                }
                 rendezvous.closeAsk(ticket.turnId());
                 return new AskResult(AskOutcome.NO_WAITER, null); // no primary is blocked on this worker
             }
@@ -399,6 +462,7 @@ public final class MessageService {
             return new AskResult(AskOutcome.ANSWERED, answer);
         } catch (TimeoutException e) {
             log.debug("bridge_ask from {} went unanswered in {}ms", workerSession, timeoutMillis);
+            clearAsyncQuestion(ticket.turnId(), true);
             return new AskResult(AskOutcome.TIMED_OUT, null);
         } catch (ExecutionException e) {
             Throwable cause = e.getCause();
@@ -441,9 +505,12 @@ public final class MessageService {
                 rendezvous.close(workerSession, reply);
                 return new Reply(Outcome.STALE_TURN, null); // lapsed between the lookup and the unblock
             }
+            clearAsyncQuestion(turnId, false);
             try {
                 Rendezvous.Resolution r = reply.get(remainingMillis(deadlineNanos), TimeUnit.MILLISECONDS);
-                return new Reply(outcomeOf(r.kind()), r.text(), r.turnId());
+                Reply result = new Reply(outcomeOf(r.kind()), r.text(), r.turnId());
+                finishAsyncTask(turnId, result);
+                return result;
             } catch (TimeoutException e) {
                 // The worker resumed but hasn't replied yet — no completion fallback arms an answered
                 // turn (it never re-entered the injector), so a silent worker rides out the window.
@@ -483,9 +550,21 @@ public final class MessageService {
      */
     public String sendAsync(String target, String content, Runnable onAccepted) {
         String ticket = "task-" + ticketSeq.incrementAndGet();
-        CompletableFuture<Reply> future = CompletableFuture.supplyAsync(
-                () -> send(target, content, ASYNC_TIMEOUT_MS, onAccepted), asyncExecutor);
-        tasks.put(ticket, new Task(target, future, System.nanoTime()));
+        Task task = new Task(target);
+        tasks.put(ticket, task);
+        asyncExecutor.submit(() -> {
+            try {
+                Reply result = send(target, content, ASYNC_TIMEOUT_MS, onAccepted, task);
+                if (result.outcome() == Outcome.QUESTION) {
+                    // Keep the accepted owner until answer() finishes it. markAsyncQuestion may run
+                    // just after resolveQuestion wakes this thread.
+                } else {
+                    finishAsyncTask(task, result);
+                }
+            } catch (Throwable t) {
+                task.future.completeExceptionally(t);
+            }
+        });
         pruneTerminalTickets();
         log.debug("async send {} -> {}", ticket, target);
         return ticket;
@@ -501,27 +580,34 @@ public final class MessageService {
         if (task == null) {
             return null;
         }
-        CompletableFuture<Reply> f = task.future();
+        CompletableFuture<Reply> f = task.future;
         if (!f.isDone()) {
-            return new TaskView(ticket, Phase.PENDING, null, null, "worker " + liveStatus(task.target()));
+            Reply question = task.question;
+            if (question != null) {
+                return new TaskView(ticket, Phase.ASKING, question.text(), null,
+                        "worker is waiting for your answer", question.turnId());
+            }
+            return new TaskView(ticket, Phase.PENDING, null, null, "worker " + liveStatus(task.target), null);
         }
         Reply r;
         try {
             r = f.getNow(null);
         } catch (CompletionException | java.util.concurrent.CancellationException e) {
             Throwable cause = (e instanceof CompletionException ce && ce.getCause() != null) ? ce.getCause() : e;
-            return new TaskView(ticket, Phase.FAILED, null, null, cause.getMessage());
+            return new TaskView(ticket, Phase.FAILED, null, null, cause.getMessage(), null);
         }
         if (r.completed()) {
             String source = r.outcome() == Outcome.REPLIED ? "reply" : "transcript";
-            return new TaskView(ticket, Phase.DONE, r.text(), source, null);
+            return new TaskView(ticket, Phase.DONE, r.text(), source, null, null);
         }
-        // A wedged worker (CB-109) carries the error context as its reason; the timeout/busy
-        // outcomes carry none, so fall back to the outcome name.
-        String detail = r.outcome() == Outcome.WORKER_FAILED && r.text() != null
+        // A wedged worker (CB-109) or a backend-exhausted classification (CB-578 stage A) carries
+        // the real cause as its reason; the timeout/busy outcomes carry none, so fall back to the
+        // outcome name.
+        boolean carriesReason = r.outcome() == Outcome.WORKER_FAILED || r.outcome() == Outcome.BACKEND_EXHAUSTED;
+        String detail = carriesReason && r.text() != null
                 ? r.text()
                 : "no reply — " + r.outcome().name().toLowerCase();
-        return new TaskView(ticket, Phase.FAILED, null, null, detail);
+        return new TaskView(ticket, Phase.FAILED, null, null, detail, null);
     }
 
     /** Best-effort live worker status for a pending poll; never throws (a lookup error is just noise). */
@@ -536,7 +622,51 @@ public final class MessageService {
     /** Drop finished tickets older than the TTL so the registry cannot grow without bound. */
     private void pruneTerminalTickets() {
         long cutoff = System.nanoTime() - TICKET_TTL_NANOS;
-        tasks.values().removeIf(t -> t.future().isDone() && t.createdNanos() < cutoff);
+        tasks.values().removeIf(t -> t.future.isDone() && t.createdNanos < cutoff);
+    }
+
+    /** Record the active question for an async ticket; blocking sends have no entry and stay unchanged. */
+    private Task markAsyncQuestion(CompletableFuture<Rendezvous.Resolution> waiter, String text, String turnId) {
+        Task task = waiter == null ? null : asyncTasksByWaiter.get(waiter);
+        if (task != null) {
+            task.question = new Reply(Outcome.QUESTION, text, turnId);
+            task.turnId = turnId;
+            asyncTasksByTurn.put(turnId, task);
+        }
+        return task;
+    }
+
+    /** Clear an answered or lapsed question, but only when it matches the ticket's current turn. */
+    private void clearAsyncQuestion(String turnId, boolean forgetTurn) {
+        Task task = asyncTasksByTurn.get(turnId);
+        if (task != null && turnId.equals(task.turnId)) {
+            task.question = null;
+            if (forgetTurn) {
+                asyncTasksByTurn.remove(turnId, task);
+                task.turnId = null;
+            }
+        }
+    }
+
+    /** Complete and detach an async ticket after its worker's actual terminal reply. */
+    private void finishAsyncTask(Task task, Reply result) {
+        task.future.complete(result);
+        if (task.turnId != null) {
+            asyncTasksByTurn.remove(task.turnId, task);
+        }
+    }
+
+    /** Complete the async ticket correlated to a specific answered turn. */
+    private void finishAsyncTask(String turnId, Reply result) {
+        Task task = asyncTasksByTurn.get(turnId);
+        if (task != null) {
+            finishAsyncTask(task, result);
+        }
+    }
+
+    /** A new send must not open a waiter while an async ticket owns this worker's paused turn. */
+    private boolean hasAsyncQuestion(String target) {
+        return asyncTasksByTurn.values().stream().anyMatch(task -> target.equals(task.target));
     }
 
     /** Release the async executor. */
@@ -550,6 +680,7 @@ public final class MessageService {
             case REPLY -> Outcome.REPLIED;
             case COMPLETION -> Outcome.COMPLETED_UNREPLIED;
             case FAILED -> Outcome.WORKER_FAILED;
+            case BACKEND_EXHAUSTED -> Outcome.BACKEND_EXHAUSTED;
             case QUESTION -> Outcome.QUESTION;
         };
     }

bridged/src/main/java/dev/ltms/bridged/msg/Rendezvous.java

@@ -33,6 +33,13 @@ public final class Rendezvous {
         COMPLETION,
         /** The worker ran the turn then wedged (CB-109); {@code text} is the failure context. */
         FAILED,
+        /**
+         * The turn finished without a {@code bridge_reply}, and the scrape matched the backend's
+         * configured usage-limit refusal pattern (CB-578 stage A); {@code text} is the reason,
+         * carrying the matched line. The pane is healthy — only the account is refusing — so this
+         * is kept separate from a session simply going {@code GONE}.
+         */
+        BACKEND_EXHAUSTED,
         /**
          * The worker paused mid-turn to ask the primary a question (CB-205 reverse rendezvous);
          * {@code text} is the question and {@code turnId} correlates the primary's answer back to
@@ -224,6 +231,19 @@ public final class Rendezvous {
         return waiter != null && waiter.complete(new Resolution(Kind.FAILED, reason));
     }
 
+    /**
+     * Resolve a specific captured {@code waiter} as {@link Kind#BACKEND_EXHAUSTED} (CB-578 stage A):
+     * the turn finished with no {@code bridge_reply} and the scrape matched the backend's configured
+     * usage-limit pattern; {@code reason} carries the matched line. Like
+     * {@link #resolveCompletion(CompletableFuture, String)} it targets the exact captured send
+     * (CB-116). A no-op if that waiter was already resolved — first resolution wins.
+     *
+     * @return {@code true} if this call resolved the waiter, {@code false} if it was null or already resolved
+     */
+    public boolean resolveExhausted(CompletableFuture<Resolution> waiter, String reason) {
+        return waiter != null && waiter.complete(new Resolution(Kind.BACKEND_EXHAUSTED, reason));
+    }
+
     private boolean complete(String session, Resolution resolution) {
         CompletableFuture<Resolution> waiter = waiters.get(session);
         return waiter != null && waiter.complete(resolution);

bridged/src/main/java/dev/ltms/bridged/msg/TurnToken.java

@@ -0,0 +1,20 @@
+package dev.ltms.bridged.msg;
+
+import java.util.concurrent.CompletableFuture;
+
+/**
+ * Identity for one accepted send. The session turn is deliberately absent: CompletionResolver's
+ * delivery callback runs before SessionManager.onDelivered, so binding it needs a later ordering design.
+ */
+public final class TurnToken {
+    private final String target;
+    private final CompletableFuture<Rendezvous.Resolution> waiter;
+
+    public TurnToken(String target, CompletableFuture<Rendezvous.Resolution> waiter) {
+        this.target = target;
+        this.waiter = waiter;
+    }
+
+    public String target() { return target; }
+    public CompletableFuture<Rendezvous.Resolution> waiter() { return waiter; }
+}

bridged/src/main/java/dev/ltms/bridged/rest/BridgedApp.java

@@ -403,9 +403,12 @@ public final class BridgedApp {
                         case TIMED_OUT_QUEUED -> "queued";
                         case BUSY -> "busy";
                         case WORKER_FAILED -> "failed";
+                        case BACKEND_EXHAUSTED -> "backend_exhausted";
                         default -> "done"; // unreachable (terminal outcomes handled above)
                     },
-                    "detail", reply.outcome() == MessageService.Outcome.WORKER_FAILED && reply.text() != null
+                    "detail", (reply.outcome() == MessageService.Outcome.WORKER_FAILED
+                                    || reply.outcome() == MessageService.Outcome.BACKEND_EXHAUSTED)
+                                    && reply.text() != null
                             ? reply.text()
                             : "no reply within " + timeout + "ms; poll status or retry"));
         }

bridged/src/main/java/dev/ltms/bridged/session/GitWorktrees.java

@@ -167,6 +167,22 @@ public final class GitWorktrees implements Worktrees {
         exec("git", "-C", repoRoot, "worktree", "remove", "--force", worktreePath);
     }
 
+    @Override
+    public boolean hasUncommitted(String worktreePath) {
+        // A worktree that is already gone holds no work to lose, and it must not break teardown:
+        // git -C <missing-dir> status exits non-zero and would throw where release() is mid-way
+        // through stopping a pane. Mirror remove()'s already-gone tolerance by treating it as clean.
+        Path p = Path.of(worktreePath);
+        if (!Files.exists(p)) {
+            log.debug("worktree {} already gone — nothing can be uncommitted", worktreePath);
+            return false;
+        }
+        // No --untracked-files=no: the exact shape of the work lost in CB-576 was a new file
+        // that was never added, so an untracked-only worktree is still dirty.
+        String out = exec("git", "-C", worktreePath, "status", "--porcelain");
+        return !out.isBlank();
+    }
+
     @Override
     public void overlayParity(String repoRoot, String worktreePath, List<String> overlay) {
         if (overlay == null || overlay.isEmpty()) {

bridged/src/main/java/dev/ltms/bridged/session/SessionManager.java

@@ -4,6 +4,7 @@ import dev.ltms.bridged.auth.MemberLifecycle;
 import dev.ltms.bridged.herdr.Agent;
 import dev.ltms.bridged.inject.TurnListener;
 import dev.ltms.bridged.inject.MemberPresence;
+import dev.ltms.bridged.msg.TurnToken;
 import dev.ltms.bridged.peer.MemberRole;
 import dev.ltms.bridged.peer.PeerHandle;
 import dev.ltms.bridged.peer.PeerLauncher;
@@ -200,6 +201,16 @@ public final class SessionManager implements TurnListener {
                     removed.paneId(), removed.terminalId(), removed.state(), cause);
             if (preserveWorktree && removed.worktree() != null) {
                 logPreservedForShutdown(removed);
+            } else if (removed.worktree() != null && worktrees.hasUncommitted(removed.worktree())) {
+                // CB-576: a release that would otherwise remove the worktree finds it holding
+                // uncommitted work the bridge cannot see. A worker that ends a turn without
+                // committing (normally because it stopped to ask a question or refused the turn)
+                // has its only copy of that work in the worktree. Remove would --force-delete it,
+                // so preserve the directory and tell an operator where to find it.
+                preserveWorktree = true;
+                log.warn("release {} preserves dirty worktree {} for pane={} terminal={}: "
+                        + "the worktree holds uncommitted changes that --force remove would destroy",
+                        cause, removed.worktree(), removed.paneId(), removed.terminalId());
             }
             // CB-516: a send still waiting on this worker can never be answered now. Tell the
             // listener BEFORE the pane is torn down, so a blocked caller fails fast with a real
@@ -425,7 +436,7 @@ public final class SessionManager implements TurnListener {
      * can be re-delivered for multi-turn reuse until it is released.
      */
     @Override
-    public void onDelivered(String target) {
+    public void onDelivered(String target, TurnToken token) {
         MemberSession current = findByTerminal(target);
         if (current == null) return;
         if (current.state() != MemberSession.State.READY && current.state() != MemberSession.State.DONE) {

bridged/src/main/java/dev/ltms/bridged/session/Worktrees.java

@@ -10,6 +10,18 @@ public interface Worktrees {
     /** git -C <repoRoot> worktree remove --force <path>. Idempotent (already-gone tolerated). */
     void remove(String repoRoot, String worktreePath);
 
+    /**
+     * True when the worktree holds uncommitted changes the bridge cannot see: tracked
+     * modifications, staged files, or untracked files. {@code git status --porcelain} is the
+     * test; an empty result means clean. Callers use this to decide whether removing the
+     * worktree would silently destroy a worker's only copy of its work.
+     *
+     * <p>An already-gone worktree is reported as clean (no throw), matching {@link #remove}'s
+     * idempotent contract: a path that does not exist holds no work to lose, and must not break
+     * a teardown that is mid-way through stopping the pane.
+     */
+    boolean hasUncommitted(String worktreePath);
+
     /** Copy each existing overlay path repoRoot→worktree; mark tracked ones --skip-worktree. */
     void overlayParity(String repoRoot, String worktreePath, List<String> overlay);
 

bridged/src/main/resources/logback.xml

@@ -1,4 +1,13 @@
 <configuration>
+    <!--
+      CB-575: MCP SDK 2.0.0 has no public notification registration API. It registers only
+      notifications/initialized and notifications/roots/list_changed, so other client notifications
+      still warn when unhandled. Clients may legitimately send notifications/cancelled; suppress only
+      that SDK WARN because it would devalue the action-needed WARN level used by M4 fleet health.
+      If a later SDK handles cancellation, this filter simply stops matching and can be removed.
+    -->
+    <turboFilter class="dev.ltms.bridged.logging.McpCancelledNotificationFilter"/>
+
     <appender name="STDOUT" class="ch.qos.logback.core.ConsoleAppender">
         <encoder>
             <pattern>%d{HH:mm:ss.SSS} %-5level [%thread] %logger{28} - %msg%n</pattern>

bridged/src/test/java/dev/ltms/bridged/config/BridgedConfigTest.java

@@ -194,7 +194,7 @@ class BridgedConfigTest {
                 fleet:
                   leaders:
                     opus:
-                      terminal: term_opus
+                      tab: "lead: opus"
                 """);
 
         BridgedConfig.Leader lead = BridgedConfig.load(f).fleet().leaders().get("opus");
@@ -212,7 +212,7 @@ class BridgedConfigTest {
                 fleet:
                   leaders:
                     opus:
-                      terminal: term_opus
+                      tab: "drive: opus"
                       tabPrefix: "drive:"
                       scanIntervalSeconds: 30
                 """);
@@ -224,7 +224,9 @@ class BridgedConfigTest {
 
     /**
      * The pane no longer has to exist before the daemon does (CB-557): a lead naming a profile may
-     * be launched, while one that names only a terminal is recognised and never created.
+     * be launched, while one that names no profile is recognised and never created. Either way it
+     * still needs its own {@code tab:} (CB-579) — that part is unconditional, see
+     * {@link #aLeadWithNoTabRefusesToStart}.
      */
     @Test
     void aLeadIsCreatableOnlyWhenItNamesAProfile(@TempDir Path dir) throws Exception {
@@ -239,8 +241,9 @@ class BridgedConfigTest {
                   leaders:
                     launched:
                       profile: opus
+                      tab: "lead: launched"
                     pinned:
-                      terminal: term_opus
+                      tab: "lead: pinned"
                 """);
 
         var leaders = BridgedConfig.load(f).fleet().leaders();
@@ -249,8 +252,12 @@ class BridgedConfigTest {
                 "no profile to launch on ⇒ recognise-only, the pre-CB-557 behaviour");
     }
 
+    /**
+     * CB-579: {@code tab} is the only field a lead's identity depends on now, so it is required
+     * whether the entry is creatable or recognise-only — without it the entry can never be found.
+     */
     @Test
-    void aLeadThatCanBeNeitherFoundNorCreatedRefusesToStart(@TempDir Path dir) throws Exception {
+    void aLeadWithNoTabRefusesToStart(@TempDir Path dir) throws Exception {
         Path f = dir.resolve("useless-lead.yaml");
         Files.writeString(f, """
                 bind:
@@ -264,6 +271,80 @@ class BridgedConfigTest {
 
         IllegalStateException e = assertThrows(IllegalStateException.class, cfg::validateMembers);
         assertTrue(e.getMessage().contains("ghost"), "the message must name the useless entry");
+        assertTrue(e.getMessage().contains("tab:"), "the message must say what is missing");
+    }
+
+    /**
+     * CB-579 acceptance (2): a config still spelling {@code fleet.leaders.<name>.terminal} must fail
+     * loudly at load, not be silently dropped by {@code Leader}'s {@code @JsonIgnoreProperties}.
+     */
+    @Test
+    void aLeaderTerminalKeyFailsLoadAndNamesTabAsTheReplacement(@TempDir Path dir) throws Exception {
+        Path f = dir.resolve("stale-terminal.yaml");
+        Files.writeString(f, """
+                bind:
+                  port: 8080
+                fleet:
+                  leaders:
+                    opus:
+                      terminal: term_opus
+                """);
+
+        IllegalStateException e =
+                assertThrows(IllegalStateException.class, () -> BridgedConfig.load(f));
+        assertTrue(e.getMessage().contains("opus"), "the message must name the offending entry");
+        assertTrue(e.getMessage().contains("tab:"), "the message must name the replacement key");
+        assertTrue(e.getMessage().contains("terminal"), "the message must name the retired key");
+    }
+
+    /** The same refusal, and it must name every offending entry, not just the first. */
+    @Test
+    void everyLeaderStillUsingTerminalIsReportedAtOnce(@TempDir Path dir) throws Exception {
+        Path f = dir.resolve("stale-terminals.yaml");
+        Files.writeString(f, """
+                bind:
+                  port: 8080
+                fleet:
+                  leaders:
+                    opus:
+                      terminal: term_opus
+                    sol:
+                      terminal: term_sol
+                """);
+
+        IllegalStateException e =
+                assertThrows(IllegalStateException.class, () -> BridgedConfig.load(f));
+        assertTrue(e.getMessage().contains("opus"));
+        assertTrue(e.getMessage().contains("sol"));
+    }
+
+    /** A {@code terminal:} anywhere else in the document (not under a leader entry) is unaffected. */
+    @Test
+    void aTerminalKeyOutsideFleetLeadersIsNotRejected(@TempDir Path dir) throws Exception {
+        Path f = dir.resolve("primary-terminal-ok.yaml");
+        Files.writeString(f, "bind:\n  port: 8080\nprimary:\n  terminal: term_fixed\n");
+
+        assertDoesNotThrow(() -> BridgedConfig.load(f));
+    }
+
+    /** CB-579 acceptance (3): distinct `tab:` labels need no shared prefix — one scanner finds both. */
+    @Test
+    void twoLeadersWithDifferentTabsAreBothConfigured(@TempDir Path dir) throws Exception {
+        Path f = dir.resolve("two-tabs.yaml");
+        Files.writeString(f, """
+                bind:
+                  port: 8080
+                fleet:
+                  leaders:
+                    opus:
+                      tab: "lead: opus"
+                    sol:
+                      tab: "captain: sol"
+                """);
+
+        var leaders = BridgedConfig.load(f).fleet().leaders();
+        assertEquals("lead: opus", leaders.get("opus").tab());
+        assertEquals("captain: sol", leaders.get("sol").tab());
     }
 
     // ── CB-551: the idle-lead heartbeat ─────────────────────────────────────────────────────────
@@ -328,7 +409,7 @@ class BridgedConfigTest {
                 fleet:
                   leaders:
                     opus:
-                      terminal: term_opus
+                      tab: "lead: opus"
                       tabPrefix: "lead:"
                 """);
         BridgedConfig cfg = BridgedConfig.load(f);
@@ -349,7 +430,7 @@ class BridgedConfigTest {
                   tabLabel: "lead: {role} {profile}"
                   leaders:
                     opus:
-                      terminal: term_opus
+                      tab: "lead: opus"
                 """);
         BridgedConfig cfg = BridgedConfig.load(f);
 
@@ -374,7 +455,7 @@ class BridgedConfigTest {
                 fleet:
                   leaders:
                     opus:
-                      terminal: term_opus
+                      tab: "lead: opus"
                 """);
 
         assertDoesNotThrow(() -> BridgedConfig.load(f).validateLeadTabPrefixes());
@@ -400,7 +481,7 @@ class BridgedConfigTest {
                 "a label that collides with a convention nobody reads is not a problem");
     }
 
-    // ── CB-530: the leaders registry ────────────────────────────────────────────────────────────
+    // ── CB-530/CB-579: the leaders registry ─────────────────────────────────────────────────────
 
     @Test
     void leadersBlockRegistersEveryPaneByName(@TempDir Path dir) throws Exception {
@@ -411,10 +492,10 @@ class BridgedConfigTest {
                 fleet:
                   leaders:
                     opus-5.0:
-                      terminal: term_opus
+                      tab: "lead: opus-5.0"
                       kind: claude
                     gpt-sol-5.6:
-                      terminal: term_sol
+                      tab: "lead: gpt-sol-5.6"
                       kind: opencode
                       model: openai/gpt-5.6-terra
                 """);
@@ -425,9 +506,9 @@ class BridgedConfigTest {
         assertEquals(Set.of("opus-5.0", "gpt-sol-5.6"), leaders.keySet());
         assertEquals("opencode", leaders.get("gpt-sol-5.6").kind());
         assertEquals("openai/gpt-5.6-terra", leaders.get("gpt-sol-5.6").model());
-        // The whole point: BOTH panes resolve as leads, so neither is demoted to worker.
-        assertEquals(Map.of("term_opus", "opus-5.0", "term_sol", "gpt-sol-5.6"),
-                cfg.leaderTerminals());
+        // Identity is the tab now (CB-579) — both entries carry their own, distinct label.
+        assertEquals("lead: opus-5.0", leaders.get("opus-5.0").tab());
+        assertEquals("lead: gpt-sol-5.6", leaders.get("gpt-sol-5.6").tab());
     }
 
     @Test
@@ -439,42 +520,6 @@ class BridgedConfigTest {
                 "configs that never migrate must behave exactly as they did before CB-530");
     }
 
-    @Test
-    void anExplicitLeadersEntryWinsOverThePinForTheSameTerminal(@TempDir Path dir) throws Exception {
-        Path f = dir.resolve("both.yaml");
-        Files.writeString(f, """
-                bind:
-                  port: 8080
-                primary:
-                  terminal: term_shared
-                fleet:
-                  leaders:
-                    opus-5.0:
-                      terminal: term_shared
-                """);
-
-        assertEquals(Map.of("term_shared", "opus-5.0"), BridgedConfig.load(f).leaderTerminals(),
-                "the pin is the older spelling of the same fact; the named entry is what was meant");
-    }
-
-    @Test
-    void bothBlocksTogetherRegisterTheUnionOfTheirTerminals(@TempDir Path dir) throws Exception {
-        Path f = dir.resolve("union.yaml");
-        Files.writeString(f, """
-                bind:
-                  port: 8080
-                primary:
-                  terminal: term_pinned
-                fleet:
-                  leaders:
-                    gpt-sol-5.6:
-                      terminal: term_sol
-                """);
-
-        assertEquals(Map.of("term_pinned", "primary", "term_sol", "gpt-sol-5.6"),
-                BridgedConfig.load(f).leaderTerminals());
-    }
-
     @Test
     void neitherBlockLeavesNothingRegistered(@TempDir Path dir) throws Exception {
         Path f = dir.resolve("none.yaml");
@@ -483,22 +528,25 @@ class BridgedConfigTest {
         assertTrue(BridgedConfig.load(f).leaderTerminals().isEmpty());
     }
 
-    /** A lead entry with no terminal identifies nothing — it must not register a null key. */
+    /**
+     * CB-579: {@code fleet.leaders} no longer feeds {@code leaderTerminals()} at all — a lead's
+     * identity comes from the live tab scan, not a config-held terminal map. This method now exists
+     * only for the {@code primary.terminal} fallback.
+     */
     @Test
-    void aLeadWithoutATerminalIsNotRegistered(@TempDir Path dir) throws Exception {
-        Path f = dir.resolve("no-terminal.yaml");
+    void fleetLeadersNeverContributesToLeaderTerminals(@TempDir Path dir) throws Exception {
+        Path f = dir.resolve("leaders-only.yaml");
         Files.writeString(f, """
                 bind:
                   port: 8080
                 fleet:
                   leaders:
-                    sketch:
-                      kind: opencode
-                    real:
-                      terminal: term_real
+                    opus-5.0:
+                      tab: "lead: opus-5.0"
                 """);
 
-        assertEquals(Map.of("term_real", "real"), BridgedConfig.load(f).leaderTerminals());
+        assertTrue(BridgedConfig.load(f).leaderTerminals().isEmpty(),
+                "no primary.terminal pin ⇒ nothing registered, even with fleet.leaders configured");
     }
 
     // ── CB-548: the architects registry ────────────────────────────────────────────────────────
@@ -1092,6 +1140,7 @@ class BridgedConfigTest {
                     gitHostEnv: GITEA_HOST
                     weight: 0.5
                     maxLoad: 2
+                    exhaustedPattern: "usage limit has been reached"
                 placement: weighted
                 lifecycle:
                   idleTtlSeconds: 300
@@ -1123,6 +1172,8 @@ class BridgedConfigTest {
         assertEquals("GITEA_HOST", w.gitHostEnv());
         assertEquals(0.5f, w.weight(), 0.0001f, "weight binds as a float");
         assertEquals(2, w.maxLoad(), "maxLoad binds as an integer");
+        assertTrue(w.hasExhaustedPattern(), "exhaustedPattern binds and enables the CB-578 stage A classification");
+        assertEquals("usage limit has been reached", w.exhaustedPattern());
 
         assertEquals("weighted", cfg.placement(), "placement binds at the top level");
         assertEquals(300, cfg.lifecycle().idleTtlSeconds());

bridged/src/test/java/dev/ltms/bridged/health/FleetHealthMonitorTest.java

@@ -0,0 +1,170 @@
+package dev.ltms.bridged.health;
+
+import ch.qos.logback.classic.Logger;
+import ch.qos.logback.classic.spi.ILoggingEvent;
+import ch.qos.logback.core.read.ListAppender;
+import dev.ltms.bridged.herdr.AgentControl;
+import dev.ltms.bridged.herdr.FakeHerdr;
+import dev.ltms.bridged.inject.Injector;
+import dev.ltms.bridged.msg.InMemoryReplyInbox;
+import dev.ltms.bridged.msg.MessageService;
+import dev.ltms.bridged.msg.Rendezvous;
+import dev.ltms.bridged.session.SessionManager;
+import dev.ltms.bridged.member.ClaudeCodeLauncher;
+import dev.ltms.bridged.herdr.WorkspaceControl;
+import dev.ltms.bridged.guard.SubscriptionGuard;
+import dev.ltms.bridged.config.BridgedConfig;
+import org.junit.jupiter.api.Test;
+import org.slf4j.LoggerFactory;
+
+import java.util.Map;
+import java.util.Set;
+import java.util.concurrent.Executors;
+import java.util.function.BiConsumer;
+
+import static org.junit.jupiter.api.Assertions.assertEquals;
+import static org.junit.jupiter.api.Assertions.assertTrue;
+
+class FleetHealthMonitorTest {
+    @Test void oneTickUsesOneFleetListForAnyRosterSize() {
+        FakeHerdr herdr = new FakeHerdr();
+        AgentControl agents = new AgentControl(herdr);
+        BridgedConfig.Profile profile = new BridgedConfig.Profile("test", "http://test:1", null,
+                null, null, null, null, null, null, null, null, null);
+        ClaudeCodeLauncher launcher = new ClaudeCodeLauncher(agents, new WorkspaceControl(herdr),
+                new SubscriptionGuard(Set.of("test")), Map.of("test", profile), "test", _ -> "token");
+        SessionManager sessions = new SessionManager(launcher);
+        sessions.acquire("test", null, null, null);
+        sessions.acquire("test", null, null, null);
+        herdr.calls.clear();
+        MessageService messages = new MessageService(agents, new Injector(agents), new Rendezvous(), new InMemoryReplyInbox());
+        var scheduler = Executors.newSingleThreadScheduledExecutor();
+        FleetHealthMonitor monitor = new FleetHealthMonitor(agents, sessions::roster, messages, scheduler, () -> 1, 60,
+                (_, _) -> { });
+        monitor.tick();
+        monitor.stop();
+        assertEquals(1, herdr.calls.stream().filter(call -> call.method().equals("agent.list")).count());
+    }
+
+    @Test void failedTickDoesNotStopTheNextTick() {
+        FakeHerdr herdr = new FakeHerdr().healthy(false);
+        AgentControl agents = new AgentControl(herdr);
+        var scheduler = Executors.newSingleThreadScheduledExecutor();
+        FleetHealthMonitor monitor = new FleetHealthMonitor(agents, java.util.List::of,
+                new MessageService(agents, new Injector(agents), new Rendezvous(), new InMemoryReplyInbox()),
+                scheduler, () -> 1, 60, (_, _) -> { });
+        monitor.tick();
+        herdr.healthy(true);
+        monitor.tick();
+        monitor.stop();
+        assertEquals(2, herdr.calls.stream().filter(call -> call.method().equals("agent.list")).count());
+    }
+
+    @Test void faultTransitionLogsOnlyOnceUntilItChanges() {
+        Logger logger = (Logger) LoggerFactory.getLogger(FleetHealthMonitor.class);
+        ListAppender<ILoggingEvent> appender = new ListAppender<>();
+        appender.start();
+        logger.addAppender(appender);
+        try {
+            FakeHerdr herdr = new FakeHerdr();
+            AgentControl agents = new AgentControl(herdr);
+            var scheduler = Executors.newSingleThreadScheduledExecutor();
+            FleetHealthMonitor monitor = new FleetHealthMonitor(agents, java.util.List::of,
+                    new MessageService(agents, new Injector(agents), new Rendezvous(), new InMemoryReplyInbox()),
+                    scheduler, () -> 1, 60, (_, _) -> { });
+            monitor.reportTransition("term_a", HealthState.TURN_BOUNDARY_LOST);
+            monitor.reportTransition("term_a", HealthState.TURN_BOUNDARY_LOST);
+            monitor.stop();
+            assertEquals(1, appender.list.stream().filter(event -> event.getFormattedMessage()
+                    .contains("member=term_a state=TURN_BOUNDARY_LOST")).count());
+        } finally {
+            logger.detachAppender(appender);
+        }
+    }
+
+    // --- CB-580: a member that reaches GONE/NEVER_READY must fail its waiting tickets
+
+    private static FleetHealthMonitor monitorWith(BiConsumer<String, String> failTarget) {
+        FakeHerdr herdr = new FakeHerdr();
+        AgentControl agents = new AgentControl(herdr);
+        var scheduler = Executors.newSingleThreadScheduledExecutor();
+        return new FleetHealthMonitor(agents, java.util.List::of,
+                new MessageService(agents, new Injector(agents), new Rendezvous(), new InMemoryReplyInbox()),
+                scheduler, () -> 1, 60, failTarget);
+    }
+
+    @Test void terminalTransitionFailsTheTargetOnce() {
+        RecordingFailTarget failTarget = new RecordingFailTarget();
+        FleetHealthMonitor monitor = monitorWith(failTarget);
+        monitor.reportTransition("term_a", HealthState.GONE);
+        monitor.stop();
+        assertEquals(1, failTarget.calls.size());
+        assertEquals("term_a", failTarget.calls.get(0).target());
+        assertTrue(failTarget.calls.get(0).reason().contains("GONE"));
+    }
+
+    @Test void neverReadyNamesItselfAsTheReason() {
+        RecordingFailTarget failTarget = new RecordingFailTarget();
+        FleetHealthMonitor monitor = monitorWith(failTarget);
+        monitor.reportTransition("term_a", HealthState.NEVER_READY);
+        monitor.stop();
+        assertEquals(1, failTarget.calls.size());
+        assertTrue(failTarget.calls.get(0).reason().contains("NEVER_READY"));
+    }
+
+    @Test void stayingInATerminalStateProducesOneFailureNotN() {
+        RecordingFailTarget failTarget = new RecordingFailTarget();
+        FleetHealthMonitor monitor = monitorWith(failTarget);
+        monitor.reportTransition("term_a", HealthState.GONE);
+        monitor.reportTransition("term_a", HealthState.GONE);
+        monitor.reportTransition("term_a", HealthState.GONE);
+        monitor.reportTransition("term_a", HealthState.GONE);
+        monitor.stop();
+        assertEquals(1, failTarget.calls.size());
+    }
+
+    @Test void aNonTerminalFaultStateDoesNotFailTheTarget() {
+        RecordingFailTarget failTarget = new RecordingFailTarget();
+        FleetHealthMonitor monitor = monitorWith(failTarget);
+        monitor.reportTransition("term_a", HealthState.TURN_BOUNDARY_LOST);
+        monitor.stop();
+        assertEquals(0, failTarget.calls.size());
+    }
+
+    @Test void failTargetRetryIsBounded() {
+        AlwaysThrowingFailTarget failTarget = new AlwaysThrowingFailTarget();
+        FleetHealthMonitor monitor = monitorWith(failTarget);
+        monitor.reportTransition("term_a", HealthState.GONE);
+        monitor.stop();
+        assertEquals(FleetHealthMonitor.MAX_FAIL_TARGET_ATTEMPTS, failTarget.calls);
+    }
+
+    @Test void exhaustedRetryStillDoesNotRefireOnAnUnchangedTick() {
+        AlwaysThrowingFailTarget failTarget = new AlwaysThrowingFailTarget();
+        FleetHealthMonitor monitor = monitorWith(failTarget);
+        monitor.reportTransition("term_a", HealthState.GONE);
+        int afterFirstTransition = failTarget.calls;
+        monitor.reportTransition("term_a", HealthState.GONE);
+        monitor.stop();
+        assertEquals(afterFirstTransition, failTarget.calls);
+    }
+
+    private record RecordedCall(String target, String reason) { }
+
+    private static final class RecordingFailTarget implements BiConsumer<String, String> {
+        final java.util.List<RecordedCall> calls = new java.util.ArrayList<>();
+
+        @Override public void accept(String target, String reason) {
+            calls.add(new RecordedCall(target, reason));
+        }
+    }
+
+    private static final class AlwaysThrowingFailTarget implements BiConsumer<String, String> {
+        int calls = 0;
+
+        @Override public void accept(String target, String reason) {
+            calls++;
+            throw new RuntimeException("boom");
+        }
+    }
+}

bridged/src/test/java/dev/ltms/bridged/health/FleetHealthTest.java

@@ -0,0 +1,56 @@
+package dev.ltms.bridged.health;
+
+import dev.ltms.bridged.herdr.AgentStatus;
+import dev.ltms.bridged.session.MemberSession;
+import dev.ltms.bridged.msg.Rendezvous;
+import org.junit.jupiter.api.Test;
+
+import static org.junit.jupiter.api.Assertions.assertEquals;
+
+class FleetHealthTest {
+    @Test void muteCountsOnlyCompletionFallbacks() {
+        MuteCounter mute = new MuteCounter();
+        mute.observe("target", "terra", Rendezvous.Kind.REPLY);
+        mute.observe("target", "terra", Rendezvous.Kind.COMPLETION);
+        assertEquals(1, mute.forTarget("target"));
+        assertEquals(1, mute.forProfile("terra"));
+    }
+    @Test void turnBoundaryNeedsTwoSnapshots() {
+        HealthSnapshot s = snapshot(MemberSession.State.BUSY, AgentStatus.DONE, true);
+        HealthDecision first = FleetHealth.decide(s, HealthPrior.NONE, 1);
+        assertEquals(HealthState.WORKING, first.state());
+        assertEquals(new HealthPrior(true), first.prior());
+        assertEquals(HealthState.TURN_BOUNDARY_LOST, FleetHealth.decide(s, first.prior(), 2).state());
+    }
+
+    @Test void unknownLiveStatusWithAcceptedDeliveryIsNotIdle() {
+        assertEquals(HealthState.WORKING, FleetHealth.decide(
+                snapshot(MemberSession.State.BUSY, AgentStatus.UNKNOWN, true), HealthPrior.NONE, 1).state());
+    }
+
+    @Test void acceptedDeliveryNeverReportsIdle() {
+        for (MemberSession.State session : MemberSession.State.values()) {
+            for (AgentStatus live : AgentStatus.values()) {
+                HealthSnapshot s = snapshot(session, live, true);
+                assertEquals(false, FleetHealth.decide(s, HealthPrior.NONE, 1).state() == HealthState.IDLE,
+                        () -> "accepted delivery returned IDLE for " + session + "/" + live);
+            }
+        }
+    }
+
+    @Test void blockedDoesNotGuessPromptKind() {
+        assertEquals(HealthState.BLOCKED_AMBIGUOUS, FleetHealth.decide(
+                snapshot(MemberSession.State.BUSY, AgentStatus.BLOCKED, true), HealthPrior.NONE, 1).state());
+    }
+
+    @Test void controlLinkOutranksMemberFault() {
+        HealthSnapshot s = new HealthSnapshot(MemberSession.State.BUSY, AgentStatus.DONE, true, false,
+                false, true, true, true, false, true, true, true);
+        assertEquals(HealthState.CONTROL_LINK_DOWN, FleetHealth.decide(s, HealthPrior.NONE, 1).state());
+    }
+
+    private static HealthSnapshot snapshot(MemberSession.State state, AgentStatus live, boolean accepted) {
+        return new HealthSnapshot(state, live, accepted, false, false, true, false, false,
+                false, false, false, false);
+    }
+}

bridged/src/test/java/dev/ltms/bridged/health/PaneBudgetTest.java

@@ -0,0 +1,14 @@
+package dev.ltms.bridged.health;
+
+import org.junit.jupiter.api.Test;
+import java.util.List;
+import static org.junit.jupiter.api.Assertions.assertEquals;
+
+class PaneBudgetTest {
+    @Test void fixedLimitsIgnoreWeakerConfig() {
+        PaneBudget budget = new PaneBudget();
+        List<String> targets = List.of("a", "b", "c");
+        assertEquals(List.of("a", "b"), budget.choose(targets, 0, 0));
+        assertEquals(List.of("c"), budget.choose(targets, 1, 0));
+    }
+}

bridged/src/test/java/dev/ltms/bridged/herdr/LeadTabScannerTest.java

@@ -15,8 +15,9 @@ import java.util.concurrent.atomic.AtomicLong;
 import static org.junit.jupiter.api.Assertions.*;
 
 /**
- * CB-531. A lead is never spawned, so the daemon has to <em>find</em> it: these assert that an
- * operator-labelled tab is what makes a pane a lead, and — just as importantly — what does not.
+ * CB-531/CB-579. A lead is never spawned, so the daemon has to <em>find</em> it: these assert that
+ * an operator-labelled tab matching a configured {@code tab:} is what makes a pane a lead, and —
+ * just as importantly — what does not, and that a stale entry does not linger forever.
  */
 class LeadTabScannerTest {
 
@@ -120,23 +121,48 @@ class LeadTabScannerTest {
                 .pane("w9:p1", "w9:t1", "term_worker");
     }
 
-    private LeadTabScanner scanner(TopologyHerdr herdr, Map<String, String> configured,
+    /** The {@code tab:} → name map {@code twoLeads()}'s two lead tabs are configured under. */
+    private static Map<String, String> twoLeadsConfigured() {
+        return Map.of("lead: opus-5.0", "opus-5.0", "lead: gpt-sol-5.6", "gpt-sol-5.6");
+    }
+
+    private LeadTabScanner scanner(TopologyHerdr herdr, Map<String, String> tabToName,
                                    AtomicLong clock) {
-        return new LeadTabScanner(herdr, "lead:", Set.of("bridged-workers"), configured, TTL,
-                clock::get);
+        return new LeadTabScanner(herdr, tabToName, Set.of("bridged-workers"), TTL, clock::get);
     }
 
     @Test
-    void everyLabelledTabBecomesALeadNamedByItsLabel() {
-        Map<String, String> leads = scanner(twoLeads(), Map.of(), new AtomicLong()).get();
+    void everyConfiguredTabBecomesALeadNamedByItsEntry() {
+        Map<String, String> leads = scanner(twoLeads(), twoLeadsConfigured(), new AtomicLong()).get();
 
         assertEquals(Map.of("term_opus", "opus-5.0", "term_gpt", "gpt-sol-5.6"), leads,
-                "two leads discovered from labels alone — no terminal_id was ever configured");
+                "two leads discovered by their configured tab — no terminal_id was ever configured");
     }
 
     @Test
-    void anUnlabelledTabContributesNothing() {
-        assertFalse(scanner(twoLeads(), Map.of(), new AtomicLong()).get().containsKey("term_notes"));
+    void anUnconfiguredTabContributesNothing() {
+        assertFalse(scanner(twoLeads(), twoLeadsConfigured(), new AtomicLong())
+                .get().containsKey("term_notes"));
+    }
+
+    /**
+     * CB-579: matching is exact against the configured map now, not a shared prefix — two leads with
+     * completely different labels are both discovered by one scanner, no convention required.
+     */
+    @Test
+    void twoLeadsWithCompletelyDifferentLabelsAreBothDiscovered() {
+        TopologyHerdr herdr = new TopologyHerdr()
+                .workspace("w1", "main")
+                .tab("w1:t1", "w1", "orchestrator: opus")
+                .tab("w1:t2", "w1", "captain: sol")
+                .pane("w1:p1", "w1:t1", "term_opus")
+                .pane("w1:p2", "w1:t2", "term_sol");
+        Map<String, String> tabToName = Map.of("orchestrator: opus", "opus", "captain: sol", "sol");
+
+        Map<String, String> leads = scanner(herdr, tabToName, new AtomicLong()).get();
+
+        assertEquals(Map.of("term_opus", "opus", "term_sol", "sol"), leads,
+                "no shared prefix needed — each lead is matched by its own configured tab");
     }
 
     /**
@@ -148,25 +174,28 @@ class LeadTabScannerTest {
     void aTabInAWorkerSpaceIsNeverALeadEvenWhenItsLabelMatches() {
         TopologyHerdr herdr = twoLeads().tab("w9:t2", "w9", "lead: impostor")
                 .pane("w9:p2", "w9:t2", "term_impostor");
+        Map<String, String> tabToName = new LinkedHashMap<>(twoLeadsConfigured());
+        tabToName.put("lead: impostor", "impostor");
 
-        assertFalse(scanner(herdr, Map.of(), new AtomicLong()).get().containsKey("term_impostor"));
+        assertFalse(scanner(herdr, tabToName, new AtomicLong()).get().containsKey("term_impostor"));
     }
 
     @Test
-    void aBarePrefixNamesNobodyAndIsRejected() {
+    void aLabelWithNoConfiguredEntryIsIgnored() {
         TopologyHerdr herdr = new TopologyHerdr().workspace("w1", "main")
-                .tab("w1:t1", "w1", "lead:").pane("w1:p1", "w1:t1", "term_a");
+                .tab("w1:t1", "w1", "lead: nobody-configured").pane("w1:p1", "w1:t1", "term_a");
 
-        assertEquals(Map.of(), scanner(herdr, Map.of(), new AtomicLong()).get(),
-                "a lead with no name would resolve as PRIMARY with nothing to attribute it to");
+        assertEquals(Map.of(), scanner(herdr, twoLeadsConfigured(), new AtomicLong()).get(),
+                "a label that names no configured lead resolves nobody");
     }
 
     @Test
-    void thePrefixMatchesCaseInsensitivelyAndTheNameIsTrimmed() {
+    void matchingIsCaseInsensitiveAndToleratesSurroundingWhitespace() {
         TopologyHerdr herdr = new TopologyHerdr().workspace("w1", "main")
-                .tab("w1:t1", "w1", "  LEAD:   opus-5.0  ").pane("w1:p1", "w1:t1", "term_a");
+                .tab("w1:t1", "w1", "  LEAD: Opus-5.0  ").pane("w1:p1", "w1:t1", "term_a");
 
-        assertEquals(Map.of("term_a", "opus-5.0"), scanner(herdr, Map.of(), new AtomicLong()).get());
+        assertEquals(Map.of("term_a", "opus-5.0"),
+                scanner(herdr, Map.of("lead: Opus-5.0", "opus-5.0"), new AtomicLong()).get());
     }
 
     @Test
@@ -175,18 +204,51 @@ class LeadTabScannerTest {
         // nothing bridged placed can land here (see the worker-space test above).
         TopologyHerdr herdr = twoLeads().pane("w1:p1b", "w1:t1", "term_opus_split");
 
-        assertEquals("opus-5.0", scanner(herdr, Map.of(), new AtomicLong()).get().get("term_opus_split"));
+        assertEquals("opus-5.0",
+                scanner(herdr, twoLeadsConfigured(), new AtomicLong()).get().get("term_opus_split"));
     }
 
+    /**
+     * CB-579 acceptance (6): this is the bug the ticket closes. A stale pin used to be merged back
+     * over every scan and never expire; now a scan is the whole answer, so a lead whose tab is gone
+     * drops out on the very next scan.
+     */
     @Test
-    void anExplicitlyConfiguredLeadIsMergedInAndOutranksALabel() {
-        Map<String, String> configured = Map.of("term_opus", "pinned-name", "term_extra", "from-config");
+    void aTabNoLongerPresentDropsTheLeadOnTheNextScan() {
+        TopologyHerdr herdr = twoLeads();
+        AtomicLong clock = new AtomicLong();
+        LeadTabScanner s = scanner(herdr, twoLeadsConfigured(), clock);
+        assertTrue(s.get().containsKey("term_opus"));
 
-        Map<String, String> leads = scanner(twoLeads(), configured, new AtomicLong()).get();
+        // The session behind term_opus restarted — herdr no longer reports that tab or pane at all.
+        herdr.tabs.remove("w1:t1");
+        herdr.panes.remove("w1:p1");
+        clock.addAndGet(TTL);
 
-        assertEquals("pinned-name", leads.get("term_opus"), "an explicit pin is the operator's last word");
-        assertEquals("from-config", leads.get("term_extra"), "a configured lead needs no tab at all");
-        assertEquals("gpt-sol-5.6", leads.get("term_gpt"));
+        assertFalse(s.get().containsKey("term_opus"),
+                "a stale entry must expire once the tab it named is gone, not be merged back forever");
+    }
+
+    /**
+     * CB-579 acceptance (5): the whole point of matching by tab instead of {@code terminal_id} — a
+     * restart changes the terminal, not the tab, so the lead resolves under the same name with no
+     * config edit.
+     */
+    @Test
+    void aLeadRestartingInTheSameTabResolvesUnderTheSameName() {
+        TopologyHerdr herdr = twoLeads();
+        AtomicLong clock = new AtomicLong();
+        LeadTabScanner s = scanner(herdr, twoLeadsConfigured(), clock);
+        assertEquals("opus-5.0", s.get().get("term_opus"));
+
+        // The session restarts: herdr assigns the pane a new terminal_id, same tab (w1:t1).
+        herdr.panes.remove("w1:p1");
+        herdr.pane("w1:p1", "w1:t1", "term_opus_v2");
+        clock.addAndGet(TTL);
+
+        Map<String, String> leads = s.get();
+        assertEquals("opus-5.0", leads.get("term_opus_v2"), "the new terminal resolves immediately");
+        assertFalse(leads.containsKey("term_opus"), "the old terminal_id is simply gone, not carried");
     }
 
     // ── caching ─────────────────────────────────────────────────────────────────────────────────
@@ -195,7 +257,7 @@ class LeadTabScannerTest {
     void aSecondLookupWithinTheTtlDoesNotTouchHerdr() {
         TopologyHerdr herdr = twoLeads();
         AtomicLong clock = new AtomicLong();
-        LeadTabScanner s = scanner(herdr, Map.of(), clock);
+        LeadTabScanner s = scanner(herdr, twoLeadsConfigured(), clock);
 
         s.get();
         int afterFirst = herdr.calls;
@@ -210,21 +272,23 @@ class LeadTabScannerTest {
     void aTabLabelledAfterStartupIsPickedUpOnceTheTtlExpires() {
         TopologyHerdr herdr = twoLeads();
         AtomicLong clock = new AtomicLong();
-        LeadTabScanner s = scanner(herdr, Map.of(), clock);
+        Map<String, String> tabToName = new LinkedHashMap<>(twoLeadsConfigured());
+        tabToName.put("lead: late-arrival", "late-arrival");
+        LeadTabScanner s = scanner(herdr, tabToName, clock);
         assertFalse(s.get().containsKey("term_notes"));
 
         herdr.tab("w1:t3", "w1", "lead: late-arrival"); // the operator renames their tab
         clock.addAndGet(TTL);
 
         assertEquals("late-arrival", s.get().get("term_notes"),
-                "the whole point over `leaders:`: no config edit, no restart");
+                "the whole point over a config-held terminal_id: no config edit, no restart");
     }
 
     @Test
     void aFailedScanKeepsTheLeadsAlreadyKnownRatherThanDemotingThem() {
         TopologyHerdr herdr = twoLeads();
         AtomicLong clock = new AtomicLong();
-        LeadTabScanner s = scanner(herdr, Map.of(), clock);
+        LeadTabScanner s = scanner(herdr, twoLeadsConfigured(), clock);
         Map<String, String> before = s.get();
 
         herdr.failing = true;
@@ -235,14 +299,14 @@ class LeadTabScannerTest {
     }
 
     @Test
-    void aFailedFirstScanStillHonoursTheConfiguredLeads() {
+    void aFailedFirstScanReturnsEmptyRatherThanThrowing() {
         TopologyHerdr herdr = twoLeads();
         herdr.failing = true;
 
-        Map<String, String> leads = scanner(herdr, Map.of("term_x", "opus-5.0"), new AtomicLong()).get();
+        Map<String, String> leads = scanner(herdr, twoLeadsConfigured(), new AtomicLong()).get();
 
-        assertEquals(Map.of("term_x", "opus-5.0"), leads,
-                "config-named leads must not depend on herdr answering at all");
+        assertEquals(Map.of(), leads,
+                "with nothing scanned yet and no override to fall back on, the map is simply empty");
     }
 
     @Test
@@ -250,7 +314,7 @@ class LeadTabScannerTest {
         TopologyHerdr herdr = twoLeads();
         herdr.failing = true;
         AtomicLong clock = new AtomicLong();
-        LeadTabScanner s = scanner(herdr, Map.of(), clock);
+        LeadTabScanner s = scanner(herdr, twoLeadsConfigured(), clock);
 
         s.get();
         int afterFirst = herdr.calls;

bridged/src/test/java/dev/ltms/bridged/inject/CompletionResolverTest.java

@@ -7,9 +7,14 @@ import ch.qos.logback.core.read.ListAppender;
 import dev.ltms.bridged.herdr.AgentControl;
 import dev.ltms.bridged.herdr.FakeHerdr;
 import dev.ltms.bridged.msg.Rendezvous;
+import dev.ltms.bridged.msg.TestTurnTokens;
+import dev.ltms.bridged.msg.TurnToken;
 import org.junit.jupiter.api.Test;
 import org.slf4j.LoggerFactory;
 
+import java.util.Set;
+import java.util.regex.Pattern;
+
 import static org.junit.jupiter.api.Assertions.assertEquals;
 import static org.junit.jupiter.api.Assertions.assertFalse;
 import static org.junit.jupiter.api.Assertions.assertTrue;
@@ -21,7 +26,7 @@ class CompletionResolverTest {
     void skipsTheScrapeWhenNoSendIsWaiting() {
         FakeHerdr herdr = new FakeHerdr();
         Rendezvous rendezvous = new Rendezvous(); // no waiter opened
-        CompletionResolver resolver = new CompletionResolver(new AgentControl(herdr), rendezvous);
+        CompletionResolver resolver = new CompletionResolver(new AgentControl(herdr), rendezvous, ExhaustedPatternLookup.none());
 
         resolver.resolve("term_a", null); // no in-flight turn captured for this target
 
@@ -33,7 +38,7 @@ class CompletionResolverTest {
     void failSkipsTheScrapeWhenNoSendIsWaiting() {
         FakeHerdr herdr = new FakeHerdr();
         Rendezvous rendezvous = new Rendezvous(); // no waiter opened
-        CompletionResolver resolver = new CompletionResolver(new AgentControl(herdr), rendezvous);
+        CompletionResolver resolver = new CompletionResolver(new AgentControl(herdr), rendezvous, ExhaustedPatternLookup.none());
 
         resolver.fail("term_a", null); // no in-flight turn, and no registered waiter to fall back to
 
@@ -45,9 +50,9 @@ class CompletionResolverTest {
     void captureBaselineSkipsTheReadWhenNoSendIsWaiting() {
         FakeHerdr herdr = new FakeHerdr().readText("⏺ X\n❯ ");
         Rendezvous rendezvous = new Rendezvous(); // no waiter opened
-        CompletionResolver resolver = new CompletionResolver(new AgentControl(herdr), rendezvous);
+        CompletionResolver resolver = new CompletionResolver(new AgentControl(herdr), rendezvous, ExhaustedPatternLookup.none());
 
-        resolver.captureBaseline("term_a"); // no send to attribute a later completion to
+        resolver.captureBaseline("term_a", TestTurnTokens.inert("term_a")); // no send to attribute a later completion to
 
         assertFalse(herdr.called("agent.read"),
                 "with no waiting send there is no turn to baseline — skip the scrape");
@@ -134,7 +139,7 @@ class CompletionResolverTest {
         // send must NOT be resolved with the stale answer.
         FakeHerdr herdr = new FakeHerdr().readText("⏺ 391\n❯ ");
         Rendezvous rendezvous = new Rendezvous();
-        CompletionResolver resolver = new CompletionResolver(new AgentControl(herdr), rendezvous);
+        CompletionResolver resolver = new CompletionResolver(new AgentControl(herdr), rendezvous, ExhaustedPatternLookup.none());
 
         var waiter = rendezvous.open("term_a");   // a send is blocked on this turn
         // The turn as captured at delivery: its waiter, and the previous turn's answer still on screen.
@@ -149,7 +154,7 @@ class CompletionResolverTest {
     void resolvesACompletionWhoseScrapeChangedSinceDelivery() {
         FakeHerdr herdr = new FakeHerdr().readText("⏺ No, 391 = 17 × 23.\n❯ "); // the worker's real answer
         Rendezvous rendezvous = new Rendezvous();
-        CompletionResolver resolver = new CompletionResolver(new AgentControl(herdr), rendezvous);
+        CompletionResolver resolver = new CompletionResolver(new AgentControl(herdr), rendezvous, ExhaustedPatternLookup.none());
 
         var waiter = rendezvous.open("term_a");
         // Delivery baseline was the previous turn's "391"; the scrape now differs → resolve.
@@ -166,7 +171,7 @@ class CompletionResolverTest {
         String block = "⏺ " + "x".repeat(CompletionResolver.MAX_SCRAPE_CHARS + 1) + "\n❯ ";
         FakeHerdr herdr = new FakeHerdr().readText(block);
         Rendezvous rendezvous = new Rendezvous();
-        CompletionResolver resolver = new CompletionResolver(new AgentControl(herdr), rendezvous);
+        CompletionResolver resolver = new CompletionResolver(new AgentControl(herdr), rendezvous, ExhaustedPatternLookup.none());
 
         var waiter = rendezvous.open("term_a");
         resolver.resolve("term_a", new CompletionResolver.InFlight(waiter, null));
@@ -180,7 +185,7 @@ class CompletionResolverTest {
     void leavesAnUnclippedCompletionPaneTailUnmarked() {
         FakeHerdr herdr = new FakeHerdr().readText("⏺ complete report\n❯ ");
         Rendezvous rendezvous = new Rendezvous();
-        CompletionResolver resolver = new CompletionResolver(new AgentControl(herdr), rendezvous);
+        CompletionResolver resolver = new CompletionResolver(new AgentControl(herdr), rendezvous, ExhaustedPatternLookup.none());
 
         var waiter = rendezvous.open("term_a");
         resolver.resolve("term_a", new CompletionResolver.InFlight(waiter, null));
@@ -192,9 +197,9 @@ class CompletionResolverTest {
     void resolvesSynchronouslyBeforePostTurnContextClearing() {
         FakeHerdr herdr = new FakeHerdr().readText("⏺ previous answer\n❯ ");
         Rendezvous rendezvous = new Rendezvous();
-        CompletionResolver resolver = new CompletionResolver(new AgentControl(herdr), rendezvous);
+        CompletionResolver resolver = new CompletionResolver(new AgentControl(herdr), rendezvous, ExhaustedPatternLookup.none());
         var waiter = rendezvous.open("term_a");
-        resolver.captureBaseline("term_a");
+        resolver.captureBaseline("term_a", new TurnToken("term_a", waiter));
         herdr.readText("⏺ answer that /clear would erase\n❯ ");
 
         resolver.resolveBeforePostAction("term_a");
@@ -214,10 +219,10 @@ class CompletionResolverTest {
         String longBlock = "⏺ " + "x".repeat(CompletionResolver.MAX_SCRAPE_CHARS + 500) + "\n❯ ";
         FakeHerdr herdr = new FakeHerdr().readText(longBlock);
         Rendezvous rendezvous = new Rendezvous();
-        CompletionResolver resolver = new CompletionResolver(new AgentControl(herdr), rendezvous);
+        CompletionResolver resolver = new CompletionResolver(new AgentControl(herdr), rendezvous, ExhaustedPatternLookup.none());
 
         var waiter = rendezvous.open("term_a");        // a send is blocked on this turn
-        resolver.captureBaseline("term_a");            // baseline is the clipped >cap block
+        resolver.captureBaseline("term_a", new TurnToken("term_a", waiter)); // baseline is the clipped >cap block
         var turn = resolver.inFlight("term_a");
         assertEquals(CompletionResolver.MAX_SCRAPE_CHARS, turn.baseline().length(),
                 "the delivery baseline is clipped to the same cap resolve() applies to the tail");
@@ -234,7 +239,7 @@ class CompletionResolverTest {
         // No delivery baseline (e.g. the pre-turn read failed) ⇒ never suppress; the completion resolves.
         FakeHerdr herdr = new FakeHerdr().readText("⏺ hello\n❯ ");
         Rendezvous rendezvous = new Rendezvous();
-        CompletionResolver resolver = new CompletionResolver(new AgentControl(herdr), rendezvous);
+        CompletionResolver resolver = new CompletionResolver(new AgentControl(herdr), rendezvous, ExhaustedPatternLookup.none());
 
         var waiter = rendezvous.open("term_a");
         resolver.resolve("term_a", new CompletionResolver.InFlight(waiter, null));
@@ -252,7 +257,7 @@ class CompletionResolverTest {
         // byte-identical guard would wrongly match the empty tail and suppress.
         FakeHerdr herdr = new FakeHerdr().healthy(false); // agent.read throws HerdrException
         Rendezvous rendezvous = new Rendezvous();
-        CompletionResolver resolver = new CompletionResolver(new AgentControl(herdr), rendezvous);
+        CompletionResolver resolver = new CompletionResolver(new AgentControl(herdr), rendezvous, ExhaustedPatternLookup.none());
 
         var waiter = rendezvous.open("term_a");
         var turn = new CompletionResolver.InFlight(waiter, ""); // empty pane baselined at delivery
@@ -272,7 +277,7 @@ class CompletionResolverTest {
         // fail must not overwrite that value, and must not even scrape the worker — nobody needs it.
         FakeHerdr herdr = new FakeHerdr().readText("an error screen");
         Rendezvous rendezvous = new Rendezvous();
-        CompletionResolver resolver = new CompletionResolver(new AgentControl(herdr), rendezvous);
+        CompletionResolver resolver = new CompletionResolver(new AgentControl(herdr), rendezvous, ExhaustedPatternLookup.none());
 
         var waiter = rendezvous.open("term_a");
         var turn = new CompletionResolver.InFlight(waiter, null);
@@ -293,7 +298,7 @@ class CompletionResolverTest {
         // fail falls back to the waiter currently registered on the Rendezvous and fails it.
         FakeHerdr herdr = new FakeHerdr().readText("stuck on an error screen");
         Rendezvous rendezvous = new Rendezvous();
-        CompletionResolver resolver = new CompletionResolver(new AgentControl(herdr), rendezvous);
+        CompletionResolver resolver = new CompletionResolver(new AgentControl(herdr), rendezvous, ExhaustedPatternLookup.none());
 
         var waiter = rendezvous.open("term_a"); // send registered, but no captureBaseline ever ran
         resolver.fail("term_a", null);          // no in-flight turn → fall back to the registered waiter
@@ -319,7 +324,7 @@ class CompletionResolverTest {
         try {
             FakeHerdr herdr = new FakeHerdr().readText("stuck on an error screen");
             Rendezvous rendezvous = new Rendezvous();
-            CompletionResolver resolver = new CompletionResolver(new AgentControl(herdr), rendezvous);
+            CompletionResolver resolver = new CompletionResolver(new AgentControl(herdr), rendezvous, ExhaustedPatternLookup.none());
             var waiter = rendezvous.open("term_a");
 
             resolver.fail("term_a", null);
@@ -347,7 +352,7 @@ class CompletionResolverTest {
         // scrape to turn N+1; targeting turn N's captured waiter makes the late completion a no-op.
         FakeHerdr herdr = new FakeHerdr().readText("⏺ turn N answer\n❯ ");
         Rendezvous rendezvous = new Rendezvous();
-        CompletionResolver resolver = new CompletionResolver(new AgentControl(herdr), rendezvous);
+        CompletionResolver resolver = new CompletionResolver(new AgentControl(herdr), rendezvous, ExhaustedPatternLookup.none());
 
         var waiterN = rendezvous.open("term_a"); // turn N's send
         // The turn as the injector captured it at delivery (waiter + pre-turn baseline).
@@ -368,4 +373,88 @@ class CompletionResolverTest {
                 "turn N stays resolved by its own reply");
         assertTrue(rendezvous.isWaiting("term_a"), "turn N+1 is still awaiting its own resolution");
     }
+
+    // --- CB-578 stage A: backend-exhausted classification ---------------------------------
+
+    @Test
+    void classifiesAMatchingScrapeAsBackendExhaustedInsteadOfACompletedReply() {
+        String block = "⏺ Working on it...\nThe usage limit has been reached. Try again later.\n❯ ";
+        FakeHerdr herdr = new FakeHerdr().readText(block);
+        Rendezvous rendezvous = new Rendezvous();
+        ExhaustedPatternLookup patterns = target -> Pattern.compile("usage limit has been reached");
+        CompletionResolver resolver = new CompletionResolver(new AgentControl(herdr), rendezvous, patterns);
+
+        var waiter = rendezvous.open("term_a");
+        resolver.resolve("term_a", new CompletionResolver.InFlight(waiter, null));
+
+        assertTrue(waiter.isDone(), "a matching scrape still resolves the blocked send");
+        assertEquals(Rendezvous.Kind.BACKEND_EXHAUSTED, waiter.getNow(null).kind(),
+                "not reported as a completed reply — the classification is distinct");
+    }
+
+    @Test
+    void theExhaustedReasonCarriesTheMatchedLine() {
+        String block = "⏺ Working on it...\nThe usage limit has been reached. Try again later.\n❯ ";
+        FakeHerdr herdr = new FakeHerdr().readText(block);
+        Rendezvous rendezvous = new Rendezvous();
+        ExhaustedPatternLookup patterns = target -> Pattern.compile("usage limit has been reached");
+        CompletionResolver resolver = new CompletionResolver(new AgentControl(herdr), rendezvous, patterns);
+
+        var waiter = rendezvous.open("term_a");
+        resolver.resolve("term_a", new CompletionResolver.InFlight(waiter, null));
+
+        assertEquals("backend exhausted (usage limit): The usage limit has been reached. Try again later.",
+                waiter.getNow(null).text(), "the reason names the real cause and carries the matched line");
+    }
+
+    @Test
+    void aNonMatchingScrapeResolvesAsAnOrdinaryCompletion() {
+        FakeHerdr herdr = new FakeHerdr().readText("⏺ complete report\n❯ ");
+        Rendezvous rendezvous = new Rendezvous();
+        ExhaustedPatternLookup patterns = target -> Pattern.compile("usage limit has been reached");
+        CompletionResolver resolver = new CompletionResolver(new AgentControl(herdr), rendezvous, patterns);
+
+        var waiter = rendezvous.open("term_a");
+        resolver.resolve("term_a", new CompletionResolver.InFlight(waiter, null));
+
+        assertEquals(Rendezvous.Kind.COMPLETION, waiter.getNow(null).kind(),
+                "a scrape that does not match the pattern is an ordinary completion");
+        assertEquals("complete report", waiter.getNow(null).text());
+    }
+
+    @Test
+    void aProfileWithNoConfiguredPatternKeepsTodaysCompletionFallbackUnchanged() {
+        // Even a scrape that WOULD have matched some other profile's pattern must resolve as a
+        // plain completion when this target's own profile has none configured (CB-578 criterion 4).
+        String block = "⏺ The usage limit has been reached.\n❯ ";
+        FakeHerdr herdr = new FakeHerdr().readText(block);
+        Rendezvous rendezvous = new Rendezvous();
+        CompletionResolver resolver =
+                new CompletionResolver(new AgentControl(herdr), rendezvous, ExhaustedPatternLookup.none());
+
+        var waiter = rendezvous.open("term_a");
+        resolver.resolve("term_a", new CompletionResolver.InFlight(waiter, null));
+
+        assertEquals(Rendezvous.Kind.COMPLETION, waiter.getNow(null).kind(),
+                "no pattern configured for this target's profile ⇒ unchanged completion-fallback behaviour");
+        assertEquals("The usage limit has been reached.", waiter.getNow(null).text());
+    }
+
+    @Test
+    void coverageIsOffWhenNoProfileHasAPatternConfigured() {
+        assertEquals("off (no profile has an exhaustedPattern configured; profiles: [terra])",
+                CompletionResolver.coverage(Set.of("terra"), Set.of()));
+    }
+
+    @Test
+    void coverageIsFullWhenEveryProfileHasAPatternConfigured() {
+        assertEquals("full (all profiles configured: [gx10, terra])",
+                CompletionResolver.coverage(Set.of("terra", "gx10"), Set.of("terra", "gx10")));
+    }
+
+    @Test
+    void coverageIsPartialAndNamesWhichProfilesAreConfigured() {
+        assertEquals("partial (configured: [terra]; not configured: [gx10])",
+                CompletionResolver.coverage(Set.of("terra", "gx10"), Set.of("terra")));
+    }
 }

bridged/src/test/java/dev/ltms/bridged/inject/InjectorTest.java

@@ -8,6 +8,7 @@ import dev.ltms.bridged.herdr.AgentControl;
 import dev.ltms.bridged.herdr.AgentStatus;
 import dev.ltms.bridged.herdr.FakeHerdr;
 import dev.ltms.bridged.herdr.HerdrException;
+import dev.ltms.bridged.msg.TestTurnTokens;
 import org.junit.jupiter.api.Test;
 import org.slf4j.LoggerFactory;
 
@@ -47,7 +48,7 @@ class InjectorTest {
 
     @Test
     void deliversWhenIdle() {
-        CompletableFuture<Void> f = injector.enqueue(T, "hello");
+        CompletableFuture<Void> f = injector.enqueue(T, "hello", TestTurnTokens.inert(T));
         assertFalse(f.isDone(), "not delivered until an injectable status arrives");
         injector.onStatus(T, AgentStatus.IDLE);
         assertTrue(f.isDone());
@@ -59,7 +60,7 @@ class InjectorTest {
         // CB-113: idle alone is not enough — hold until the worker's MCP is connected (ready).
         java.util.Set<String> ready = new java.util.HashSet<>();
         Injector inj = new Injector(new AgentControl(herdr), TurnListener.NOOP, ready::contains);
-        inj.enqueue(T, "task");
+        inj.enqueue(T, "task", TestTurnTokens.inert(T));
 
         inj.onStatus(T, AgentStatus.IDLE); // idle but not yet available → held out of the boot window
         assertEquals(List.of(), sent(), "must not deliver into a not-yet-available worker");
@@ -79,7 +80,7 @@ class InjectorTest {
     void resubmitsEnterWhenADeliveredMessageIsNotPickedUp() {
         // CB-113: the Enter at delivery can race the paste; while the worker stays idle (not picked
         // up), the injector re-nudges Enter so the pending paste submits.
-        injector.enqueue(T, "task");
+        injector.enqueue(T, "task", TestTurnTokens.inert(T));
         injector.onStatus(T, AgentStatus.IDLE); // deliver: paste + one Enter
         long afterDeliver = enterKeystrokes();
 
@@ -95,7 +96,7 @@ class InjectorTest {
 
     @Test
     void holdsWhileWorkingThenDeliversOnIdle() {
-        injector.enqueue(T, "later");
+        injector.enqueue(T, "later", TestTurnTokens.inert(T));
         injector.onStatus(T, AgentStatus.WORKING);
         assertEquals(List.of(), sent(), "must not inject mid-turn");
         injector.onStatus(T, AgentStatus.IDLE);
@@ -104,7 +105,7 @@ class InjectorTest {
 
     @Test
     void blockedIsInjectableButUnknownIsNot() {
-        injector.enqueue(T, "answer");
+        injector.enqueue(T, "answer", TestTurnTokens.inert(T));
         injector.onStatus(T, AgentStatus.UNKNOWN);
         assertEquals(List.of(), sent(), "unknown status is not safe to inject");
         injector.onStatus(T, AgentStatus.BLOCKED);
@@ -113,8 +114,8 @@ class InjectorTest {
 
     @Test
     void twoRapidDeliveriesNeverInterleave() {
-        injector.enqueue(T, "m1");
-        injector.enqueue(T, "m2");
+        injector.enqueue(T, "m1", TestTurnTokens.inert(T));
+        injector.enqueue(T, "m2", TestTurnTokens.inert(T));
 
         // First idle window delivers only m1, even if idle is observed twice before pickup.
         injector.onStatus(T, AgentStatus.IDLE);
@@ -129,8 +130,8 @@ class InjectorTest {
 
     @Test
     void transientUnknownDoesNotReleaseThePickupLatch() {
-        injector.enqueue(T, "m1");
-        injector.enqueue(T, "m2");
+        injector.enqueue(T, "m1", TestTurnTokens.inert(T));
+        injector.enqueue(T, "m2", TestTurnTokens.inert(T));
         injector.onStatus(T, AgentStatus.IDLE); // m1 sent, awaiting pickup
         assertEquals(List.of("m1"), sent());
 
@@ -145,8 +146,8 @@ class InjectorTest {
 
     @Test
     void missedPickupEdgeIsReleasedByGraceSoTheQueueNeverWedges() {
-        injector.enqueue(T, "m1");
-        injector.enqueue(T, "m2");
+        injector.enqueue(T, "m1", TestTurnTokens.inert(T));
+        injector.enqueue(T, "m2", TestTurnTokens.inert(T));
         injector.onStatus(T, AgentStatus.IDLE); // m1 sent
         assertEquals(List.of("m1"), sent());
 
@@ -158,9 +159,9 @@ class InjectorTest {
 
     @Test
     void fifoOrderAcrossManyTurns() {
-        injector.enqueue(T, "a");
-        injector.enqueue(T, "b");
-        injector.enqueue(T, "c");
+        injector.enqueue(T, "a", TestTurnTokens.inert(T));
+        injector.enqueue(T, "b", TestTurnTokens.inert(T));
+        injector.enqueue(T, "c", TestTurnTokens.inert(T));
         for (int i = 0; i < 3; i++) {
             injector.onStatus(T, AgentStatus.IDLE);   // deliver one
             injector.onStatus(T, AgentStatus.WORKING); // pickup
@@ -172,7 +173,7 @@ class InjectorTest {
     @Test
     void activeWhileQueuedOrInFlightThenQuietAfterTurnCompletes() {
         assertTrue(injector.activeTargets().isEmpty());
-        injector.enqueue(T, "x");
+        injector.enqueue(T, "x", TestTurnTokens.inert(T));
         assertEquals(Set.of(T), injector.activeTargets(), "active while a message is queued");
 
         injector.onStatus(T, AgentStatus.IDLE); // delivers; awaiting pickup
@@ -191,7 +192,7 @@ class InjectorTest {
     void firesTurnCompleteOnAConfirmedWorkingThenIdle() {
         List<String> completed = new ArrayList<>();
         Injector inj = new Injector(new AgentControl(herdr), completed::add);
-        inj.enqueue(T, "task");
+        inj.enqueue(T, "task", TestTurnTokens.inert(T));
 
         inj.onStatus(T, AgentStatus.IDLE);    // deliver
         inj.onStatus(T, AgentStatus.WORKING); // pickup + turn running
@@ -226,8 +227,8 @@ class InjectorTest {
         }
         ResetListener listener = new ResetListener();
         Injector inj = new Injector(agents, listener);
-        inj.enqueue(T, "first");
-        inj.enqueue(T, "second");
+        inj.enqueue(T, "first", TestTurnTokens.inert(T));
+        inj.enqueue(T, "second", TestTurnTokens.inert(T));
 
         inj.onStatus(T, AgentStatus.IDLE);    // first delegation
         inj.onStatus(T, AgentStatus.WORKING);
@@ -246,7 +247,7 @@ class InjectorTest {
     void doesNotSynthesizeCompletionFromAnUnconfirmedTurn() {
         List<String> completed = new ArrayList<>();
         Injector inj = new Injector(new AgentControl(herdr), completed::add);
-        inj.enqueue(T, "task");
+        inj.enqueue(T, "task", TestTurnTokens.inert(T));
 
         // Deliver, then only ever idle — a `working` sample is never seen. The pickup grace unwedges
         // the queue but must NOT invent a completion: without a sampled turn there is no trustworthy
@@ -259,6 +260,7 @@ class InjectorTest {
     private static final class Captor implements TurnListener {
         final List<String> completed = new ArrayList<>();
         final List<String> failed = new ArrayList<>();
+        final List<String> failureReasons = new ArrayList<>();
 
         @Override
         public void onTurnComplete(String target) {
@@ -269,6 +271,12 @@ class InjectorTest {
         public void onTurnFailed(String target) {
             failed.add(target);
         }
+
+        @Override
+        public void onTurnFailed(String target, String reason) {
+            failed.add(target);
+            failureReasons.add(reason);
+        }
     }
 
     // ~30s of unknown at the 250ms prod poll interval; enough onStatus samples to trip the stall.
@@ -278,7 +286,7 @@ class InjectorTest {
     void failsAnOutstandingDelegationWhoseWorkerWedgesInUnknown() {
         Captor cap = new Captor();
         Injector inj = new Injector(new AgentControl(herdr), cap);
-        inj.enqueue(T, "task");
+        inj.enqueue(T, "task", TestTurnTokens.inert(T));
 
         inj.onStatus(T, AgentStatus.IDLE);    // deliver
         inj.onStatus(T, AgentStatus.WORKING); // worker starts the turn
@@ -293,7 +301,7 @@ class InjectorTest {
     void aTransientUnknownGlitchNeitherFailsNorBlocksCompletion() {
         Captor cap = new Captor();
         Injector inj = new Injector(new AgentControl(herdr), cap);
-        inj.enqueue(T, "task");
+        inj.enqueue(T, "task", TestTurnTokens.inert(T));
 
         inj.onStatus(T, AgentStatus.IDLE);    // deliver
         inj.onStatus(T, AgentStatus.WORKING); // confirmed turn
@@ -308,7 +316,7 @@ class InjectorTest {
     void sendFailureDropsMessageAndFailsItsFuture() {
         FakeHerdr failing = new FakeHerdr().agentSendFailsWith("send_failed");
         Injector inj = new Injector(new AgentControl(failing));
-        CompletableFuture<Void> f = inj.enqueue(T, "boom");
+        CompletableFuture<Void> f = inj.enqueue(T, "boom", TestTurnTokens.inert(T));
 
         inj.onStatus(T, AgentStatus.IDLE);
         assertTrue(f.isCompletedExceptionally());
@@ -317,18 +325,35 @@ class InjectorTest {
 
     @Test
     void dropFailsPendingWaiters() {
-        CompletableFuture<Void> f = injector.enqueue(T, "orphan");
+        CompletableFuture<Void> f = injector.enqueue(T, "orphan", TestTurnTokens.inert(T));
         injector.drop(T, new HerdrException("worker gone", "pane_not_found", null));
         assertTrue(f.isCompletedExceptionally(), "queued waiters unblock when the worker vanishes");
     }
 
+    @Test
+    void dropPassesTheRealCauseForQueuedAndDeliveredWork() {
+        Captor cap = new Captor();
+        Injector inj = new Injector(new AgentControl(herdr), cap);
+        CompletableFuture<Void> delivered = inj.enqueue(T, "delivered", TestTurnTokens.inert(T));
+        CompletableFuture<Void> queued = inj.enqueue(T, "queued", TestTurnTokens.inert(T));
+
+        inj.onStatus(T, AgentStatus.IDLE);    // deliver the first message
+        inj.onStatus(T, AgentStatus.WORKING); // its turn is now in flight; one remains queued
+        inj.drop(T, new HerdrException("agent target sol not found", "agent_not_found", null));
+
+        assertEquals(List.of(T), cap.failed, "drop signals one turn failure for both affected states");
+        assertEquals(List.of("agent target sol not found"), cap.failureReasons);
+        assertTrue(delivered.isDone(), "the delivered future has already completed");
+        assertTrue(queued.isCompletedExceptionally(), "the queued future fails with the drop cause");
+    }
+
     @Test
     void dropFailsTheTurnOfADeliveredMessageWhenTheWorkerVanishes() {
         // CB-110: the message was delivered (no longer queued), so failing queued waiters alone would
         // leave its send hanging. A vanished worker must fail that in-flight turn too.
         Captor cap = new Captor();
         Injector inj = new Injector(new AgentControl(herdr), cap);
-        inj.enqueue(T, "task");
+        inj.enqueue(T, "task", TestTurnTokens.inert(T));
         inj.onStatus(T, AgentStatus.IDLE);    // deliver
         inj.onStatus(T, AgentStatus.WORKING); // turn running
 
@@ -343,7 +368,7 @@ class InjectorTest {
         // by an off-sub worker's review of CB-110, delegated through the bridge.)
         Captor cap = new Captor();
         Injector inj = new Injector(new AgentControl(herdr), cap);
-        inj.enqueue(T, "task");
+        inj.enqueue(T, "task", TestTurnTokens.inert(T));
         inj.onStatus(T, AgentStatus.IDLE); // deliver; pickup never confirmed
 
         inj.drop(T, new HerdrException("worker gone", "pane_not_found", null));
@@ -362,7 +387,7 @@ class InjectorTest {
         Captor cap = new Captor();
         List<String> forgotten = new ArrayList<>();
         Injector inj = new Injector(new AgentControl(herdr), cap, _ -> false, forgotten::add);
-        CompletableFuture<Void> f = inj.enqueue(T, "task");
+        CompletableFuture<Void> f = inj.enqueue(T, "task", TestTurnTokens.inert(T));
 
         for (int i = 0; i < READINESS_SAMPLES; i++) inj.onStatus(T, AgentStatus.IDLE);
 
@@ -390,7 +415,7 @@ class InjectorTest {
         try {
             Injector inj = new Injector(new AgentControl(herdr), TurnListener.NOOP, _ -> false, _ -> {
             });
-            inj.enqueue(T, "task");
+            inj.enqueue(T, "task", TestTurnTokens.inert(T));
 
             for (int i = 0; i < READINESS_SAMPLES; i++) inj.onStatus(T, AgentStatus.IDLE);
 
@@ -415,7 +440,7 @@ class InjectorTest {
         Set<String> ready = new java.util.HashSet<>();
         Injector inj = new Injector(new AgentControl(herdr), TurnListener.NOOP, ready::contains, _ -> {
         });
-        inj.enqueue(T, "task");
+        inj.enqueue(T, "task", TestTurnTokens.inert(T));
 
         for (int i = 0; i < 100; i++) inj.onStatus(T, AgentStatus.IDLE); // still booting, well under grace
         assertEquals(List.of(), sent());
@@ -431,7 +456,7 @@ class InjectorTest {
         // linger past the worker's life (MemberPresence.forget had no caller before this).
         List<String> forgotten = new ArrayList<>();
         Injector inj = new Injector(new AgentControl(herdr), TurnListener.NOOP, _ -> true, forgotten::add);
-        inj.enqueue(T, "orphan");
+        inj.enqueue(T, "orphan", TestTurnTokens.inert(T));
         inj.drop(T, new HerdrException("worker gone", "pane_not_found", null));
         assertEquals(List.of(T), forgotten, "drop clears the gone worker's presence");
     }
@@ -452,7 +477,7 @@ class InjectorTest {
         try {
             Injector inj = new Injector(new AgentControl(herdr), TurnListener.NOOP, _ -> true, _ -> {
             });
-            inj.enqueue(T, "orphan");
+            inj.enqueue(T, "orphan", TestTurnTokens.inert(T));
             inj.drop(T, new HerdrException("worker gone", "pane_not_found", null));
 
             String warn = appender.list.stream()
@@ -476,7 +501,7 @@ class InjectorTest {
         StatusPoller poller = new StatusPoller(new AgentControl(idle), inj, 10);
         poller.start();
         try {
-            CompletableFuture<Void> delivered = inj.enqueue(T, "via-poller");
+            CompletableFuture<Void> delivered = inj.enqueue(T, "via-poller", TestTurnTokens.inert(T));
             delivered.get(2, TimeUnit.SECONDS); // completes when the poller drives the send
         } finally {
             poller.stop();
@@ -495,7 +520,7 @@ class InjectorTest {
     void deliveredFutureCarriesSendFailure() {
         FakeHerdr failing = new FakeHerdr().agentSendFailsWith("send_failed");
         Injector inj = new Injector(new AgentControl(failing));
-        CompletableFuture<Void> f = inj.enqueue(T, "boom");
+        CompletableFuture<Void> f = inj.enqueue(T, "boom", TestTurnTokens.inert(T));
         inj.onStatus(T, AgentStatus.IDLE);
         ExecutionException ex = assertThrows(ExecutionException.class, f::get);
         assertInstanceOf(HerdrException.class, ex.getCause());

bridged/src/test/java/dev/ltms/bridged/lead/LeadLauncherTest.java

@@ -28,7 +28,7 @@ class LeadLauncherTest {
                 List.of("ccs", "ltms"), "tab", "bridged-workers", null,
                 "http://127.0.0.1:8765/mcp", null, null,
                 null, null, null,
-                Map.of("CLAUDE_CODE_AUTO_COMPACT_WINDOW", "300000"), null, null, true);
+                Map.of("CLAUDE_CODE_AUTO_COMPACT_WINDOW", "300000"), null, null, true, null);
     }
 
     private static BridgedConfig configWith(BridgedConfig.Leader lead) {
@@ -41,8 +41,8 @@ class LeadLauncherTest {
                 null, null, fleet, null, "fixed", null).withDefaults();
     }
 
-    private static BridgedConfig.Leader lead(String profile, String terminal, int instances) {
-        return new BridgedConfig.Leader(profile, terminal, instances, "lead:", 10, null, null,
+    private static BridgedConfig.Leader lead(String profile, String tab, int instances) {
+        return new BridgedConfig.Leader(profile, tab, instances, "lead:", 10, null, null,
                 "leads", "/repo");
     }
 
@@ -70,16 +70,16 @@ class LeadLauncherTest {
     void startsTheDeclaredLeadWhenNoneIsRunning() {
         FakeHerdr herdr = new FakeHerdr();
 
-        assertEquals(1, launcher(herdr, configWith(lead("opus", null, 1))).ensureLeads());
+        assertEquals(1, launcher(herdr, configWith(lead("opus", "lead: opus", 1))).ensureLeads());
         assertTrue(herdr.called("agent.start"), "a lead must actually be started");
         assertEquals("lead-opus", startedName(herdr));
     }
 
-    /** The tab is labelled so the scanner finds the lead on the next resolve. */
+    /** The tab is labelled with the configured `tab:` so the scanner finds the lead on the next resolve. */
     @Test
-    void labelsTheTabWithThePrefixTheScannerReadsBack() {
+    void labelsTheTabWithTheConfiguredTabValue() {
         FakeHerdr herdr = new FakeHerdr();
-        launcher(herdr, configWith(lead("opus", null, 1))).ensureLeads();
+        launcher(herdr, configWith(lead("opus", "lead: opus", 1))).ensureLeads();
 
         assertEquals("lead: opus",
                 ((Map<?, ?>) herdr.lastCall("tab.rename").params()).get("label"));
@@ -90,7 +90,7 @@ class LeadLauncherTest {
     void startsAsManyInstancesAsAreDeclared() {
         FakeHerdr herdr = new FakeHerdr();
 
-        assertEquals(2, launcher(herdr, configWith(lead("opus", null, 2))).ensureLeads());
+        assertEquals(2, launcher(herdr, configWith(lead("opus", "lead: opus", 2))).ensureLeads());
         assertEquals(2, herdr.calls.stream().filter(c -> c.method().equals("agent.start")).count());
     }
 
@@ -104,7 +104,7 @@ class LeadLauncherTest {
                 .withTab("wL", "wL:t1", "lead: opus")
                 .withAgent("lead-opus", "term_lead", "wL:p1", "wL:t1");
 
-        assertEquals(0, launcher(herdr, configWith(lead("opus", null, 1))).ensureLeads());
+        assertEquals(0, launcher(herdr, configWith(lead("opus", "lead: opus", 1))).ensureLeads());
         assertFalse(herdr.called("agent.start"), "the live lead must not be duplicated");
     }
 
@@ -118,20 +118,23 @@ class LeadLauncherTest {
                 .withWorkspace("wL", "leads")
                 .withTab("wL", "wL:t1", "lead: opus"); // label only — nothing running in it
 
-        assertEquals(1, launcher(herdr, configWith(lead("opus", null, 1))).ensureLeads(),
+        assertEquals(1, launcher(herdr, configWith(lead("opus", "lead: opus", 1))).ensureLeads(),
                 "a stale label is not a lead; the lead must be relaunched");
     }
 
     /**
-     * A lead the operator opened by hand and pinned with `terminal:` is live even though its tab
-     * carries no matching label. Counting labels alone would relaunch it on every boot.
+     * A lead the operator opened by hand is live once its tab carries the configured `tab:` label —
+     * CB-579 retired the `terminal:` pin, so a hand-opened lead is found the same way an
+     * auto-launched one is, by its tab, not by a terminal id nobody wrote down in advance.
      */
     @Test
-    void aPinnedTerminalWithARunningAgentCountsAsLive() {
+    void aHandOpenedLeadWithTheConfiguredTabLabelCountsAsLive() {
         FakeHerdr herdr = new FakeHerdr()
-                .withAgent("hand-opened", "term_pinned", "wX:p1", "wX:t1");
+                .withWorkspace("wX", "main")
+                .withTab("wX", "wX:t1", "lead: opus")
+                .withAgent("hand-opened", "term_hand", "wX:p1", "wX:t1");
 
-        assertEquals(0, launcher(herdr, configWith(lead("opus", "term_pinned", 1))).ensureLeads());
+        assertEquals(0, launcher(herdr, configWith(lead("opus", "lead: opus", 1))).ensureLeads());
         assertFalse(herdr.called("agent.start"));
     }
 
@@ -143,7 +146,7 @@ class LeadLauncherTest {
                 .withTab("wM", "wM:t1", "lead: opus")              // a member tab that looks like a lead
                 .withAgent("claude-opus-x", "term_m", "wM:p1", "wM:t1");
 
-        assertEquals(1, launcher(herdr, configWith(lead("opus", null, 1))).ensureLeads(),
+        assertEquals(1, launcher(herdr, configWith(lead("opus", "lead: opus", 1))).ensureLeads(),
                 "a member in a lead-labelled tab is not a lead, so the real lead is still missing");
     }
 
@@ -152,7 +155,7 @@ class LeadLauncherTest {
     void anUncountableHerdrStartsNothing() {
         FakeHerdr herdr = new FakeHerdr().healthy(false);
 
-        assertEquals(0, launcher(herdr, configWith(lead("opus", null, 1))).ensureLeads());
+        assertEquals(0, launcher(herdr, configWith(lead("opus", "lead: opus", 1))).ensureLeads());
         assertFalse(herdr.called("agent.start"));
     }
 
@@ -166,7 +169,7 @@ class LeadLauncherTest {
     @Test
     void theLeadNeverReceivesTheWorkerReplyCharter() {
         FakeHerdr herdr = new FakeHerdr();
-        launcher(herdr, configWith(lead("opus", null, 1))).ensureLeads();
+        launcher(herdr, configWith(lead("opus", "lead: opus", 1))).ensureLeads();
 
         List<String> args = startedArgs(herdr);
         assertFalse(args.contains("--append-system-prompt"),
@@ -178,7 +181,7 @@ class LeadLauncherTest {
     @Test
     void theLeadMountsTheBridgeMcpAndPinsItsModel() {
         FakeHerdr herdr = new FakeHerdr();
-        launcher(herdr, configWith(lead("opus", null, 1))).ensureLeads();
+        launcher(herdr, configWith(lead("opus", "lead: opus", 1))).ensureLeads();
 
         List<String> args = startedArgs(herdr);
         assertTrue(args.contains("--mcp-config"));
@@ -192,7 +195,7 @@ class LeadLauncherTest {
     @Test
     void theLeadEnvCarriesNoAnthropicBinding() {
         FakeHerdr herdr = new FakeHerdr();
-        launcher(herdr, configWith(lead("opus", null, 1))).ensureLeads();
+        launcher(herdr, configWith(lead("opus", "lead: opus", 1))).ensureLeads();
 
         Map<String, String> env = tabEnv(herdr);
         assertNull(env.get("ANTHROPIC_BASE_URL"));
@@ -205,7 +208,7 @@ class LeadLauncherTest {
     @Test
     void theLeadTabIsCreatedOutsideEveryMemberWorkspace() {
         FakeHerdr herdr = new FakeHerdr();
-        launcher(herdr, configWith(lead("opus", null, 1))).ensureLeads();
+        launcher(herdr, configWith(lead("opus", "lead: opus", 1))).ensureLeads();
 
         String label = (String) ((Map<?, ?>) herdr.lastCall("workspace.create").params()).get("label");
         assertEquals("leads", label);
@@ -214,12 +217,12 @@ class LeadLauncherTest {
 
     // ── recognise-only and misconfiguration ───────────────────────────────────────────────────
 
-    /** A lead with a pin but no profile is recognise-only by design — not an error, not a launch. */
+    /** A lead with a tab but no profile is recognise-only by design — not an error, not a launch. */
     @Test
     void aLeadThatNamesNoProfileIsRecognisedButNeverLaunched() {
         FakeHerdr herdr = new FakeHerdr();
 
-        assertEquals(0, launcher(herdr, configWith(lead(null, "term_dead", 1))).ensureLeads());
+        assertEquals(0, launcher(herdr, configWith(lead(null, "lead: dead", 1))).ensureLeads());
         assertFalse(herdr.called("agent.start"));
     }
 
@@ -228,7 +231,7 @@ class LeadLauncherTest {
     void zeroInstancesLaunchesNothing() {
         FakeHerdr herdr = new FakeHerdr();
 
-        assertEquals(0, launcher(herdr, configWith(lead("opus", null, 0))).ensureLeads());
+        assertEquals(0, launcher(herdr, configWith(lead("opus", "lead: opus", 0))).ensureLeads());
         assertFalse(herdr.called("agent.start"));
     }
 
@@ -237,7 +240,7 @@ class LeadLauncherTest {
     void anUnknownProfileIsSkippedRatherThanThrown() {
         FakeHerdr herdr = new FakeHerdr();
 
-        assertEquals(0, launcher(herdr, configWith(lead("nope", null, 1))).ensureLeads());
+        assertEquals(0, launcher(herdr, configWith(lead("nope", "lead: opus", 1))).ensureLeads());
         assertFalse(herdr.called("agent.start"));
     }
 

bridged/src/test/java/dev/ltms/bridged/logging/McpCancelledNotificationFilterTest.java

@@ -0,0 +1,37 @@
+package dev.ltms.bridged.logging;
+
+import ch.qos.logback.classic.Level;
+import ch.qos.logback.classic.Logger;
+import ch.qos.logback.classic.spi.ILoggingEvent;
+import ch.qos.logback.core.read.ListAppender;
+import io.modelcontextprotocol.spec.McpSchema;
+import org.junit.jupiter.api.Test;
+import org.slf4j.LoggerFactory;
+
+import java.util.Map;
+
+import static org.junit.jupiter.api.Assertions.assertEquals;
+import static org.junit.jupiter.api.Assertions.assertTrue;
+
+class McpCancelledNotificationFilterTest {
+
+    @Test
+    void suppressesOnlyTheCancelledNotificationWarning() {
+        Logger logger = (Logger) LoggerFactory.getLogger(McpCancelledNotificationFilter.LOGGER);
+        ListAppender<ILoggingEvent> appender = new ListAppender<>();
+        appender.start();
+        logger.addAppender(appender);
+        try {
+            logger.warn(McpCancelledNotificationFilter.UNHANDLED_NOTIFICATION,
+                    new McpSchema.JSONRPCNotification("notifications/cancelled", Map.of("requestId", 7)));
+            logger.warn(McpCancelledNotificationFilter.UNHANDLED_NOTIFICATION,
+                    new McpSchema.JSONRPCNotification("notifications/progress", Map.of("progress", 1)));
+
+            assertEquals(1, appender.list.size());
+            assertEquals(Level.WARN, appender.list.getFirst().getLevel());
+            assertTrue(appender.list.getFirst().getFormattedMessage().contains("notifications/progress"));
+        } finally {
+            logger.detachAppender(appender);
+        }
+    }
+}

bridged/src/test/java/dev/ltms/bridged/mcp/BridgeMcpAuthzTest.java

@@ -72,7 +72,7 @@ class BridgeMcpAuthzTest {
                 new PrimaryRegistry(null),
                 enforce ? CallerResolver.withLeadsAndMembers(identity, false, null,
                 Map::of, new MemberRegistry(null)) : null,
-                metrics);
+                metrics, BridgeMcp.CapacitySource.none(), new BridgeMcp.HealthCoverageSource(() -> "off"));
         return mcp;
     }
 

bridged/src/test/java/dev/ltms/bridged/mcp/BridgeMcpTest.java

@@ -53,6 +53,18 @@ class BridgeMcpTest {
         return ((McpSchema.TextContent) r.content().getFirst()).text();
     }
 
+    private void assertSendRoundTrips(String target, Set<String> profiles) throws Exception {
+        CompletableFuture<McpSchema.CallToolResult> send = CompletableFuture.supplyAsync(
+                () -> BridgeMcp.send(messages, target, "hi", 4000L, null, profiles));
+        long deadline = System.currentTimeMillis() + 3000;
+        while (!rendezvous.isWaiting(target) && System.currentTimeMillis() < deadline) {
+            Thread.sleep(5);
+        }
+        assertTrue(rendezvous.isWaiting(target), "send should be accepted for " + target);
+        BridgeMcp.reply(messages, target, "received");
+        assertEquals("received", textOf(send.get(6, TimeUnit.SECONDS)));
+    }
+
     private static ClaudeCodeLauncher workerService(FakeHerdr h, String baseUrl, Set<String> allow) {
         BridgedConfig.Profile cfg = new BridgedConfig.Profile(
                 "ltms-local", baseUrl, "coder", null, "BRIDGED_WORKER_TOKEN", null,
@@ -69,7 +81,7 @@ class BridgeMcpTest {
     void sendThenReplyRoundTrips() throws Exception {
         // bridge_send blocks; bridge_reply resolves it with the worker's structured answer.
         CompletableFuture<McpSchema.CallToolResult> send = CompletableFuture.supplyAsync(
-                () -> BridgeMcp.send(messages, "term_a", "review this", 4000L));
+                () -> BridgeMcp.send(messages, "term_a", "review this", 4000L, null, Set.of()));
 
         // Wait until the send has opened its waiter so the reply resolves it (CB-307: reply now
         // queues in the inbox if no waiter is open, which would break the round-trip).
@@ -91,7 +103,7 @@ class BridgeMcpTest {
     @Test
     void asyncSendReturnsATicketThenPollReportsTheReply() throws Exception {
         // wait:false parity — a ticket is issued, resolved by a reply, and surfaced by bridge_poll.
-        McpSchema.CallToolResult accepted = BridgeMcp.sendAsync(messages, "term_a", "do it");
+        McpSchema.CallToolResult accepted = BridgeMcp.sendAsync(messages, "term_a", "do it", null, Set.of());
         assertNotEquals(Boolean.TRUE, accepted.isError());
         String out = textOf(accepted);
         assertTrue(out.contains("ticket="), out);
@@ -120,6 +132,131 @@ class BridgeMcpTest {
         assertEquals("async LGTM", textOf(polled));
     }
 
+    @Test
+    void asyncSendSurfacesAnAskThenKeepsTheTicketForTheFinalReply() throws Exception {
+        McpSchema.CallToolResult accepted = BridgeMcp.sendAsync(messages, "term_a", "do it", null, Set.of());
+        String ticket = textOf(accepted).substring(textOf(accepted).indexOf("ticket=") + "ticket=".length()).trim();
+
+        long deadline = System.currentTimeMillis() + 3000;
+        while (!rendezvous.isWaiting("term_a") && System.currentTimeMillis() < deadline) {
+            Thread.sleep(5);
+        }
+        assertTrue(rendezvous.isWaiting("term_a"));
+
+        CompletableFuture<McpSchema.CallToolResult> ask = CompletableFuture.supplyAsync(
+                () -> BridgeMcp.ask(messages, "term_a", "which config?", 5000L));
+        McpSchema.CallToolResult question = BridgeMcp.poll(messages, ticket, null);
+        deadline = System.currentTimeMillis() + 3000;
+        while (!textOf(question).contains("[question") && System.currentTimeMillis() < deadline) {
+            Thread.sleep(5);
+            question = BridgeMcp.poll(messages, ticket, null);
+        }
+        assertTrue(textOf(question).contains("which config?"), textOf(question));
+        String questionText = textOf(question);
+        String afterTurnId = questionText.substring(questionText.indexOf("turnId=\"") + "turnId=\"".length());
+        String turnId = afterTurnId.substring(0, afterTurnId.indexOf('"'));
+
+        CompletableFuture<McpSchema.CallToolResult> answer = CompletableFuture.supplyAsync(
+                () -> BridgeMcp.answer(messages, turnId, "config.yaml", 5000L));
+        assertEquals("config.yaml", textOf(ask.get(6, TimeUnit.SECONDS)));
+
+        deadline = System.currentTimeMillis() + 3000;
+        while (!rendezvous.isWaiting("term_a") && System.currentTimeMillis() < deadline) {
+            Thread.sleep(5);
+        }
+        assertTrue(rendezvous.isWaiting("term_a"));
+        BridgeMcp.reply(messages, "term_a", "done");
+        assertEquals("done", textOf(answer.get(6, TimeUnit.SECONDS)));
+
+        McpSchema.CallToolResult done = BridgeMcp.poll(messages, ticket, null);
+        deadline = System.currentTimeMillis() + 3000;
+        while (!"done".equals(textOf(done)) && System.currentTimeMillis() < deadline) {
+            Thread.sleep(5);
+            done = BridgeMcp.poll(messages, ticket, null);
+        }
+        assertEquals("done", textOf(done));
+    }
+
+    @Test
+    void unansweredAsyncAskReturnsTheTicketToPending() throws Exception {
+        McpSchema.CallToolResult accepted = BridgeMcp.sendAsync(messages, "term_a", "do it", null, Set.of());
+        String ticket = textOf(accepted).substring(textOf(accepted).indexOf("ticket=") + "ticket=".length()).trim();
+
+        long deadline = System.currentTimeMillis() + 3000;
+        while (!rendezvous.isWaiting("term_a") && System.currentTimeMillis() < deadline) {
+            Thread.sleep(5);
+        }
+        assertTrue(rendezvous.isWaiting("term_a"));
+
+        McpSchema.CallToolResult ask = BridgeMcp.ask(messages, "term_a", "still there?", 50L);
+        assertTrue(textOf(ask).contains("no answer"), textOf(ask));
+        assertTrue(textOf(BridgeMcp.poll(messages, ticket, null)).startsWith("[pending"));
+
+        BridgeMcp.reply(messages, "term_a", "finished after timeout");
+        assertEquals("finished after timeout", messages.drainReplies("term_a").getFirst().content());
+    }
+
+    @Test
+    void asyncSendFailureDoesNotLeaveItsTicketPending() throws Exception {
+        String ticket = messages.sendAsync("term_a", "do it", () -> {
+            throw new IllegalStateException("accept failed");
+        });
+
+        long deadline = System.currentTimeMillis() + 3000;
+        MessageService.TaskView view = messages.poll(ticket);
+        while (view.phase() == MessageService.Phase.PENDING && System.currentTimeMillis() < deadline) {
+            Thread.sleep(5);
+            view = messages.poll(ticket);
+        }
+        assertEquals(MessageService.Phase.FAILED, view.phase());
+        assertEquals("accept failed", view.detail());
+    }
+
+    @Test
+    void anotherAsyncTicketCannotCaptureAReplyWhileTheFirstTicketIsAsking() throws Exception {
+        McpSchema.CallToolResult firstAccepted = BridgeMcp.sendAsync(messages, "term_a", "first", null, Set.of());
+        String firstTicket = textOf(firstAccepted).substring(textOf(firstAccepted).indexOf("ticket=") + "ticket=".length()).trim();
+
+        long deadline = System.currentTimeMillis() + 3000;
+        while (!rendezvous.isWaiting("term_a") && System.currentTimeMillis() < deadline) {
+            Thread.sleep(5);
+        }
+        assertTrue(rendezvous.isWaiting("term_a"));
+
+        CompletableFuture<McpSchema.CallToolResult> ask = CompletableFuture.supplyAsync(
+                () -> BridgeMcp.ask(messages, "term_a", "which config?", 5000L));
+        MessageService.TaskView first = messages.poll(firstTicket);
+        deadline = System.currentTimeMillis() + 3000;
+        while (first.phase() != MessageService.Phase.ASKING && System.currentTimeMillis() < deadline) {
+            Thread.sleep(5);
+            first = messages.poll(firstTicket);
+        }
+        assertEquals(MessageService.Phase.ASKING, first.phase());
+        String firstTurnId = first.turnId();
+
+        McpSchema.CallToolResult secondAccepted = BridgeMcp.sendAsync(messages, "term_a", "second", null, Set.of());
+        String secondTicket = textOf(secondAccepted).substring(textOf(secondAccepted).indexOf("ticket=") + "ticket=".length()).trim();
+        MessageService.TaskView second = messages.poll(secondTicket);
+        deadline = System.currentTimeMillis() + 3000;
+        while (second.phase() == MessageService.Phase.PENDING && System.currentTimeMillis() < deadline) {
+            Thread.sleep(5);
+            second = messages.poll(secondTicket);
+        }
+        assertEquals(MessageService.Phase.FAILED, second.phase());
+
+        BridgeMcp.reply(messages, "term_a", "late reply");
+        assertEquals("late reply", messages.drainReplies("term_a").getFirst().content());
+
+        CompletableFuture<McpSchema.CallToolResult> answer = CompletableFuture.supplyAsync(
+                () -> BridgeMcp.answer(messages, firstTurnId, "config.yaml", 5000L));
+        assertEquals("config.yaml", textOf(ask.get(6, TimeUnit.SECONDS)));
+        while (!rendezvous.isWaiting("term_a") && System.currentTimeMillis() < deadline) {
+            Thread.sleep(5);
+        }
+        BridgeMcp.reply(messages, "term_a", "done");
+        assertEquals("done", textOf(answer.get(6, TimeUnit.SECONDS)));
+    }
+
     @Test
     void pollUnknownTicketIsAnError() {
         McpSchema.CallToolResult res = BridgeMcp.poll(messages, "task-999", null);
@@ -129,15 +266,40 @@ class BridgeMcpTest {
 
     @Test
     void sendTimesOutWithAWorkingNote() {
-        McpSchema.CallToolResult res = BridgeMcp.send(messages, "term_a", "hi", 120L);
+        McpSchema.CallToolResult res = BridgeMcp.send(messages, "term_a", "hi", 120L, null, Set.of());
         assertNotEquals(Boolean.TRUE, res.isError(), "a timeout is informational, not a tool error");
         assertTrue(textOf(res).contains("no reply"), "got: " + textOf(res));
     }
 
     @Test
     void sendRejectsMissingArgs() {
-        assertTrue(BridgeMcp.send(messages, null, "hi", null).isError());
-        assertTrue(BridgeMcp.send(messages, "term_a", "  ", null).isError());
+        assertTrue(BridgeMcp.send(messages, null, "hi", null, null, Set.of()).isError());
+        assertTrue(BridgeMcp.send(messages, "term_a", "  ", null, null, Set.of()).isError());
+    }
+
+    @Test
+    void sendRejectsAConfiguredProfileNameBeforeAcceptingIt() {
+        McpSchema.CallToolResult blocking = BridgeMcp.send(messages, "sol", "hi", 100L, null, Set.of("sol"));
+        McpSchema.CallToolResult async = BridgeMcp.sendAsync(messages, "sol", "hi", null, Set.of("sol"));
+
+        assertTrue(blocking.isError());
+        assertTrue(async.isError());
+        assertTrue(textOf(blocking).contains("sol"));
+        assertTrue(textOf(blocking).contains("configured profile name"));
+        assertTrue(textOf(blocking).contains("bridge_list"));
+        assertFalse(textOf(async).contains("ticket="));
+    }
+
+    @Test
+    void sendAllowsPeerLeadMemberAndUnclassifiedTargets() throws Exception {
+        Set<String> profiles = Set.of("sol");
+
+        assertSendRoundTrips("term_peer_lead", profiles);
+        assertSendRoundTrips("term_live_member", profiles);
+
+        // A herdr-owned pane outside the bridge roster cannot be classified at accept time.
+        McpSchema.CallToolResult result = BridgeMcp.send(messages, "external-pane", "hi", 10L, null, profiles);
+        assertFalse(result.isError(), "an unclassified target must not be rejected at acceptance time");
     }
 
     @Test
@@ -173,7 +335,7 @@ class BridgeMcpTest {
     void askThenAnswerRoundTrips() throws Exception {
         // The primary delegates and blocks; wait until its waiter is open before the worker asks.
         CompletableFuture<McpSchema.CallToolResult> send = CompletableFuture.supplyAsync(
-                () -> BridgeMcp.send(messages, "term_a", "do X", 5000L));
+                () -> BridgeMcp.send(messages, "term_a", "do X", 5000L, null, Set.of()));
         long deadline = System.currentTimeMillis() + 3000;
         while (!rendezvous.isWaiting("term_a") && System.currentTimeMillis() < deadline) {
             //noinspection BusyWait
@@ -305,6 +467,42 @@ class BridgeMcpTest {
         assertTrue(out.contains("\"liveStatus\":\"unknown\""), out);
     }
 
+    @Test
+    void capacityUsesThePlacementLiveCount() {
+        FakeHerdr h = new FakeHerdr();
+        SessionManager sessions = new SessionManager(workerService(h, "http://gx00.gw:8000", Set.of("gx00.gw")));
+        sessions.acquire("ltms-local", null, null, null);
+        McpSchema.CallToolResult res = BridgeMcp.listFleet(workerService(h, "http://gx00.gw:8000", Set.of("gx00.gw")),
+                sessions, null, new BridgeMcp.CapacitySource(profile -> 2, profile -> 2,
+                () -> Set.of("ltms-local"), () -> 0), new BridgeMcp.HealthCoverageSource(() -> "off"), Map.of(), "");
+        String out = textOf(res);
+        assertTrue(out.contains("\"maxLoad\":2"), out);
+        assertTrue(out.contains("\"live\":2"), out);
+        assertTrue(out.contains("\"free\":0"), out);
+    }
+
+    @Test
+    void capacityIncludesConfiguredProfileWithoutMembers() {
+        FakeHerdr h = new FakeHerdr();
+        SessionManager sessions = new SessionManager(workerService(h, "http://gx00.gw:8000", Set.of("gx00.gw")));
+        String out = textOf(BridgeMcp.listFleet(workerService(h, "http://gx00.gw:8000", Set.of("gx00.gw")),
+                sessions, null, new BridgeMcp.CapacitySource(profile -> 0, profile -> 2,
+                () -> Set.of("terra"), () -> 0), new BridgeMcp.HealthCoverageSource(() -> "off"), Map.of(), ""));
+        assertTrue(out.contains("\"profile\":\"terra\""), out);
+        assertTrue(out.contains("\"live\":0"), out);
+        assertTrue(out.contains("\"free\":2"), out);
+        assertTrue(out.contains("\"reclaimable\":0"), out);
+    }
+
+    @Test
+    void inertCapacitySourceOmitsCapacityBlock() {
+        FakeHerdr h = new FakeHerdr();
+        String out = textOf(BridgeMcp.listFleet(workerService(h, "http://gx00.gw:8000", Set.of("gx00.gw")),
+                new SessionManager(workerService(h, "http://gx00.gw:8000", Set.of("gx00.gw"))), null,
+                BridgeMcp.CapacitySource.none(), new BridgeMcp.HealthCoverageSource(() -> "off"), Map.of(), ""));
+        assertFalse(out.contains("\"capacity\":"), out);
+    }
+
     @Test
     void listReportsLeadsAndFlagsTheCallersOwnRow() {
         FakeHerdr h = new FakeHerdr();
@@ -578,7 +776,7 @@ class BridgeMcpTest {
         assertTrue(out.contains("\"role\":\"dev\""), out);
         assertTrue(out.contains("\"role\":\"reviewer\""), out);
         assertEquals(2, out.split("\"profile\":\"ltms-local\"", -1).length - 1,
-                "both members share one profile — that is the point: " + out);
+                "inert capacity is omitted, leaving the two member rows: " + out);
     }
 
     @Test

bridged/src/test/java/dev/ltms/bridged/member/ClaudeCodeLauncherTest.java

@@ -102,6 +102,53 @@ class ClaudeCodeLauncherTest {
                 "the operator's own args are preserved, in order, ahead of the model flag");
     }
 
+    @Test
+    void appendsTheBaseComposedRoleAndReplyCharter() {
+        FakeHerdr herdr = new FakeHerdr();
+        String roleCharter = "You review changes.";
+        BridgedConfig.Profile cfg = new BridgedConfig.Profile(
+                "sonnet", "http://gx00.gw:8000", null, null, "BRIDGED_WORKER_TOKEN",
+                List.of("claude"), "tab", "bridged-workers", "w #{n}",
+                "http://127.0.0.1:8765/mcp", null, null);
+        ClaudeCodeLauncher svc = new ClaudeCodeLauncher(new AgentControl(herdr), new WorkspaceControl(herdr),
+                new SubscriptionGuard(Set.of("gx00.gw")), Map.of(cfg.profile(), cfg), cfg.profile(), _ -> null,
+                0, 0L, () -> fleet(Map.of("reviewer", roleCharter), null));
+
+        svc.spawn(new SpawnRequest("sonnet", null, null, null, null, MemberRole.REVIEWER));
+
+        List<String> args = spawnedArgs(herdr);
+        int flag = args.indexOf("--append-system-prompt");
+        assertEquals(1, args.stream().filter("--append-system-prompt"::equals).count(),
+                "the composed charter is passed once");
+        assertEquals(roleCharter + "\n\n" + HerdrPeerLauncher.REPLY_CHARTER, args.get(flag + 1),
+                "the role charter comes first and the reply rule comes last");
+    }
+
+    @Test
+    void profileWithoutMcpOrRoleCharterGetsNoSystemPrompt() {
+        FakeHerdr herdr = new FakeHerdr();
+        ClaudeCodeLauncher svc = labelService(herdr, () -> fleet(Map.of(), null));
+
+        svc.spawn(new SpawnRequest("sonnet", null, null, null, null, MemberRole.DEV));
+
+        assertFalse(spawnedArgs(herdr).contains("--append-system-prompt"));
+    }
+
+    @Test
+    void profileWithoutMcpStillGetsItsRoleCharter() {
+        FakeHerdr herdr = new FakeHerdr();
+        String roleCharter = "You design changes.";
+        ClaudeCodeLauncher svc = labelService(herdr, () -> fleet(Map.of("architect", roleCharter), null));
+
+        svc.spawn(new SpawnRequest("sonnet", null, null, null, null, MemberRole.ARCHITECT));
+
+        List<String> args = spawnedArgs(herdr);
+        int flag = args.indexOf("--append-system-prompt");
+        assertTrue(flag >= 0, "a role charter does not need an MCP mount");
+        assertEquals(roleCharter, args.get(flag + 1));
+        assertFalse(args.contains("--mcp-config"));
+    }
+
     private ClaudeCodeLauncher multiProfile(FakeHerdr herdr) {
         BridgedConfig.Profile gx10 = new BridgedConfig.Profile("gx10", "http://gx10.gw:8000", "coder",
                 null, "BRIDGED_WORKER_TOKEN", List.of("claude"), "tab", "bridged-workers", "w #{n}", null, null, null);
@@ -686,7 +733,7 @@ class ClaudeCodeLauncherTest {
         return new BridgedConfig.Profile(
                 profile, baseUrl, "sonnet", null, "BRIDGED_WORKER_TOKEN",
                 List.of("ccs", profile), "tab", "bridged-workers", "w #{n}", null, null, null,
-                null, null, null, Map.of(), null, null, true);
+                null, null, null, Map.of(), null, null, true, null);
     }
 
     @Test
@@ -772,7 +819,7 @@ class ClaudeCodeLauncherTest {
                 null, null, null,
                 Map.of("ANTHROPIC_BASE_URL", "http://evil.example.com",
                         "ANTHROPIC_AUTH_TOKEN", "sk-ant-bad", "JAVA_HOME", "/opt/jdk"),
-                null, null, true);
+                null, null, true, null);
         new ClaudeCodeLauncher(new AgentControl(herdr), new WorkspaceControl(herdr),
                 new SubscriptionGuard(Set.of("gx00.gw")), Map.of(cfg.profile(), cfg), cfg.profile(),
                 _ -> "would-be-token").spawn();

bridged/src/test/java/dev/ltms/bridged/member/OpenCodeLauncherTest.java

@@ -17,6 +17,10 @@ import java.nio.file.Files;
 import java.nio.file.Path;
 import java.util.List;
 import java.util.Map;
+import java.util.concurrent.ExecutorService;
+import java.util.concurrent.Executors;
+import java.util.concurrent.Future;
+import java.util.function.Supplier;
 
 import static org.junit.jupiter.api.Assertions.*;
 
@@ -34,12 +38,19 @@ class OpenCodeLauncherTest {
     }
 
     /** Gate-disabled launcher whose per-spawn config dirs land under an inspectable temp root. */
-    private OpenCodeLauncher service(FakeHerdr herdr, Path configRoot, BridgedConfig.Profile cfg) {
+    private static OpenCodeLauncher service(FakeHerdr herdr, Path configRoot, BridgedConfig.Profile cfg) {
         return new OpenCodeLauncher(new AgentControl(herdr), new WorkspaceControl(herdr),
                 Map.of(cfg.profile(), cfg), cfg.profile(), k -> "GITEA_ACCESS_TOKEN".equals(k) ? "tok" : null,
                 0, System::currentTimeMillis, () -> { }, configRoot, configRoot);
     }
 
+    private static OpenCodeLauncher service(FakeHerdr herdr, Path configRoot, BridgedConfig.Profile cfg,
+                                            Supplier<BridgedConfig.Fleet> fleet) {
+        return new OpenCodeLauncher(new AgentControl(herdr), new WorkspaceControl(herdr),
+                Map.of(cfg.profile(), cfg), cfg.profile(), _ -> null,
+                0, System::currentTimeMillis, () -> { }, configRoot, configRoot, fleet);
+    }
+
     @SuppressWarnings("unchecked")
     private static Map<String, Object> lastStart(FakeHerdr herdr) {
         return (Map<String, Object>) herdr.lastCall("agent.start").params();
@@ -62,8 +73,10 @@ class OpenCodeLauncherTest {
     @Test
     void writesRemoteMcpConfigAndCharterInstructionsWhenMcpUrlSet(@TempDir Path root) throws Exception {
         FakeHerdr herdr = new FakeHerdr();
-        service(herdr, root, opencodeCfg("google/gemini-2.5-pro", "http://127.0.0.1:8765/mcp", null))
-                .spawn();
+        BridgedConfig.Fleet fleet = new BridgedConfig.Fleet(Map.of(), Map.of(), Map.of(), Map.of(),
+                Map.of("dev", "role rule"), null);
+        service(herdr, root, opencodeCfg("google/gemini-2.5-pro", "http://127.0.0.1:8765/mcp", null),
+                () -> fleet).spawn();
 
         Map<String, String> env = startEnv(herdr);
         assertNull(env.get("ANTHROPIC_BASE_URL"), "opencode carries no ANTHROPIC_* / subscription boundary");
@@ -82,13 +95,15 @@ class OpenCodeLauncherTest {
                 "the profile's bridge MCP url is present");
         assertTrue(bridge.path("enabled").asBoolean(), "the bridge server is enabled");
         assertTrue(json.path("instructions").isArray() && !json.path("instructions").isEmpty(),
-                "the reply charter is mounted via instructions");
+                "the member charter is mounted via instructions");
 
-        // The instructions entry is a real file path holding the reply charter.
-        Path charter = Path.of(cfgPath).resolveSibling("reply-charter.md");
+        // The instructions entry is a real file path holding the composed member charter.
+        Path charter = Path.of(cfgPath).resolveSibling("member-charter.md");
         assertTrue(Files.exists(charter), "the charter file the config references was written");
-        assertTrue(Files.readString(charter).contains("bridge_reply"),
-                "the charter instructs the worker to answer via bridge_reply");
+        assertEquals("role rule\n\n" + HerdrPeerLauncher.REPLY_CHARTER, Files.readString(charter),
+                "the composed charter keeps the role rule first and the reply rule last");
+        assertEquals(charter.toAbsolutePath().toString(), json.path("instructions").get(0).asText(),
+                "instructions names the charter file by its absolute path");
     }
 
     @Test
@@ -100,6 +115,69 @@ class OpenCodeLauncherTest {
                 "no bridge MCP url → no config file and no OPENCODE_CONFIG");
     }
 
+    @Test
+    void roleCharterWithoutMcpOrCustomProviderStillWritesAConfig(@TempDir Path root) throws Exception {
+        FakeHerdr herdr = new FakeHerdr();
+        BridgedConfig.Fleet fleet = new BridgedConfig.Fleet(Map.of(), Map.of(), Map.of(), Map.of(),
+                Map.of("dev", "role rule"), null);
+        Path configRoot = Files.createDirectory(root.resolve("configs"));
+        Path checkout = Files.createDirectory(root.resolve("checkout"));
+        service(herdr, configRoot, opencodeCfg("google/gemini-2.5-pro", null, null), () -> fleet).spawn();
+
+        String cfgPath = startEnv(herdr).get("OPENCODE_CONFIG");
+        assertNotNull(cfgPath, "a role charter needs a config even without MCP or custom provider");
+        JsonNode json = new ObjectMapper().readTree(Path.of(cfgPath).toFile());
+        Path charter = Path.of(json.path("instructions").get(0).asText());
+        assertEquals("role rule", Files.readString(charter), "the base-composed role charter is unchanged");
+        assertTrue(json.path("mcp").isMissingNode(), "a charter does not add an MCP mount");
+        assertTrue(charter.startsWith(configRoot), "the charter is written under the temp config root");
+        try (var files = Files.walk(checkout)) {
+            assertFalse(files.anyMatch(path -> path.getFileName().toString().equals("member-charter.md")),
+                    "the worker checkout receives no charter file");
+        }
+    }
+
+    @Test
+    void nullCharterWritesNoCharterFileOrInstructions(@TempDir Path root) throws Exception {
+        FakeHerdr herdr = new FakeHerdr();
+        service(herdr, root, pinnedCfg("local-vllm/model", "http://127.0.0.1:8000", null),
+                () -> new BridgedConfig.Fleet(Map.of(), Map.of(), Map.of(), Map.of(), Map.of(), null)).spawn();
+
+        String config = startEnv(herdr).get("OPENCODE_CONFIG");
+        assertNotNull(config, "the custom provider still needs a config");
+        JsonNode json = new ObjectMapper().readTree(Path.of(config).toFile());
+        assertTrue(json.path("instructions").isMissingNode(), "a null charter adds no instructions entry");
+        try (var files = Files.walk(root)) {
+            assertFalse(files.anyMatch(path -> path.getFileName().toString().equals("member-charter.md")),
+                    "a null charter creates no charter file");
+        }
+    }
+
+    @Test
+    void concurrentSpawnsWriteSeparateCharterDirectories(@TempDir Path root) throws Exception {
+        BridgedConfig.Profile cfg = opencodeCfg("google/gemini-2.5-pro", "http://127.0.0.1:8765/mcp", null);
+        ExecutorService executor = Executors.newFixedThreadPool(2);
+        try {
+            Future<String> first = executor.submit(() -> spawnConfigPath(root, cfg));
+            Future<String> second = executor.submit(() -> spawnConfigPath(root, cfg));
+
+            Path firstCharter = Path.of(first.get()).resolveSibling("member-charter.md");
+            Path secondCharter = Path.of(second.get()).resolveSibling("member-charter.md");
+            assertNotEquals(firstCharter.getParent(), secondCharter.getParent(),
+                    "each concurrent spawn owns a separate config directory");
+            assertTrue(Files.exists(firstCharter));
+            assertTrue(Files.exists(secondCharter));
+        } finally {
+            executor.shutdownNow();
+        }
+    }
+
+    private static String spawnConfigPath(Path root, BridgedConfig.Profile cfg) {
+        FakeHerdr herdr = new FakeHerdr();
+        service(herdr, root, cfg).spawn();
+        return startEnv(herdr).get("OPENCODE_CONFIG");
+    }
+
     @Test
     void passesTheModelAsDashMFlagAlongsideAutoApprove(@TempDir Path root) {
         FakeHerdr herdr = new FakeHerdr();

bridged/src/test/java/dev/ltms/bridged/msg/MessageServiceTest.java

@@ -5,6 +5,7 @@ import dev.ltms.bridged.herdr.AgentStatus;
 import dev.ltms.bridged.herdr.FakeHerdr;
 import dev.ltms.bridged.herdr.HerdrException;
 import dev.ltms.bridged.inject.CompletionResolver;
+import dev.ltms.bridged.inject.ExhaustedPatternLookup;
 import dev.ltms.bridged.mcp.PrimaryRegistry;
 import dev.ltms.bridged.inject.Injector;
 import org.junit.jupiter.api.BeforeEach;
@@ -31,7 +32,8 @@ class MessageServiceTest {
     private final FakeHerdr herdr = new FakeHerdr().readText("BUILD GREEN: 391 files");
     private final AgentControl agents = new AgentControl(herdr);
     private final Rendezvous rendezvous = new Rendezvous();
-    private final CompletionResolver completion = new CompletionResolver(agents, rendezvous);
+    private final CompletionResolver completion =
+            new CompletionResolver(agents, rendezvous, ExhaustedPatternLookup.none());
     private final Injector injector = new Injector(agents, completion);
     private final InMemoryReplyInbox inbox = new InMemoryReplyInbox();
     private final MessageService messages = new MessageService(agents, injector, rendezvous, inbox);
@@ -120,6 +122,36 @@ class MessageServiceTest {
         assertFalse(reply.completed());
     }
 
+    @Test
+    void droppedQueuedAndDeliveredTurnsExposeTheRealCauseExactlyOnce() throws Exception {
+        CompletableFuture<MessageService.Reply> first = sendAsync();
+        awaitWaiting();
+        injector.onStatus(T, AgentStatus.IDLE);    // first delivery
+        injector.onStatus(T, AgentStatus.WORKING); // first turn in flight
+
+        CompletableFuture<Void> queued = injector.enqueue(T, "second task", TestTurnTokens.inert(T));
+        CompletableFuture<Rendezvous.Resolution> waiter = rendezvous.currentWaiter(T);
+        injector.drop(T, new HerdrException("agent target sol not found", "agent_not_found", null));
+
+        MessageService.Reply reply = first.get(5, TimeUnit.SECONDS);
+        assertEquals(MessageService.Outcome.WORKER_FAILED, reply.outcome());
+        assertEquals("agent target sol not found", reply.text());
+        assertTrue(queued.isCompletedExceptionally(), "the queued delivery future also fails");
+        assertFalse(rendezvous.resolveFailure(waiter, "second failure"), "the waiter fails exactly once");
+    }
+
+    @Test
+    void noDropReasonKeepsTheExistingFallbackText() throws Exception {
+        herdr.readText("");
+        CompletableFuture<Rendezvous.Resolution> waiter = rendezvous.open(T);
+
+        completion.onTurnFailed(T);
+
+        Rendezvous.Resolution resolution = waiter.get(2, TimeUnit.SECONDS);
+        assertEquals("worker did not reply; its turn ended in an unrecoverable state "
+                + "(worker unreachable or stuck)", resolution.text());
+    }
+
     // --- bridge_ask reverse rendezvous (CB-205) ------------------------------------------------
 
     @Test
@@ -609,4 +641,94 @@ class MessageServiceTest {
         assertTrue(view.detail() != null && view.detail().contains("released"),
                 "and the detail says why, rather than 'worker unknown'");
     }
+
+    @Test
+    void abandonFailsEveryPendingAsyncTicketForTheReleasedTarget() throws Exception {
+        String first = messages.sendAsync(T, "first task");
+        awaitWaiting(); // first task owns the target lock and rendezvous waiter
+        String second = messages.sendAsync(T, "second task"); // parked on the same lock, not yet queued
+        String third = messages.sendAsync(T, "third task"); // a second queued ticket proves the full sweep
+
+        assertTrue(messages.abandon(T, "agent target term_a not found"));
+
+        assertFailedTicket(first, "agent target term_a not found");
+        assertFailedTicket(second, "agent target term_a not found");
+        assertFailedTicket(third, "agent target term_a not found");
+    }
+
+    @Test
+    void abandonDoesNotFailAnAsyncTicketWaitingForAnAnswer() throws Exception {
+        String ticket = messages.sendAsync(T, "task that asks");
+        awaitWaiting();
+        injectDelivery();
+
+        CompletableFuture<MessageService.AskResult> ask =
+                CompletableFuture.supplyAsync(() -> messages.ask(T, "which config?", 5000));
+        MessageService.TaskView asking = awaitTicketPhase(ticket, MessageService.Phase.ASKING);
+
+        assertFalse(messages.abandon(T, "agent target term_a not found"),
+                "an asking ticket is an active turn, not a pending send to sweep");
+        assertEquals(MessageService.Phase.ASKING, messages.poll(ticket).phase());
+
+        CompletableFuture<MessageService.Reply> answer = CompletableFuture.supplyAsync(
+                () -> messages.answer(asking.turnId(), "config.yaml", 5000));
+        assertEquals("config.yaml", ask.get(5, TimeUnit.SECONDS).answer());
+        awaitWaiting();
+        assertTrue(rendezvous.resolve(T, "done"));
+        assertEquals(MessageService.Outcome.REPLIED, answer.get(5, TimeUnit.SECONDS).outcome());
+    }
+
+    @Test
+    void unansweredAsyncQuestionReturnsTheTicketToPendingAndReleasesItsTarget() throws Exception {
+        String ticket = messages.sendAsync(T, "task that asks");
+        awaitWaiting();
+        injectDelivery();
+
+        assertEquals(MessageService.AskOutcome.TIMED_OUT,
+                messages.ask(T, "which config?", 200).outcome());
+        assertEquals(MessageService.Phase.PENDING, messages.poll(ticket).phase(),
+                "only the question wait ended; the delegated turn may still finish");
+
+        String next = messages.sendAsync(T, "next task");
+        awaitWaiting();
+        assertTrue(rendezvous.resolve(T, "done"));
+        assertEquals(MessageService.Phase.DONE, awaitTicketPhase(next, MessageService.Phase.DONE).phase());
+    }
+
+    @Test
+    void asyncQuestionBelongsToTheTaskThatOwnsItsForwardWaiter() throws Exception {
+        String first = messages.sendAsync(T, "first task");
+        awaitWaiting();
+
+        CompletableFuture<MessageService.AskResult> ask =
+                CompletableFuture.supplyAsync(() -> messages.ask(T, "which config?", 5000));
+        assertEquals(MessageService.Phase.ASKING, awaitTicketPhase(first, MessageService.Phase.ASKING).phase());
+
+        MessageService.TaskView asking = messages.poll(first);
+        CompletableFuture<MessageService.Reply> answer = CompletableFuture.supplyAsync(
+                () -> messages.answer(asking.turnId(), "config.yaml", 5000));
+        assertEquals("config.yaml", ask.get(5, TimeUnit.SECONDS).answer());
+        awaitWaiting();
+        assertTrue(rendezvous.resolve(T, "done"));
+        assertEquals(MessageService.Outcome.REPLIED, answer.get(5, TimeUnit.SECONDS).outcome());
+    }
+
+    private void assertFailedTicket(String ticket, String reason) throws Exception {
+        MessageService.TaskView view = awaitTicketPhase(ticket, MessageService.Phase.FAILED);
+        assertEquals(reason, view.detail());
+    }
+
+    private MessageService.TaskView awaitTicketPhase(String ticket, MessageService.Phase phase) throws Exception {
+        long deadline = System.currentTimeMillis() + 3000;
+        MessageService.TaskView view;
+        do {
+            view = messages.poll(ticket);
+            if (view.phase() == phase) {
+                return view;
+            }
+            Thread.sleep(5);
+        } while (System.currentTimeMillis() < deadline);
+        assertEquals(phase, view.phase());
+        return view;
+    }
 }

bridged/src/test/java/dev/ltms/bridged/msg/RendezvousTest.java

@@ -114,6 +114,17 @@ class RendezvousTest {
                 "the first resolution wins; the stored value is unchanged");
     }
 
+    @Test
+    void resolveExhaustedTwiceIsANoOpTheSecondTime() {
+        CompletableFuture<Rendezvous.Resolution> waiter = rendezvous.open(W);
+        assertTrue(rendezvous.resolveExhausted(waiter, "first reason"), "the first classification resolves");
+        assertFalse(rendezvous.resolveExhausted(waiter, "second reason"),
+                "a second exhausted resolution on an already-resolved waiter returns false");
+        assertEquals(Rendezvous.Kind.BACKEND_EXHAUSTED, waiter.getNow(null).kind());
+        assertEquals("first reason", waiter.getNow(null).text(),
+                "the first resolution wins; the stored value is unchanged");
+    }
+
     @Test
     void closeAskRemovesTheTurn() {
         Rendezvous.AskTicket t = rendezvous.openAsk(W);

bridged/src/test/java/dev/ltms/bridged/msg/TestTurnTokens.java

@@ -0,0 +1,20 @@
+package dev.ltms.bridged.msg;
+
+/**
+ * Explicit unbound tokens for tests that exercise delivery without an accepted send.
+ *
+ * <p>The waiter is {@code null} on purpose. "No accepted send" is an <em>absence</em>, and a helper
+ * that handed back a fresh {@code CompletableFuture} would invent one — which is how the first
+ * version of this class turned {@code captureBaselineSkipsTheReadWhenNoSendIsWaiting} red: the
+ * resolver saw a non-null waiter, decided a turn was in flight, and scraped a pane that no send was
+ * blocked on. An inert value must omit the fact, never fabricate it.
+ */
+public final class TestTurnTokens {
+    private TestTurnTokens() {
+    }
+
+    /** A token for a delivery that no send is waiting on: it authorises nothing. */
+    public static TurnToken inert(String target) {
+        return new TurnToken(target, null);
+    }
+}

bridged/src/test/java/dev/ltms/bridged/session/FakeWorktrees.java

@@ -29,6 +29,7 @@ public final class FakeWorktrees implements Worktrees {
     private final Set<String> existingPaths = ConcurrentHashMap.newKeySet();
     private final Set<String> trackedPaths = ConcurrentHashMap.newKeySet();
     private volatile RuntimeException addFailure;
+    private volatile boolean dirty = false;
     private volatile String repoRoot = "/repo";
     private volatile String prefix = "/worktrees";
 
@@ -61,6 +62,12 @@ public final class FakeWorktrees implements Worktrees {
         return this;
     }
 
+    /** Mark the worktree dirty so {@link #hasUncommitted} reports true (simulates uncommitted work). */
+    public FakeWorktrees withDirty(boolean dirty) {
+        this.dirty = dirty;
+        return this;
+    }
+
     @Override
     public String add(String repoRoot, String branch, String baseRef) {
         addCalls.add(new AddCall(repoRoot, branch, baseRef));
@@ -77,6 +84,11 @@ public final class FakeWorktrees implements Worktrees {
         removeCalls.add(new RemoveCall(repoRoot, worktreePath));
     }
 
+    @Override
+    public boolean hasUncommitted(String worktreePath) {
+        return dirty;
+    }
+
     @Override
     public void overlayParity(String repoRoot, String worktreePath, List<String> overlay) {
         List<String> copied = new java.util.ArrayList<>();

bridged/src/test/java/dev/ltms/bridged/session/GitWorktreesTest.java

@@ -175,6 +175,46 @@ class GitWorktreesTest {
         assertTrue(Files.exists(Path.of(wt).resolve(".mcp.json")), ".mcp.json stub was dropped");
     }
 
+    /**
+     * CB-576. {@code hasUncommitted} must treat a freshly-provisioned worktree as clean, but a
+     * worktree holding a brand-new, never-added file as dirty. The untracked-file-only shape is
+     * exactly the work lost in the incident — a worker's draft that compiled but was never
+     * committed because it stopped to ask its lead a question.
+     */
+    @Test
+    void anUntrackedOnlyWorktreeCountsAsDirty(@TempDir Path tmp) throws Exception {
+        Path repo = initRepo(tmp.resolve("repo"));
+        GitWorktrees gitWorktrees = new GitWorktrees(tmp.resolve("wts").toString());
+        String wt = gitWorktrees.add(repo.toString(), "cb-576-u", "HEAD");
+
+        assertFalse(gitWorktrees.hasUncommitted(wt),
+                "a freshly provisioned worktree must read as clean");
+
+        Files.writeString(Path.of(wt).resolve("brand-new.txt"), "draft that was never added\n");
+
+        assertTrue(gitWorktrees.hasUncommitted(wt),
+                "an untracked-only file must count as dirty");
+
+        Files.writeString(Path.of(wt).resolve("README.md"), "edited tracked file\n");
+        assertTrue(gitWorktrees.hasUncommitted(wt),
+                "a tracked modification must also count as dirty");
+    }
+
+    /**
+     * CB-576 review. {@code hasUncommitted} must tolerate a missing worktree exactly like
+     * {@code remove}: an already-gone directory holds no work to lose, and throwing here would
+     * break teardown — SessionManager.release() calls it before stopping the pane, so an
+     * exception would orphan a live pane and skip the release notification (CB-516).
+     */
+    @Test
+    void hasUncommittedOnAMissingWorktreeReturnsFalseWithoutThrowing(@TempDir Path tmp) {
+        GitWorktrees gitWorktrees = new GitWorktrees(tmp.resolve("wts").toString());
+        String gone = tmp.resolve("wts").resolve("does-not-exist").toString();
+
+        assertFalse(gitWorktrees.hasUncommitted(gone),
+                "a missing worktree is reported clean, not an error");
+    }
+
     /** All three protected configs are covered: each one present in a worktree is neutralized and hidden. */
     @Test
     void allThreeConfigsAreNeutralizedWhenPresent(@TempDir Path tmp) throws Exception {

bridged/src/test/java/dev/ltms/bridged/session/SessionManagerTest.java

@@ -10,6 +10,7 @@ import dev.ltms.bridged.herdr.AgentControl;
 import dev.ltms.bridged.herdr.FakeHerdr;
 import dev.ltms.bridged.herdr.WorkspaceControl;
 import dev.ltms.bridged.member.ClaudeCodeLauncher;
+import dev.ltms.bridged.msg.TestTurnTokens;
 import dev.ltms.bridged.peer.PeerUnreachableException;
 import org.junit.jupiter.api.Test;
 import org.slf4j.LoggerFactory;
@@ -101,7 +102,7 @@ class SessionManagerTest {
 
         assertDoesNotThrow(() -> sessions.asPresence().markPresent(null),
                 "the primary's null terminal must not blow up an unrelated tool call");
-        assertDoesNotThrow(() -> sessions.onDelivered(null));
+        assertDoesNotThrow(() -> sessions.onDelivered(null, TestTurnTokens.inert(null)));
         assertDoesNotThrow(() -> sessions.onTurnComplete(null));
         assertDoesNotThrow(() -> sessions.onTurnFailed(null));
 
@@ -121,7 +122,7 @@ class SessionManagerTest {
                 "MCP presence moves SPAWNING → READY");
         assertTrue(sessions.asPresence().isPresent(terminal), "presence is also recorded");
 
-        sessions.onDelivered(terminal);
+        sessions.onDelivered(terminal, TestTurnTokens.inert(terminal));
         assertEquals(MemberSession.State.BUSY, sessions.get(session.paneId()).orElseThrow().state(),
                 "delivery moves READY → BUSY");
 
@@ -153,7 +154,7 @@ class SessionManagerTest {
         MemberSession session = sessions.acquire("ltms-local", null, "/caller", "term_primary");
         String terminal = session.terminalId();
         sessions.asPresence().markPresent(terminal);
-        sessions.onDelivered(terminal);
+        sessions.onDelivered(terminal, TestTurnTokens.inert(terminal));
 
         sessions.onTurnFailed(terminal);
 
@@ -181,7 +182,7 @@ class SessionManagerTest {
             MemberSession session = sessions.acquire("ltms-local", null, "/caller", "term_primary");
             String terminal = session.terminalId();
             sessions.asPresence().markPresent(terminal);
-            sessions.onDelivered(terminal);
+            sessions.onDelivered(terminal, TestTurnTokens.inert(terminal));
 
             sessions.onTurnFailed(terminal);
 
@@ -264,7 +265,7 @@ class SessionManagerTest {
         MemberSession session = sessions.acquire("ltms-local", null, "/caller", "term_primary");
         String terminal = session.terminalId();
         sessions.asPresence().markPresent(terminal);
-        sessions.onDelivered(terminal);
+        sessions.onDelivered(terminal, TestTurnTokens.inert(terminal));
 
         clock[0] = 100;
         assertEquals(0, sessions.reapIdle(10), "BUSY session past TTL is never reaped");
@@ -281,7 +282,7 @@ class SessionManagerTest {
         MemberSession session = sessions.acquire("ltms-local", null, "/caller", "term_primary");
         String terminal = session.terminalId();
         sessions.asPresence().markPresent(terminal);
-        sessions.onDelivered(terminal);
+        sessions.onDelivered(terminal, TestTurnTokens.inert(terminal));
         sessions.onTurnComplete(terminal);
 
         clock[0] = 21;
@@ -299,7 +300,7 @@ class SessionManagerTest {
         MemberSession busy = sessions.acquire("ltms-local", "/busy", "/caller", "owner2");
         sessions.asPresence().markPresent(ready.terminalId());
         sessions.asPresence().markPresent(busy.terminalId());
-        sessions.onDelivered(busy.terminalId());
+        sessions.onDelivered(busy.terminalId(), TestTurnTokens.inert(busy.terminalId()));
 
         clock[0] = 50;
         assertEquals(1, sessions.reapIdle(30), "only READY past TTL is reaped");
@@ -317,9 +318,9 @@ class SessionManagerTest {
         String terminal = session.terminalId();
         sessions.asPresence().markPresent(terminal);
 
-        sessions.onDelivered(terminal);
+        sessions.onDelivered(terminal, TestTurnTokens.inert(terminal));
         sessions.onTurnComplete(terminal);
-        sessions.onDelivered(terminal);
+        sessions.onDelivered(terminal, TestTurnTokens.inert(terminal));
         sessions.onTurnComplete(terminal);
 
         MemberSession updated = sessions.get(session.paneId()).orElseThrow();
@@ -337,13 +338,13 @@ class SessionManagerTest {
         String terminal = session.terminalId();
         sessions.asPresence().markPresent(terminal);
 
-        sessions.onDelivered(terminal);
+        sessions.onDelivered(terminal, TestTurnTokens.inert(terminal));
         sessions.onTurnComplete(terminal);
         assertEquals(MemberSession.State.DONE,
                 sessions.get(session.paneId()).orElseThrow().state(),
                 "first turn completes without release");
 
-        sessions.onDelivered(terminal);
+        sessions.onDelivered(terminal, TestTurnTokens.inert(terminal));
         sessions.onTurnComplete(terminal);
 
         assertTrue(sessions.get(session.paneId()).isEmpty(), "session released after cap reached");
@@ -359,7 +360,7 @@ class SessionManagerTest {
         MemberSession session = sessions.acquire("ltms-local", null, "/caller", "term_primary");
         sessions.asPresence().markPresent(session.terminalId());
 
-        sessions.onDelivered(session.terminalId());
+        sessions.onDelivered(session.terminalId(), TestTurnTokens.inert(session.terminalId()));
         assertTrue(sessions.onTurnCompleteWithPostAction(session.terminalId()));
 
         MemberSession updated = sessions.get(session.paneId()).orElseThrow();
@@ -373,7 +374,7 @@ class SessionManagerTest {
         SessionManager sessions = sessionManager(herdr, () -> 0L, 1, true);
         MemberSession session = sessions.acquire("ltms-local", null, "/caller", "term_primary");
         sessions.asPresence().markPresent(session.terminalId());
-        sessions.onDelivered(session.terminalId());
+        sessions.onDelivered(session.terminalId(), TestTurnTokens.inert(session.terminalId()));
 
         assertFalse(sessions.hasPostTurnAction(session.terminalId()),
                 "a session at its cap will be released, not reset for reuse");
@@ -388,7 +389,7 @@ class SessionManagerTest {
         SessionManager sessions = sessionManager(herdr, () -> 0L, 0, false);
         MemberSession session = sessions.acquire("ltms-local", null, "/caller", "term_primary");
         sessions.asPresence().markPresent(session.terminalId());
-        sessions.onDelivered(session.terminalId());
+        sessions.onDelivered(session.terminalId(), TestTurnTokens.inert(session.terminalId()));
 
         sessions.onTurnComplete(session.terminalId());
 
@@ -406,7 +407,7 @@ class SessionManagerTest {
         MemberSession busy = sessions.acquire("ltms-local", "/busy", "/caller", "ownerB");
         sessions.asPresence().markPresent(ready.terminalId());
         sessions.asPresence().markPresent(busy.terminalId());
-        sessions.onDelivered(busy.terminalId());
+        sessions.onDelivered(busy.terminalId(), TestTurnTokens.inert(busy.terminalId()));
 
         sessions.drainAll(TimeUnit.MILLISECONDS.toNanos(100));
 

bridged/src/test/java/dev/ltms/bridged/session/WorktreeSessionManagerTest.java

@@ -6,9 +6,15 @@ import dev.ltms.bridged.guard.SubscriptionGuard;
 import dev.ltms.bridged.herdr.AgentControl;
 import dev.ltms.bridged.herdr.FakeHerdr;
 import dev.ltms.bridged.herdr.WorkspaceControl;
+import ch.qos.logback.classic.Level;
+import ch.qos.logback.classic.LoggerContext;
+import ch.qos.logback.classic.spi.ILoggingEvent;
+import ch.qos.logback.core.read.ListAppender;
 import dev.ltms.bridged.member.ClaudeCodeLauncher;
+import dev.ltms.bridged.msg.TestTurnTokens;
 import dev.ltms.bridged.peer.MemberRole;
 import org.junit.jupiter.api.Test;
+import org.slf4j.LoggerFactory;
 
 import java.util.List;
 import java.util.Map;
@@ -178,6 +184,49 @@ class WorktreeSessionManagerTest {
         assertTrue(sessions.get(paneId).isEmpty(), "released session is no longer retrievable");
     }
 
+    /**
+     * CB-576. A normal {@code COMPLETED} release whose worktree holds uncommitted work must NOT
+     * remove it — {@code --force} would destroy the worker's only copy. The bridge cannot see
+     * uncommitted files, so the worktree is preserved and the release logged at WARN naming the
+     * path, the session, and the cause an operator needs to find the work.
+     */
+    @Test
+    void releasePreservesDirtyWorktreeAndLogsWarn() {
+        FakeHerdr herdr = new FakeHerdr();
+        FakeWorktrees worktrees = new FakeWorktrees().withRepoRoot("/repo").withPrefix("/wt")
+                .withDirty(true);
+        SessionManager sessions = new SessionManager(workerService(herdr), worktrees);
+        MemberSession s = sessions.acquire("ltms-local", null, "/caller/proj", null,
+                new WorktreeRequest("cb-576", null));
+
+        LoggerContext ctx = (LoggerContext) LoggerFactory.getILoggerFactory();
+        ch.qos.logback.classic.Logger sessionLog =
+                (ch.qos.logback.classic.Logger) LoggerFactory.getLogger(SessionManager.class);
+        ListAppender<ILoggingEvent> appender = new ListAppender<>();
+        appender.setContext(ctx);
+        appender.start();
+        sessionLog.addAppender(appender);
+        sessionLog.setLevel(Level.WARN);
+        try {
+            sessions.release(s.paneId());
+
+            assertTrue(herdr.called("pane.close"), "release still tears the worker pane down");
+            assertTrue(worktrees.removeCalls().isEmpty(),
+                    "a dirty worktree is never removed — it holds the only copy of the work");
+            String warn = appender.list.stream()
+                    .filter(e -> e.getLevel().equals(Level.WARN))
+                    .map(ILoggingEvent::getFormattedMessage)
+                    .filter(m -> m.contains("dirty worktree"))
+                    .findFirst()
+                    .orElse("no dirty-release WARN logged");
+            assertTrue(warn.contains(s.worktree()), "the WARN names the worktree path: " + warn);
+            assertTrue(warn.contains(s.terminalId()), "the WARN names the session: " + warn);
+            assertTrue(warn.contains("COMPLETED"), "the WARN names the release cause: " + warn);
+        } finally {
+            sessionLog.detachAppender(appender);
+        }
+    }
+
     @Test
     void drainAllPreservesWorktreeOfIdleSession() {
         FakeHerdr herdr = new FakeHerdr();
@@ -204,7 +253,7 @@ class WorktreeSessionManagerTest {
                 new WorktreeRequest("cb-544", null));
         String terminal = s.terminalId();
         sessions.asPresence().markPresent(terminal);
-        sessions.onDelivered(terminal); // BUSY, never completes → still BUSY when the timeout hits
+        sessions.onDelivered(terminal, TestTurnTokens.inert(terminal)); // BUSY, never completes → still BUSY when the timeout hits
 
         sessions.drainAll(TimeUnit.MILLISECONDS.toNanos(100));
 

bridged/src/test/resources/logback-test.xml

@@ -16,6 +16,9 @@
           AuditLogTest, which attaches its own ListAppender and asserts on emitted records.
     -->
 
+    <!-- Keep the test logger behaviour aligned with the production cancellation filter. -->
+    <turboFilter class="dev.ltms.bridged.logging.McpCancelledNotificationFilter"/>
+
     <appender name="STDOUT" class="ch.qos.logback.core.ConsoleAppender">
         <encoder>
             <pattern>%d{HH:mm:ss.SSS} [%thread] %-5level %logger{36} - %msg%n</pattern>
@@ -34,4 +37,4 @@
         <appender-ref ref="STDOUT"/>
     </root>
 
-</configuration>
+</configuration>

docs/M4-Fleet-Health.md

@@ -0,0 +1,937 @@
+# M4 - Fleet health, recovery, routing, and capacity
+
+**Status:** Design accepted on 2026-08-15. CB-573 part 1 has shipped the classification model and
+the `bridge_list` capacity view; the remaining M4 units are not yet shipped.
+**Scope:** Fleet evidence, safe mechanical repair, lead routing, capacity reporting, and optional
+human notification.
+**Grounded in:** `health/FleetHealth`, `health/PaneBudget`, `inject/StatusPoller`,
+`inject/StatusRefiner`, `inject/CompletionResolver`, `inject/Injector`, `session/SessionManager`,
+`msg/MessageService`, `msg/ReplyInbox`, `msg/ReplyPushLoop`, `msg/LeadHeartbeatLoop`,
+`mcp/PrimaryRegistry`, and `herdr/AgentControl`.
+
+## 1. Problem and decision boundary
+
+The operator asked the bridge to detect idle agents, exceptions, stopped work, and broken
+communication. The bridge may read an agent pane from time to time. It must notify a person when
+the fleet cannot move forward.
+
+The four operator terms are not four equal health states. `IDLE` is a normal mode. An exception is
+sometimes visible only as pane text. Stopped work may look the same as slow work. Broken
+communication can occur on several links.
+
+M4 uses this boundary:
+
+- The bridge detects facts and joins evidence.
+- The bridge repairs only mechanical failures with no judgement.
+- The lead decides whether to stop, retry, replace, or reassign a member.
+- A human is notified only when no healthy lead can act.
+- n8n may route an outbound incident. It never classifies state or chooses recovery.
+
+An inbound n8n decider would need bridge authority. No narrow machine-decider role exists. Giving a
+workflow engine lead authority is unsafe, while adding a new role is a separate authorization
+design. An outbound sink needs no bridge role.
+
+The bridge must never replay a delivered task. That task may already have changed files, pushed a
+branch, opened a pull request, or changed external state. A replay can run those side effects twice.
+This rule must remain true even if later code stores delivered prompt text.
+
+## 2. Evidence model
+
+A health state is mainly a comparison between two views:
+
+- **herdr view:** current agents and raw live status from one `AgentControl.list()` call.
+- **bridge view:** session FSM, MCP presence, accepted turns, tasks, inbox state, and lead ownership.
+
+A strong fault often appears as a disagreement between those views. For example, `BUSY` in the
+session FSM and `DONE` in herdr means the bridge missed a turn boundary. Pane reads support this
+model, but they are not the main monitor.
+
+`SessionManager.rosterView` already joins session state and live status. `AgentControl.list()`
+already gets the whole live fleet in one call. M4 makes that join persistent and adds timers,
+accepted-turn state, and incident state.
+
+### 2.1 Real traces behind the design
+
+The first trace was an architect that stopped making progress:
+
+```text
+profile=opus role=architect state=busy liveStatus=done
+```
+
+The session moved from `DONE` to `BUSY` for turn 2. Eighteen minutes later, the session still said
+`BUSY`, herdr still said `DONE`, the async task still said `PENDING`, and no completion fallback had
+run. This is `TURN_BOUNDARY_LOST`, not a general slow-turn guess.
+
+The second trace had two async sends to the same pane, one second apart. The pane was then stopped.
+One ticket became failed. The other stayed `pending - worker unknown`. Current
+`MessageService.abandon` resolves only `Rendezvous.currentWaiter(target)`, while async tasks live in
+a separate ticket map. CB-568 is intended to fix that bug. M4 still keeps an independent
+post-teardown invariant so a later regression becomes `DELEGATION_ORPHANED`.
+
+### 2.2 Corrections made during design
+
+The first state table missed `BUSY` in bridged plus `IDLE` or `DONE` in herdr. It would have found
+the real trace only through a late, weak stall timer. The final model adds
+`TURN_BOUNDARY_LOST` as a strong disagreement state.
+
+The first notification design also required a webhook before `health.enabled` could turn on. That
+removed useful local detection to avoid a narrower human-notification gap. The final design splits
+detection from notification. Missing human escalation is shown as partial coverage instead of
+disabling health.
+
+## 3. Classification precedence
+
+Evidence is applied in this order. A lower rule cannot hide a higher one.
+
+1. **Control link:** failed fleet list plus failed ping becomes `CONTROL_LINK_DOWN`.
+2. **Definitive target loss:** `_not_found` becomes `GONE` or `LEAD_UNREACHABLE` when the control
+   link is healthy.
+3. **Startup and teardown invariants:** readiness expiry becomes `NEVER_READY`; surviving tasks
+   after teardown become `DELEGATION_ORPHANED`.
+4. **Bridge/live disagreement:** `BUSY` plus stable raw `IDLE` or `DONE` becomes
+   `TURN_BOUNDARY_LOST`.
+5. **Known screen evidence:** a tested fatal signature becomes `ERROR_ON_SCREEN`.
+6. **Timed suspicion:** unchanged sparse pane probes may become `STALL_SUSPECTED`.
+7. **Communication quality:** completion fallback becomes `MUTE`; an old inbox entry becomes
+   `REPLY_STRANDED`.
+8. **Normal mode:** `STARTING`, `IDLE`, `WORKING`, `WORK_PENDING`, or `BLOCKED_AMBIGUOUS`.
+
+The member flow in Figure 1 shows lifecycle states and the main fault exits. Fault states are
+reported beside the session FSM; most are not new FSM values.
+
+```mermaid
+flowchart TD
+    Registered["Member registered"] --> Starting["STARTING"]
+    Starting -->|"MCP presence"| Idle["IDLE"]
+    Starting -->|"Readiness grace expires"| NeverReady["NEVER_READY"]
+    Idle -->|"Accepted delivery"| Working["WORKING"]
+    Working -->|"Trusted turn boundary"| Idle
+    Working -->|"Bridge BUSY and herdr IDLE or DONE"| Lost["TURN_BOUNDARY_LOST"]
+    Working -->|"Known fatal screen"| Error["ERROR_ON_SCREEN"]
+    Working -->|"Long age and unchanged sparse probes"| Stall["STALL_SUSPECTED"]
+    Working -->|"Target not found"| Gone["GONE"]
+    Idle -->|"Inbox or queued delivery exists"| Pending["WORK_PENDING"]
+    Pending -->|"Delivery or collection finishes"| Idle
+    Idle -->|"Raw BLOCKED with an open turn"| Blocked["BLOCKED_AMBIGUOUS"]
+    Lost -->|"Strict guarded repair"| Repaired["DONE with reconciled completion"]
+    Lost -->|"Repair refused"| LeadDecision["Lead decision required"]
+```
+
+*Figure 1. The member lifecycle and the main health exits. Pane-based states never authorise an
+automatic retry of the task.*
+
+## 4. State model
+
+### 4.1 Normal and transitional member states
+
+| State | Exact evidence | Meaning and certainty |
+|---|---|---|
+| `STARTING` | Session is `SPAWNING`; MCP presence is absent | Normal inside the startup grace. MCP contact is the readiness signal. |
+| `IDLE` | Session is `READY` or `DONE`; live status is `IDLE` or `DONE`; no open turn or inbox item exists | Normal. Idle is not a fault. |
+| `WORKING` | Session is `BUSY`; raw live status is `WORKING`; the accepted turn is open | Certain that herdr sees work. It does not prove useful progress. |
+| `WORK_PENDING` | Queued delivery or inbox content exists while the target is injectable | Transitional. Existing injector or push logic should move it. |
+| `BLOCKED_AMBIGUOUS` | An open turn exists and raw live status is `BLOCKED` | The bridge cannot tell whether this is permission, input, or a settled screen. |
+
+Idle may drive configured resource cleanup. It never opens an incident and never pages a person.
+
+### 4.2 Member fault and quality states
+
+| State | Exact evidence | Certainty and action |
+|---|---|---|
+| `NEVER_READY` | `SPAWNING`, no MCP presence, and an accepted delivery waits through the existing readiness grace | Delivery never became possible. The exact cause is unknown. Fail the send, stop the process, and preserve a provisioned worktree. |
+| `GONE` | Per-target herdr call returns `_not_found` while fleet list or ping works | Certain target loss. Fail all target work. Do not replay it. |
+| `TURN_BOUNDARY_LOST` | Same session turn stays `BUSY`; same accepted task stays open; two raw snapshots show `IDLE` or `DONE` | Strong disagreement. Strict reconciliation may repair it. |
+| `ERROR_ON_SCREEN` | Suspicious non-working state survives grace; `detection` matches a tested adapter-specific fatal signature | Certain only for the matched signature. A bare word such as `Exception` is not enough. |
+| `STALL_SUSPECTED` | Open turn is older than the configured threshold; two normalised `recent_unwrapped` digests are unchanged; no boundary or reply occurs | Not certain. A long valid API call can look the same. Lead decides. |
+| `MUTE` | Turn resolves through completion fallback instead of `bridge_reply` | Certain that no structured reply won. It does not prove an MCP failure. A single event is a metric, not an incident. |
+| `REPLY_STRANDED` | Typed reply or health message remains after owning-lead push reaches its cap | Collection failed. This does not explain whether the lead is busy, dead, or ignoring the nudge. |
+| `DELEGATION_ORPHANED` | Target is gone, failed, or released, but one or more tasks remain `PENDING` after reconciliation grace | Certain bridge invariant failure. This is not an inbox-drain fault. |
+| `WORK_PRODUCT_AT_RISK` | Provisioned branch has commits after its recorded base; member is `DONE`, `FAILED`, or preserved after release; no turn or inbox item remains; long-idle threshold passed | A warning, not proof of loss. Work may already have an open pull request or a squash merge. |
+
+`MUTE` opens an incident only after a small fixed rate threshold for one target or profile, or when
+it appears with another fault.
+
+`WORK_PRODUCT_AT_RISK` must not become `WORK_PRODUCT_UNCOLLECTED`. The bridge does not know pull
+request or merge state. If committed work appears with `REPLY_STRANDED` or
+`DELEGATION_ORPHANED`, the existing incident gains `committedWorkAtRisk: true`.
+
+### 4.3 Control-link state
+
+| State | Exact evidence | Certainty and action |
+|---|---|---|
+| `CONTROL_LINK_DOWN` | Two full-fleet `agent.list` calls fail across the grace, and herdr `ping` also fails | Certain for the bridged-to-herdr link. Retry calls, record the incident, and use human escalation if no lead can be reached. |
+
+A failed fleet list alone is not a dead-member claim. A single `_not_found` with a healthy global
+link is a target fault, not a control-link fault.
+
+### 4.4 Lead states
+
+| State | Exact evidence | Meaning and action |
+|---|---|---|
+| `LEAD_IDLE` | Expected lead is present with raw injectable status; no actionable state waits | Normal. Existing heartbeat may run under its own policy. |
+| `LEAD_WORKING` | Expected lead is present with raw `WORKING`; stall threshold is not met | Reachable and busy. Never inject into the live turn. |
+| `LEAD_STATUS_UNKNOWN` | Expected lead is present with raw `UNKNOWN` | Neither dead nor a healthy routing target. Retain evidence and retry. |
+| `LEAD_UNREACHABLE` | Expected lead is absent from two successful live-agent snapshots while ping works, or targeted lookup returns `_not_found` with a healthy control link | Route to a healthy peer. If none exists, use human escalation. |
+| `LEAD_UNRESPONSIVE` | Actionable state waits; lead stays injectable; bounded nudges exhaust; inbox remains uncollected | Route to a healthy peer or a person. |
+| `LEAD_STALL_SUSPECTED` | Lead stays `WORKING` past threshold; two sparse pane probes show no progress | Not certain. Never kill or restart automatically. Route to peer or person. |
+
+The monitor retains the lead name and terminal, last successful sighting, raw status and age,
+consecutive list absences, targeted errors, pane-probe facts, pending incident age, and nudge
+outcomes. Current heartbeat and push loops discard much of this history.
+
+Expected lead identity comes from the same supplier used by `CallerResolver`. It is not liveness
+evidence. `LeadTabScanner` keeps cached identity after a failed scan, so the health monitor compares
+that identity with a fresh successful agent list. A dynamic identity also survives a two-successful-
+snapshot retirement grace. This stops a dead lead from escaping health by disappearing from one map.
+
+### 4.5 Evidence limits
+
+M4 cannot tell these cases apart with current evidence:
+
+- A valid long call and a hung call may have the same status and pane digest.
+- `BLOCKED` does not explain which input is needed.
+- An idle prompt after failure may look like an idle prompt after success.
+- A missing structured reply does not prove a broken MCP connection.
+- An undrained inbox does not explain why the lead did not collect it.
+- Arbitrary pane text cannot safely classify arbitrary exceptions.
+- A branch ahead of its base does not prove that work was not collected.
+
+Logs are outputs, not classifier inputs. The monitor never parses its own logs.
+
+## 5. Automatic action and lead action
+
+### 5.1 Actions the bridge may take
+
+The bridge may:
+
+- retry transient herdr status, list, ping, and pane-read failures with bounded backoff;
+- re-submit Enter after the existing paste/submit race;
+- fail queued delivery after `NEVER_READY`;
+- stop a never-ready process while preserving its provisioned worktree;
+- fail all queued, accepted, and async tasks for a gone or released target;
+- reconcile one lost boundary when every strict gate in Section 8 passes;
+- hold typed messages, nudge the owning lead, and stop at the configured cap;
+- use the existing bounded idle-lead heartbeat;
+- deduplicate, route, update, and resolve incidents.
+
+These actions do not choose new work and do not replay old work.
+
+### 5.2 Decisions reserved for the lead
+
+Only the lead may:
+
+- stop or continue `BLOCKED_AMBIGUOUS`;
+- stop, inspect, or wait on `ERROR_ON_SCREEN`;
+- kill or continue `STALL_SUSPECTED`;
+- spawn a replacement or reassign work;
+- retry a delivered task;
+- choose how to use partial work in a worktree;
+- restart herdr or change network, model, credentials, backend, or configuration.
+
+Reports include literal safe tool calls such as `bridge_status(sessionId="...")`,
+`bridge_poll(ticket="...")`, `bridge_list()`, and optional `bridge_stop(paneId="...")`. A judgement
+state never presents stop as the only action.
+
+### 5.3 Release causes and worktree safety
+
+| Release cause | Process action | Provisioned worktree |
+|---|---|---|
+| `SPAWN_ROLLBACK` before registration or delivery | Stop and clean up | Remove |
+| `COMPLETED` for `READY` or `DONE` without pending work, idle TTL, or successful context-cap completion | Stop | Remove only if clean; preserve a dirty worktree (CB-576) |
+| `NEVER_READY` | Stop | Preserve |
+| `GONE` | Best-effort stop | Preserve |
+| `TURN_FAILED` or lead abort while `BUSY` or `FAILED` | Stop | Preserve |
+| `RELEASE_WITH_PENDING_TASKS` | Stop | Preserve |
+| `SHUTDOWN` | Stop | Preserve |
+
+Explicit stop is state-aware. `SPAWNING`, `BUSY`, `FAILED`, or any target with pending tasks uses a
+preserving cause.
+
+Before abnormal release removes the live session, M4 writes an atomic manifest under the worktree
+root. It records session identity, owner, role, profile, repository, path, branch, base commit,
+release cause, release time, state, and pending task ids. `bridge_list.preservedWorktrees` loads these
+manifests after restart. Stop output and WARN logs also name the path and cause. M4 never
+auto-deletes a preserved worktree.
+
+## 6. Fleet health monitor
+
+Add `FleetHealthMonitor`. Do not widen `StatusPoller` into a policy loop.
+
+`StatusPoller` has a 250 ms delivery cadence and samples only injector targets with outstanding
+work. Health needs all sessions, all leads, task state, inbox age, and global control evidence. One
+loop cannot serve both cadences safely.
+
+Build the monitor like `LeadHeartbeatLoop`:
+
+- pure `decide(snapshot, priorState, now)` logic;
+- a thin scheduler;
+- an injected clock;
+- edge-triggered state changes;
+- no network work in the pure function;
+- no sleeping in tests.
+
+Each enabled fleet tick reads:
+
+- one `AgentControl.list()` result for the whole fleet;
+- one in-memory `SessionManager.roster()` snapshot;
+- accepted turns and async task state;
+- typed inbox depth, kind, and age;
+- push and heartbeat outcomes;
+- configured and discovered leads.
+
+Existing failure paths publish structured evidence to the monitor. The monitor does not infer events
+from log text.
+
+### 6.1 Pane budget
+
+Healthy idle members, recent working members, and quiet leads cause no pane reads.
+
+A pane is eligible only for a stable lost boundary, sustained `BLOCKED` or `UNKNOWN`, work older
+than the suspect threshold, or one final evidence read for a confirmed fault when the pane exists.
+
+Compiled brakes apply even if config asks for more:
+
+- per-target pane cooldown is at least 60 seconds;
+- working age before the first progress probe is at least 300 seconds;
+- at most two pane reads occur in one fleet tick;
+- targets rotate fairly;
+- only a normalised digest and optional clipped local excerpt are stored;
+- no pane excerpt leaves bridged in a human webhook.
+
+Use `detection` for tested screen signatures. Use normalised `recent_unwrapped` only for progress
+comparison.
+
+## 7. Typed inbox and routing
+
+### 7.1 Semantic record
+
+The typed inbox record carries:
+
+```text
+schemaVersion
+kind: reply | health
+msgId, target, subjectTerminal, recipientLead
+severity, state, evidence
+createdAtEpochMillis, firstSeenEpochMillis, lastSeenEpochMillis
+recoveryTried, suggestedToolCalls, content
+```
+
+A health message never calls `Rendezvous.resolve`. It cannot look like the member's task result.
+
+Both inbox adapters share field preservation, first-id-wins dedup, FIFO among decoded messages,
+explicit ownership, ack, and release rules. The in-memory adapter stores typed records directly. It
+does not copy AMQP migration logic.
+
+### 7.2 AMQP migration
+
+The reader uses AMQP `content_type`, never body sniffing:
+
+```text
+Legacy v0: text/plain
+Typed family: application/vnd.ltms.bridged.inbox-message+json
+```
+
+A legacy reply may begin with `{`. It remains plain text because its media type is `text/plain`.
+Legacy text becomes `kind=reply` with exact UTF-8 content and absent typed metadata.
+
+Typed JSON has required integer `schemaVersion: 1`. Version 1 ignores unknown optional fields.
+Missing required fields, invalid enums, malformed UTF-8 or JSON, and property/body identity mismatch
+are invalid data.
+
+An unknown schema version is not partly decoded. It remains unacknowledged on the original queue and
+creates one operator-visible `unsupported_version` failure. A newer daemon may read it later.
+
+Invalid known-format data is copied byte-for-byte to durable queue
+`agent.<target>.inbox.quarantine`. A dedicated confirm-mode publisher confirms the persistent copy
+before the original is acknowledged. A failed quarantine handoff leaves the original unacknowledged.
+The raw body never enters logs.
+
+Decode failure creates a redacted WARN, metric, `bridge_list` summary, and routed health incident.
+One bad entry never escapes the consumer callback and never stops later valid messages.
+
+Safe downgrade is not supported. The previous build ignores `content_type` and would show typed JSON
+as ordinary reply text. If drained, it would acknowledge the message and lose typed meaning. Typed
+queues must be drained or preserved before an old jar runs.
+
+The existing contract suite uses RabbitMQ. Production uses LavinMQ. The migration and lead-key
+ownership cases must run once against production LavinMQ before release, or the release must state
+that LavinMQ was not checked.
+
+### 7.3 Member routing
+
+A member incident first goes to the exact lead that owns its accepted delegation.
+`PrimaryRegistry` needs a no-fallback `delegatingLeadFor(memberTarget)` query. Health routing must not
+use the old singular-primary fallback when several leads exist.
+
+Publish the incident under the affected member target. Trigger the existing bounded push route. The
+push waits until the owning lead is injectable, so it does not interrupt a live lead turn.
+
+### 7.4 Peer lead routing
+
+Figure 2 shows the route from incident to lead, peer, or person.
+
+```mermaid
+flowchart TD
+    Incident["Open incident"] --> Member{"Member incident?"}
+    Member -->|"yes"| Known{"Exact delegation owner known?"}
+    Known -->|"no"| Sink{"Human webhook enabled and healthy?"}
+    Known -->|"yes"| Owner{"Owner lead healthy?"}
+    Owner -->|"yes"| OwnerInbox["Publish to owner lead path"]
+    Owner -->|"no"| PeerSet["Build healthy peer candidate set"]
+    Member -->|"no, lead incident"| PeerSet
+    PeerSet --> Peer{"Healthy peer exists?"}
+    Peer -->|"yes"| Select["Choose fewest assigned incidents<br/>then stable name and terminal id"]
+    Select --> PeerInbox["Publish to peer lead inbox<br/>and status-gated push"]
+    Peer -->|"no"| Sink
+    Sink -->|"yes"| Webhook["Send classified outbound incident"]
+    Sink -->|"no"| Passive["Keep incident open<br/>show partial coverage on local surfaces"]
+```
+
+*Figure 2. Routing keeps delegation ownership separate from temporary peer fallback.*
+
+Peer candidates exclude the incident subject, failed owner, absent leads, raw-unknown leads, and
+leads with an open unhealthy state. A reachable `WORKING` peer may be selected; its push waits for an
+injectable window.
+
+Choose the candidate with the fewest assigned foreign incidents. Break ties by stable lead name,
+then terminal id. Pin the recipient. Reassign only if that peer becomes unhealthy or retires. A
+routing generation marks a reassignment, and old pending assignments become superseded.
+
+`bridge_list` lead rows show health, health age, assigned foreign incident count, and a bounded list
+of incident id, subject, state, severity, age, and routing generation. The top-level view also shows
+owner, recipient, and routing reason.
+
+A peer incident is published under the recipient lead's inbox key, not the failed subject's key. Its
+status-gated nudge names the failed lead and gives the exact
+`bridge_poll(target="<recipient-terminal>")` call.
+
+### 7.5 Lead inbox ownership
+
+Add `LeadInboxRegistry`, driven by the same expected-lead supplier as `CallerResolver`.
+
+It calls `replyInbox.own(leadTerminal)` at startup for configured leads, after successful discovery,
+after config adds a lead, and before publication. Ownership is not an authorization side effect.
+
+A missing lead keeps its key owned. Release happens only after confirmed retirement, all incidents
+are reassigned or resolved, typed health messages move or ack, and the queue is empty. Own a
+replacement terminal before moving messages from the old key. Never release a non-empty in-memory
+lead key, because in-memory release clears local data.
+
+### 7.6 Single-lead deployment
+
+One lead and no peer is a normal mode, not an edge case.
+
+An idle, reachable lead may receive the existing bounded nudge. An unreachable or stalled sole lead
+has no safe in-loop recovery. The bridge must not restart or replace it. A new lead would not have the
+failed lead's plan or context, and an uncertain relaunch could create two orchestrators.
+
+With no webhook, only `bridge_list`, `/healthz`, metrics, WARN logs, and the incident journal remain.
+These are passive surfaces. They are not a human notification.
+
+## 8. Lost-boundary reconciliation
+
+This is the only M4 path that reconstructs a result. It must prefer a visible stall over a fabricated
+reply.
+
+### 8.1 Why normal completion rules are not enough
+
+Current `CompletionResolver.resolve` has two fail-open rules. It resolves when the delivery baseline
+is missing. It also resolves an empty completion when the pane read fails. Those choices are valid
+after a trusted `WORKING -> IDLE` boundary because the bridge knows the turn ran. They are unsafe
+when health only guesses that a boundary was lost.
+
+M4 gives each accepted send an internal `TurnToken`. It ties target, exact waiter, session turn,
+delivery baseline, and task outcome together.
+
+### 8.2 Delivery baseline
+
+Capture the baseline immediately after prompt send and before the delivery future completes. Store:
+
+```text
+TurnToken
+exact waiter identity
+capture time and pane source
+normalised assistant block clipped to MAX_SCRAPE_CHARS
+whether a supported assistant marker was recognised
+capture result: PRESENT | READ_FAILED | UNRECOGNISED
+```
+
+A failed or missing baseline never authorises repair. A late baseline is not valid evidence. After a
+daemon restart, the old waiter, task, token, and baseline are gone, so the old turn cannot be
+repaired.
+
+Automatic repair is enabled only for agent kinds with tested assistant-block fixtures. Current
+extraction is Claude Code-specific and falls back to arbitrary raw text without `⏺`. That raw fallback
+cannot authorise repair. OpenCode repair stays disabled until live pane fixtures exist.
+
+### 8.3 Strict gates and resolver result
+
+Figure 3 shows the repair gates. Any failed gate keeps the waiter unchanged.
+
+The two raw snapshots must describe the same `TurnToken` and session turn. No `WORKING`,
+`BLOCKED`, `UNKNOWN`, missing-agent, reply, failure, or new-delivery observation may occur between
+them.
+
+```mermaid
+flowchart TD
+    Candidate["TURN_BOUNDARY_LOST candidate"] --> Stable{"Same TurnToken and BUSY turn<br/>across two raw IDLE or DONE snapshots?"}
+    Stable -->|"no"| Resnapshot["Take a fresh snapshot"]
+    Stable -->|"yes"| Waiter{"Exact captured waiter<br/>still open by identity?"}
+    Waiter -->|"no"| Stale["STALE_TURN or ALREADY_RESOLVED"]
+    Waiter -->|"yes"| Baseline{"Successful recognised<br/>delivery baseline exists?"}
+    Baseline -->|"no"| Refuse["Refuse repair<br/>leave ticket pending"]
+    Baseline -->|"yes"| Read{"Fresh pane read succeeds?"}
+    Read -->|"no"| Refuse
+    Read -->|"yes"| Output{"Recognised non-blank assistant block<br/>differs from clipped baseline?"}
+    Output -->|"no"| Refuse
+    Output -->|"yes"| Resolve["Shared CompletionResolver guard core<br/>resolves exact waiter"]
+    Resolve -->|"won race"| Repaired["RECONCILED_COMPLETION<br/>same turn becomes DONE"]
+    Resolve -->|"lost race"| Resnapshot
+```
+
+*Figure 3. Repair needs stronger evidence than a normal observed turn boundary.*
+
+Refactor the current resolver into one guard core with two policies:
+
+```text
+resolveCaptured(target, inFlight, OBSERVED_BOUNDARY)
+resolveCaptured(target, inFlight, LOST_BOUNDARY_REPAIR)
+```
+
+The health monitor calls only:
+
+```text
+CompletionResolver.reconcileLostBoundary(target, expectedTurnToken)
+```
+
+It returns `REPAIRED`, `ALREADY_RESOLVED`, `REFUSED_NO_CAPTURE`, `REFUSED_NO_BASELINE`,
+`REFUSED_UNREADABLE`, `REFUSED_UNCHANGED`, `REFUSED_AMBIGUOUS_OUTPUT`, `STALE_TURN`, or
+`RACE_LOST`.
+
+Only `REPAIRED` and same-turn `ALREADY_RESOLVED` may move that turn from `BUSY` to `DONE`. A
+per-target reconciliation gate stops a queued second send from being accepted between waiter
+resolution and the FSM transition.
+
+### 8.4 Lead-visible marker and refusal
+
+A repaired result uses distinct `RECONCILED_COMPLETION` values in `Rendezvous`, `MessageService`,
+task poll source, and metrics. The lead sees:
+
+```text
+[repaired completion - bridged detected a lost turn boundary. The member did not call
+bridge_reply; pane-derived text follows and may be partial]
+```
+
+Clipped text also keeps the existing clipped-tail marker.
+
+A refused repair leaves `TURN_BOUNDARY_LOST` open and the ticket pending. The report states that no
+reply was reconstructed and no task was replayed. `UNCHANGED`, `UNREADABLE`, and
+`AMBIGUOUS_OUTPUT` get at most one delayed retry for the same token. Missing capture or baseline gets
+no retry. After two refused scrapes, automatic repair stops for that token.
+
+### 8.5 Target-wide teardown invariant
+
+CB-568 owns the multi-ticket cancellation mechanism. M4 routes every terminal cause through that one
+idempotent operation and checks this independent invariant after teardown:
+
+- no injector entry exists for the target;
+- no accepted turn or completion record exists;
+- no rendezvous waiter or ask exists;
+- every async task is terminal or was already terminal;
+- no thread waiting for the target send lock can later accept it;
+- new sends fail immediately;
+- each old task has one terminal outcome and one metric count.
+
+A violation becomes `DELEGATION_ORPHANED`. The monitor may call the same idempotent target-wide
+failure operation once. It never recreates the task.
+
+## 9. Capacity and utilisation
+
+Capacity is a view, not a health state.
+
+`bridge_list` adds one block per profile:
+
+```text
+profile, maxLoad, live, free, reclaimable
+```
+
+For an unlimited profile, `maxLoad` and `free` are null. `free` is
+`max(0, maxLoad - live)` for a capped profile.
+
+The view must use the exact live-count function used by placement. A second calculation could show a
+free slot that placement then refuses. Member rows add `idleForSeconds` only when state is `READY` or
+`DONE`, no accepted turn exists, and the inbox is empty. `reclaimable` means only that the member
+holds capacity without open bridge work.
+
+The existing idle-lead nudge gains a bounded capacity summary. It lists per-profile live, cap, free,
+and reclaimable counts, plus at most three long-idle members. Capacity does not make
+`FleetState.hasPending()` true. A changed capacity fingerprint may re-arm one capped heartbeat
+sequence. The fingerprint excludes changing idle durations, so a static idle fleet cannot reset the
+cap forever. Reply-push stand-down remains first.
+
+The bridge must never:
+
+- spawn a member because a slot is free;
+- generate a task or acceptance criteria;
+- move queued work to another member or profile;
+- treat a free slot or idle member as an incident;
+- stop an idle member only to improve utilisation.
+
+The bridge knows capacity facts but has no work list. Only the lead has the plan, task context,
+side-effect history, and acceptance criteria.
+
+Capacity calculation is in memory and adds no pane reads. Work-product checks run on a terminal
+session edge, not every fleet tick.
+
+This capacity design adds no automatic stop. The accepted `NEVER_READY` cleanup can still stop a
+very slow startup after the existing grace, which is a known risk. Free capacity and long idle time
+never trigger that path.
+
+## 10. Human escalation and notification
+
+### 10.1 Escalation rule
+
+Notify a person only when no healthy lead can act:
+
+- `CONTROL_LINK_DOWN` survives grace;
+- a lead is unhealthy and no healthy peer can receive the incident;
+- a member incident has no known owning lead;
+- the only owning lead becomes unreachable, unresponsive, or stalled;
+- incident publication or routing itself fails.
+
+Do not page a person for a member fault while a healthy owning lead exists. An uncollected member
+incident feeds lead-health evidence. If the lead then becomes unhealthy, peer or human routing starts.
+
+### 10.2 Detection and notification switches
+
+`health.enabled` controls detection and bridge-local reporting. It does not require a webhook.
+
+`health.notifications.mode` is `disabled` or `webhook`. Disabled is valid and is the default.
+Webhook mode requires a resolved environment variable. Turning notification off stops outbound
+attempts but keeps incidents. Turning it back on resumes still-open human incidents.
+
+Without a sink, `bridge_list.healthCoverage` states that human escalation is unavailable. `/healthz`
+keeps its existing HTTP liveness result and adds a nested `fleetHealth.status=partial` component.
+Metrics and one startup or reload WARN expose the same limit.
+
+### 10.3 Incident and delivery deduplication
+
+One open incident uses this key:
+
+```text
+(scope, subjectStableId, state, causeFingerprint)
+```
+
+The cause fingerprint includes stable error codes, dependency names, signature ids, or invariant
+names. It excludes times, ages, retry counts, pane text, and changing digests. A later recurrence
+after resolution gets a new generation and incident id.
+
+Each outbound event uses:
+
+```text
+Idempotency-Key = hash(incidentId, eventType, eventRevision)
+```
+
+Event types are `open`, `severity_changed`, `reminder`, and `resolved`. Transport retries keep the
+same key.
+
+An atomic owner-only journal beside the active config stores open incidents, routing, delivered
+revisions, retry state, and resolution state. It stores no pane or task content. Journal failure does
+not stop detection, but notification coverage becomes degraded.
+
+### 10.4 Retry, reminder, and resolve
+
+Send the first event immediately. Retry network errors, timeouts, HTTP 408, HTTP 429, and HTTP 5xx
+with full-jitter exponential backoff:
+
+```text
+base: 5 seconds
+factor: 3
+maximum delay: 15 minutes
+one outstanding attempt per event
+```
+
+Respect `Retry-After` up to 15 minutes. Other HTTP 4xx responses are permanent for that event until
+config changes or a person requests replay.
+
+Transport retry is not an incident reminder. `humanRepeatSeconds` creates a new reminder revision
+for an unresolved critical incident after the last successful human event. Disabled mode does not
+build an unbounded reminder queue.
+
+Send `resolved` only if at least one human event for that incident was delivered. If an incident
+resolves before its first successful delivery, cancel the pending open event and record local
+resolution.
+
+### 10.5 Outbound payload boundary
+
+An outbound payload may contain incident id and event type, severity, state, scope, stable bridge
+ids, role or profile, times, duration, structured evidence type and counts, recovery attempted,
+routing reason, coverage, and safe tool calls.
+
+It must never contain:
+
+- raw pane text, pane excerpts, or pane digests;
+- task briefs, prompts, or member reply content;
+- source files, diffs, or worktree file content;
+- worktree paths;
+- environment values, tokens, credentials, headers, or webhook URL;
+- raw exception messages or stack traces;
+- arbitrary model output.
+
+The sink response body is ignored. A webhook cannot direct recovery. n8n remains outbound-only.
+
+### 10.6 Metrics
+
+M4 adds bounded-label series:
+
+```text
+bridged_health_incidents{scope,state,severity}
+bridged_health_incidents_total{event}
+bridged_health_notifications_total{event,outcome}
+bridged_health_notification_queue_depth
+bridged_health_notification_last_success_seconds
+bridged_health_notification_capability{mode,status}
+bridged_lead_health{lead,state}
+bridged_lead_assigned_incidents{lead}
+```
+
+Metric labels never include terminal ids, incident ids, URLs, or error text.
+
+## 11. Configuration
+
+The optional `health:` block is absent or disabled by default. The dormant monitor scheduler does no
+herdr or pane work while disabled. Every listed key is hot because the monitor reads `ConfigRef` on
+each tick or notification.
+
+| Key | Class | Default and hard bound | Purpose |
+|---|---|---|---|
+| `health.enabled` | Hot | `false` | Enable detection and bridge-local reporting. |
+| `health.snapshotIntervalSeconds` | Hot | default 30, minimum 15 | Whole-fleet comparison cadence. |
+| `health.workingSuspectAfterSeconds` | Hot | default 600, minimum 300 | Age before working-pane probes. |
+| `health.paneProbeIntervalSeconds` | Hot | default 60, minimum 60 | Per-target pane cooldown. |
+| `health.leadUnresponsiveAfterSeconds` | Hot | default 300, minimum 120 | Delay after exhausted actionable nudges before lead fault. |
+| `health.humanRepeatSeconds` | Hot | default 3600, minimum 900 | Minimum repeat period for one open human incident. |
+| `health.capacityLongIdleAfterSeconds` | Hot | default 900, minimum 300 | Long-idle threshold for capacity summaries. |
+| `health.includePaneExcerpt` | Hot | `false` | Allow a clipped excerpt in local lead reports only. Human payloads still exclude it. |
+| `health.notifications.mode` | Hot | `disabled` | Select `disabled` or `webhook`. |
+| `health.notifications.webhookUrlEnv` | Hot | required in webhook mode | Name of the environment variable that holds the sink URL. |
+| `health.notifications.requestTimeoutMs` | Hot | default 10000, range 1000-30000 | Whole webhook request limit. |
+
+Two consecutive snapshots are compiled floors for lost boundary, lead disappearance, and control
+link failure. The two-pane-reads-per-tick limit is also compiled and cannot be weakened by config.
+
+## 12. Delivery units and acceptance
+
+### Unit 1 - Evidence model and fleet snapshot
+
+Scope: health state model, fleet join, clocks, evidence retention, and pane budget.
+
+Acceptance criteria:
+
+1. One `agent.list` call covers one enabled fleet tick.
+2. Pure decision tests cover every state and every evidence limit in Section 4.
+3. `BUSY` plus stable raw `DONE` opens `TURN_BOUNDARY_LOST` after two snapshots.
+4. Healthy fleet snapshots perform zero pane reads.
+5. Pane cooldown, two-read fleet budget, and fair rotation cannot be disabled by config.
+6. Logs are outputs only; no log parsing exists.
+7. Fleet snapshots expose the same profile live-count calculation that placement uses.
+8. Capacity rows report cap, live, free, and reclaimable values without opening incidents.
+
+### Unit 2 - Lost boundary and task reconciliation
+
+Scope: accepted-turn identity, guarded repair, target-wide teardown, release causes, and preserved
+worktree discovery.
+
+Acceptance criteria:
+
+1. Every accepted send receives a stable `TurnToken` tied to target, exact waiter, and delivery
+   baseline.
+
+   **Corrected during implementation (2026-08-15).** This criterion first also required the session
+   turn number and the task outcome. That is not implementable at this layer, and the implementer
+   refused it three times rather than fabricate a value — correctly. The reason is an ordering fact
+   that is invisible from any single class: `MessageService` owns acceptance and holds the waiter and
+   the async `Task`, but it learns nothing about delivery, because the delivery event goes to
+   `CompletionResolver` through `TurnListener.onDelivered`. And `CompletionResolver.onDelivered` runs
+   *before* `SessionManager.onDelivered`, so the session turn number does not exist yet at the only
+   point where the token could capture it.
+
+   Two ways out were rejected. A shared registry keyed by target reintroduces exactly the "whichever
+   send happens to be waiting" ambiguity the token exists to remove — the same weak claim
+   `Rendezvous.currentWaiter` warns about. Injecting a turn counter into `MessageService` adds a
+   required cross-layer dependency to populate a field that nothing in this slice reads, which is
+   speculative coupling across a boundary already shown to be fragile.
+
+   So the token identifies the **accepted send**, and `SessionManager` keeps verifying its own
+   delivery separately. Repair (criterion 2) does need the session turn; binding it means resolving
+   that acceptance-versus-delivery ordering first, and that work belongs to the repair unit, not
+   here. The token record carries a comment saying the field is deliberately absent.
+2. Repair requires the same `BUSY` token, two raw `IDLE` or `DONE` snapshots, no conflicting
+   observation, exact open waiter, successful baseline, and new recognised assistant output.
+3. Missing, failed, late, or post-restart baseline never authorises repair.
+4. Repair is enabled only for agent kinds with tested assistant-block extraction. Raw-text fallback
+   without a recognised marker refuses repair.
+5. Normal completion and repair use one resolver guard core. Waiter, scrape, clipping, unchanged, and
+   exact-turn guards are not duplicated.
+6. `reconcileLostBoundary` returns every typed result named in Section 8.3.
+7. Only `REPAIRED` and same-turn `ALREADY_RESOLVED` may move the same turn to `DONE`.
+8. A per-target reconciliation gate blocks a queued second send during repair and FSM update.
+9. Repaired completion has distinct rendezvous kind, message outcome, poll source, lead marker, and
+   metric. Clipping keeps its extra marker.
+10. Unchanged, unreadable, or ambiguous evidence gets at most one delayed retry. Missing capture or
+    baseline gets none.
+11. Refusal leaves the ticket pending and tells the lead that no result was rebuilt or replayed.
+12. Release, gone, never-ready, and abnormal stop use CB-568's one idempotent target-wide failure
+    operation.
+13. The post-teardown invariant in Section 8.5 is tested independently of CB-568 internals.
+14. A violated teardown invariant creates `DELEGATION_ORPHANED` and retries only the idempotent
+    failure operation.
+15. `SPAWN_ROLLBACK` and normal `COMPLETED` remove worktrees. Abnormal and shutdown causes preserve
+    them.
+16. Explicit stop is state-aware. Any pending task or non-terminal state preserves the worktree.
+17. Atomic preserved-worktree manifests reload after restart and appear in lead-only
+    `bridge_list.preservedWorktrees`.
+18. Manifest failure preserves the worktree and opens an operator-visible health failure.
+19. Provision records the base commit. Terminal, long-idle worktrees report
+    `WORK_PRODUCT_AT_RISK` only under the evidence in Section 4.2 and never auto-delete work.
+20. No path replays a delivered task, rebuilds its brief, or retargets it, even when prompt text is
+    available.
+21. Tests cover both real traces, all repair refusals, clipping, explicit-reply and next-turn races,
+    restart without capture, concurrent send and release, and preserved discovery after restart.
+
+### Unit 3 - Typed inbox and member routing
+
+Scope: semantic record, AMQP migration, both adapters, member routing, polling, and member health in
+`bridge_list`.
+
+Acceptance criteria:
+
+1. AMQP selects legacy or typed decoding only from `content_type`; it never sniffs the body.
+2. Persistent `text/plain` from the old build becomes `kind=reply` with exact UTF-8 content,
+   including content beginning with `{`.
+3. New entries use the vendor media type, `schemaVersion: 1`, UTF-8, persistent delivery, and AMQP
+   message ids.
+4. Version 1 ignores unknown optional fields but rejects missing fields and identity mismatch.
+5. Unknown versions are not decoded or acked. They remain on the original queue and create one
+   deduplicated failure.
+6. Invalid known data never escapes the callback, appears as a reply, or blocks later valid messages.
+7. Invalid data reaches durable per-target quarantine before original ack. Failed handoff leaves the
+   original unacked.
+8. Decode failures create redacted WARN, metric, `bridge_list` summary, and routed incident without
+   raw content.
+9. Both adapters pass one semantic contract for fields, FIFO, dedup, ownership, ack, and release.
+10. Lead keys require explicit ownership. Publication never claims a queue.
+11. Unit codec tests cover legacy `{`, Unicode, malformed UTF-8, typed round trip, additive fields,
+    malformed JSON, missing fields, identity mismatch, media type, version, and dedup.
+12. A live broker contract writes old wire data and reads it with the new adapter after reconnect.
+13. Live contract tests cover mixed entries, quarantine confirm-before-ack, unsupported redelivery,
+    later progress past poison, property persistence, lead ownership, and ack removal.
+14. Safe downgrade is documented as unsupported.
+15. RabbitMQ contract tests pass with `mvn test -Pcontract`. The same cases run once on production
+    LavinMQ, or the release states that LavinMQ was not checked.
+16. Member incidents route to the exact delegating lead and never resolve a task rendezvous.
+17. `bridge_list` shows compact member health and capacity without pane content. Member
+    `idleForSeconds` is present only when no accepted turn or inbox item exists.
+
+### Unit 4 - Lead health and peer routing
+
+Scope: lead evidence, exact ownership, peer selection, explicit-recipient push, and lead inbox
+lifecycle.
+
+Acceptance criteria:
+
+1. Lead identity uses the `CallerResolver` supplier. Liveness uses successful current agent data.
+2. Two successful-list absences with healthy ping become `LEAD_UNREACHABLE`; global link failure does
+   not.
+3. Raw `WORKING`, raw `UNKNOWN`, first-seen time, failures, last success, and error class persist
+   across ticks.
+4. Heartbeat and push publish status and nudge outcomes before safe no-injection decisions.
+5. Dynamic lead identity survives a two-successful-snapshot retirement grace.
+6. Member incidents first use exact delegation ownership with no singular-primary fallback.
+7. Peer selection follows the exclusions, load rule, and stable tie break in Section 7.4.
+8. A selected working peer is not interrupted. Its push waits for an injectable window.
+9. Recipient assignment stays pinned. Reassignment increments generation and supersedes old pending
+   assignment.
+10. `bridge_list` shows bounded foreign assignments, recipient, reason, and generation without pane
+    content.
+11. `LeadInboxRegistry` owns configured and discovered lead keys before publication.
+12. Missing leads keep ownership. Retirement needs an empty queue and handled incidents.
+13. Replacement owns the new key before messages move. Non-empty in-memory keys are not released.
+14. Tests cover dead versus busy, unknown, global failure, stale scan cache, disappearance, one peer,
+    several peers, reassignment, and no peer.
+15. Adapter tests cover lead ownership, restart re-ownership, retirement, and terminal replacement.
+    LavinMQ is checked or named as unchecked.
+16. A sole unreachable or stalled lead is never restarted or replaced. Without a sink, only passive
+    evidence remains and every coverage surface says so.
+
+### Unit 5 - Human sink, hot config, metrics, and operator coverage
+
+Scope: generic webhook, config split, incident journal, retry, resolve, metrics, example config, and
+operator documentation.
+
+Acceptance criteria:
+
+1. `health.enabled` works without a human sink.
+2. Notification mode is hot, defaults to disabled, and supports disabled or webhook.
+3. Webhook mode requires a resolved environment value. Bad notification config does not disable an
+   already valid detector.
+4. Mode changes keep open incidents. Re-enable resumes eligible incidents.
+5. `bridge_list`, `/healthz`, metrics, and one WARN show partial coverage without a sink. HTTP
+   liveness behavior stays unchanged.
+6. One-lead, no-sink coverage states that lead failure has no active notification or recovery.
+7. Incident and outbound dedupe use the stable keys in Section 10.3.
+8. The owner-only local journal survives restart and contains no pane or task content.
+9. Journal failure keeps detection running but marks notification coverage degraded.
+10. Retry tests cover network failure, timeout, 408, 429, `Retry-After`, 5xx, permanent 4xx, jitter,
+    delay cap, config re-arm, and one outstanding attempt.
+11. Reminders and transport retries remain separate. Disabled mode does not build an unbounded queue.
+12. Resolve sends only after an earlier human event succeeded. Resolve-before-delivery cancels stale
+    open delivery.
+13. Metrics use bounded labels and exclude ids, URLs, and error text.
+14. Payload tests reject every content type forbidden in Section 10.5.
+15. Webhook response bodies are ignored and cannot direct recovery.
+16. Tests cover disabled mode, one lead without sink, open/update/reminder/resolve, restart, dedup,
+    reassignment, disable/re-enable, and sink failure while local health continues.
+17. `bridged.example.yaml` documents all hot keys and compiled floors.
+18. The operator Features wiki is updated separately. The portable `CLAUDE.md` block is checked and
+    changed only if shipped tool or inbox semantics make it untrue.
+19. `mvn clean install` passes.
+
+## 13. Not checked and release gates
+
+These limits are part of the design, not optional follow-up notes.
+
+- **OpenCode pane status and assistant markers were not checked.** OpenCode lost-boundary repair is
+  disabled until live fixtures exist.
+- **Permission-prompt status was not checked** for Claude Code or OpenCode. `BLOCKED` remains
+  ambiguous and has no automatic action.
+- **`recent_unwrapped` stability was not checked** across all supported agent kinds. If normalisation
+  is not stable, `STALL_SUSPECTED` must say its evidence is weaker.
+- **The real `BUSY + DONE` trace was not replayed against live herdr.** The design uses the observed
+  production trace and current poller behavior.
+- **CB-568 was not present when Unit 2 was designed.** Unit 2 must inspect the landed API and keep its
+  independent teardown invariant.
+- **Production LavinMQ was not checked.** Existing durable-inbox contracts use RabbitMQ. Migration,
+  quarantine, redelivery, lead ownership, and reassignment must run on LavinMQ before release or be
+  recorded as unchecked.
+- **Live multi-lead routing was not checked.** Peer choice and reassignment are design rules backed by
+  fake-clock and adapter tests until a live exercise runs.
+- **A live sole-lead failure with a webhook was not checked.** The no-peer path is a design result,
+  not a tested recovery.
+- **No n8n, Slack, PagerDuty, or other receiver was checked.** The webhook remains generic and
+  outbound-only.
+- **Deployment supervisor behavior for nested `/healthz` fields was not checked.** HTTP liveness
+  status stays unchanged to reduce this risk.
+- **Incident-journal crash behavior was not checked** because the journal does not exist yet. Unit 5
+  must test atomic replacement and restart recovery.
+- **Worktree merge state cannot be checked reliably** without forge or explicit collection evidence.
+  `WORK_PRODUCT_AT_RISK` stays a warning.
+
+## 14. Locked exclusions
+
+M4 does not expose `agent.read` as a bridge tool. It does not add a workflow engine, inbound n8n
+authority, automatic task assignment, task replay, automatic lead replacement, or automatic member
+spawn for free capacity.
+
+The bridge remains a message bus with evidence and bounded mechanical repair. The lead remains the
+place where judgement and work planning happen.