fleetd#330: split reload class for health/coordinator + top-level coverage

Unit 1: ConfigRef gets a fourth reload class, `split`, for keys read both
off the startup snapshot and live off config.get() at different sites
(health:, coordinator:). A split change is accepted (Outcome.applied()
stays true) and reported by name, naming which half is live and which
needs a restart, via a new Outcome.split() field kept separate from
deferred() since the two carry different guarantees for any caller that
branches on them, not just prose in summary(). Class doc updated: four
classes now, denominator note no longer calls health/coordinator
undecided.

Unit 2: ConfigRefTopLevelCoverageTest enumerates FleetConfig's 22
top-level record components and requires each to sit in exactly one of
COLD_KEYS, a pinned "compared in changedDeferredKeys" set, SPLIT_KEYS, or
a pinned hot-exclusion escape hatch — printing its own denominator and
pinning the escape hatch's exact contents the way #323 asked for.

Deviates from the issue's starting values by one key: `profiles` moves
from the suggested Hot bucket into the deferred bucket, because
changedDeferredKeys demonstrably compares it (add/remove and launch
settings), and citing "read live off the config supplier" for the whole
key would be false — most Profile fields are not read live, only
weight/maxLoad/credentialId are (and those are already covered by
ConfigRefProfileCoverageTest). Cold=5, split=2, deferred=11, hot=4,
total=22 — verified against the record and against ConfigRef's code, not
copied from the issue.

.claude/skills/hunter/SKILL.md

@@ -0,0 +1,102 @@
+---
+name: hunter
+description: Defect-hunt procedure for a fleetd worker — sweep an assigned package for real bugs and report several ranked findings without fixing anything. Load this when the lead asks you to hunt or audit a scope rather than review one diff. Do NOT load `reviewer` for this; the two want different output.
+---
+
+# Hunter worker — procedure
+
+The turn contract (one `fleet_reply`, `fleet_ask` for the lead's decisions, honest reporting,
+never merge) is in **`CLAUDE.md` → Bridge communication → Worker** and already applies.
+
+**This skill is not `reviewer`.** `reviewer` judges one diff and reports the *single* most
+important issue in about 90 words. A hunt sweeps a whole package and reports *several* findings
+in a long structured form. Loading both gives you two contradictory output contracts, and the
+usual result is a worker that writes a good report into its terminal and ends the turn without
+sending it. Load exactly one.
+
+## 0. Read this before you read code: how the report gets home
+
+Your terminal reaches nobody. The lead sees **only** the text inside your `fleet_reply` call.
+
+A long report is exactly the case where this goes wrong, so plan for it:
+
+- **Write the report into the `fleet_reply` argument itself.** Do not compose it in your terminal
+  and then summarise it into the call.
+- If the report is long, **send it anyway** — one `fleet_reply` with everything.
+- If you end the turn without replying, the bridge scrapes your pane instead. That scrape carries
+  at most the last 4000 characters, and on a hunt it usually captures the tail of the lead's own
+  brief rather than your findings. The lead then has nothing and has to ask you again.
+
+## 1. Change nothing
+
+A hunt is read-only. Do not edit a production file, do not "quickly fix" what you find, and do
+not run a formatter. You may run the build and tests to *check* a claim, and you should say so
+when you did.
+
+## 2. Read the whole scope first
+
+Read every file in the assigned package before you judge any of it. A defect that a caller
+elsewhere in the same package makes unreachable is not a defect, and you cannot know that from
+one file.
+
+Stay inside the scope. If a defect there depends on a class outside it, read that class to
+confirm — but the defect itself must live in the scope you were given.
+
+## 3. The bar — this matters more than the count
+
+**Name the path into the bad state.** Say which caller, in which state, reaches it. A defect on
+paper is not a reachable defect. If you cannot name that path, keep the finding but mark it
+`unproven` and say exactly what you could not check. Do not drop it, and do not dress it up.
+
+**Say which direction the harm goes.** Data loss, privilege escalation and silent wrong answers
+are worth reporting even when the window is narrow. A finding whose worst outcome is a worse log
+line is not worth a block.
+
+Two workers once ran the same scope: the one that applied the direction-of-harm filter found ten
+real defects, the one that did not found none. Fewer findings the lead can act on beat many the
+lead has to triage.
+
+## 4. Shapes that have produced real merged fixes here
+
+Read for these first:
+
+1. **A one-way gate.** A guard added after an incident closes only the direction that incident
+   came from. Do not only ask what closes the gate — ask **which states still open it**.
+2. **A value read once, then used later to authorise something destructive**, after something
+   else has had a chance to change it.
+3. **A failure downgraded to a value that looks like a legitimate result** — `-1`, `null`, an
+   empty list, `false` — which a caller then trusts.
+4. **A lock held for one half of a read-modify-write and not the other**, or two collections
+   updated under different locks.
+5. **A comment or javadoc stating an invariant the code no longer keeps.** Comments are
+   load-bearing in this repo; a stale one has already caused a bug.
+
+## 5. What you cannot check, and must not claim you did
+
+- `fleetd/fleetd.yaml` is gitignored and **absent from your worktree**. You cannot read it. If a
+  finding depends on live configuration, name the key and say you could not check it.
+- `.mcp.json`, `opencode.json` and `.autoenv` in your worktree are neutralised stubs, not the
+  repo's real files.
+- The `wiki/` submodule pointer is months old. Do not cite it.
+
+Reporting a fact you took from the lead's brief as something you measured yourself is a false
+report, even when the fact is correct. Say where each fact came from.
+
+## 6. The report — what goes in `fleet_reply`
+
+One block per finding, most severe first:
+
+```
+FINDING N — <one line>
+file:line
+Path in:        <which caller, in which state, reaches this>
+Direction:      <data loss | escalation | silent wrong answer | outage | ...>
+Window/trigger: <when it actually happens>
+Confidence:     <confirmed by reading | unproven — say what you could not check>
+Why nothing else catches it: <the guard or test you checked, and why it misses>
+```
+
+End with one line naming every file you read, so the lead knows the denominator.
+
+**Nothing clears the bar?** Reply `NO FINDINGS`, name the files you read, and say what you ruled
+out. A clean sweep is a valid result; an invented defect is worse than none.

.claude/skills/reviewer/SKILL.md

@@ -10,6 +10,11 @@ never merge) is in **`CLAUDE.md` → Bridge communication → Worker** and alrea
 skill is only the *review procedure*: how to work the scope, and the exact shape of what you
 send back.
 
+**Wrong skill for a sweep.** This one reviews *one* diff or scope and reports the *single* most
+important issue. If the lead asked you to hunt or audit a whole package for several defects, load
+`hunter` instead and ignore this file — the two want different output, and following both is how a
+worker ends its turn with a good report that never gets sent.
+
 ## 1. Read the whole scope before you judge
 
 The delegation names your scope — a file, a diff, a PR, a function. **Read all of it first.**

CLAUDE.md

@@ -200,8 +200,13 @@ must obey belongs in the charter, not here.
   adapter, with a message naming the credential and the remaining seconds ("cooling off after
   repeated backend errors") — distinct wording from a quarantine refusal, so don't conflate the
   two when reading a spawn failure.
-- **Skills available to delegate:** `implementer` (worktree → commit → push → own PR) and
-  `reviewer` (scoped review → one structured finding). Name one in every delegation.
+- **Skills available to delegate:** `implementer` (worktree → commit → push → own PR),
+  `reviewer` (one diff → one structured finding) and `hunter` (sweep a package → several ranked
+  findings, change nothing). Name exactly one in every delegation. **`reviewer` and `hunter` are
+  not interchangeable** — `reviewer` caps the answer at one finding in about 90 words, so naming
+  it for a multi-finding sweep hands the worker two contradictory output contracts. That has
+  already cost three workers' turns: each wrote a good report to its terminal and ended the turn
+  with no `fleet_reply`, and the scrape returned the tail of the brief instead.
 - **Primary-side skills** (not delegation playbooks — a worker cannot use them):
   `port-to-opencode` (make an OpenCode session a participant in this workspace) and
   `fleets-status` (report every fleet that shares one LavinMQ instance).

fleetd/src/main/java/dev/ltms/fleet/auth/CallerResolver.java

@@ -236,7 +236,15 @@ public final class CallerResolver {
         // loopback-trust: same-host callers that are not workers are the primary. A non-loopback
         // caller is anonymous even here — and startup refuses that combination anyway
         // (FleetConfig.validateAuthExposure), so this is defence in depth, not the control.
-        return isLoopback(remoteAddr) ? Principal.primary(c.pid()) : Principal.anonymous();
+        //
+        // fleetd #317: "not a worker" must not be conflated with "identity unresolved". The real
+        // primary is a real process — its pid resolves (c.resolved()), it just owns no herdr pane.
+        // A caller whose peer-PID lookup failed (LsofPeerPidLookup's -1 sentinel — on any failure,
+        // silently including "lsof found no match") has no such pid, and PaneLocator's own javadoc
+        // already names what happens if that case is handed the primary role: a worker→primary
+        // escalation. So an unresolved caller is refused (ANONYMOUS — the same clean, already-tested
+        // "authenticated as nothing" outcome used everywhere else in this method), never promoted.
+        return isLoopback(remoteAddr) && c.resolved() ? Principal.primary(c.pid()) : Principal.anonymous();
     }
 
     private boolean presentedTokenMatches(String authorizationHeader) {
@@ -262,11 +270,15 @@ public final class CallerResolver {
         return token.isEmpty() ? null : token;
     }
 
+    /**
+     * fleetd #305: delegates to {@link ConnectionIdentity#isLoopback}. This used to be a second,
+     * independent copy of the same rule, and the two drifted: this one accepted all of
+     * {@code 127.0.0.0/8}, {@code ConnectionIdentity}'s accepted only {@code 127.0.0.1}. A caller
+     * from {@code 127.0.0.2} therefore had its identity skipped (so it had no terminal) and was
+     * then read as loopback here — which under loopback-trust is the primary. Sharing the inputs
+     * would not have prevented that; only sharing the computation does.
+     */
     private static boolean isLoopback(String remoteAddr) {
-        if (remoteAddr == null) {
-            return false;
-        }
-        return remoteAddr.equals("127.0.0.1") || remoteAddr.equals("::1")
-                || remoteAddr.equals("0:0:0:0:0:0:0:1") || remoteAddr.startsWith("127.");
+        return ConnectionIdentity.isLoopback(remoteAddr);
     }
 }

fleetd/src/main/java/dev/ltms/fleet/config/ConfigRef.java

@@ -22,7 +22,7 @@ import java.util.function.Supplier;
  * choice rather than an accident of where the field was initialised.
  *
  * <h2>Not every key can change under a running daemon</h2>
- * Keys fall into three classes, and the difference is about what already exists when the reload
+ * Keys fall into four classes, and the difference is about what already exists when the reload
  * happens — not about how important the key is.
  *
  * <ul>
@@ -39,29 +39,95 @@ import java.util.function.Supplier;
  *       keeps the old value until a restart: {@code lifecycle:}, {@code leadHeartbeat:},
  *       {@code spawnReadyTimeoutMs} / {@code spawnReadyPollMs}, {@code quarantineCooldownSeconds}
  *       (CB-578 stage B — baked once into the {@code BackendQuarantine} built at startup),
- *       {@code guard:}, {@code worktreeRoot:}, adding or removing a profile (a new backend needs its own launcher,
+ *       {@code guard:}, {@code worktreeRoot:} and {@code worktreeGroup:} (both baked once into the
+ *       {@code GitWorktrees} built at {@code Fleetd.java:251} and never rebuilt — fleetd #323
+ *       instance 2 found {@code worktreeGroup} missing from this list and from
+ *       {@link #changedDeferredKeys}), {@code primary:} (fleetd #326 — {@code Fleetd.java:506, 519,
+ *       520} read {@code cfg.primary()} only off the startup snapshot to build {@code
+ *       PrimaryRegistry} and size {@code ReplyPushLoop}'s reminder cap/backoff, and neither is
+ *       rebuilt on reload. Say the consequence exactly: {@code primary.terminal} is DEPRECATED
+ *       (CB-532, and {@code Fleetd.java:511} warns about it at startup) — a lead's identity comes
+ *       from {@code leaders:}/{@code leadScan:}, so changing this pin does not demote or promote a
+ *       lead that uses those. What a changed pin still does not take effect on until a restart is
+ *       the fallback nudge destination the pin remains, the deprecated identity path for an operator
+ *       who still relies on it, and {@code pushReminders}/{@code pushBackoffMs}), {@code configReload:} (fleetd #326 — {@code
+ *       Fleetd.java:679-680} read it only at startup to decide whether to build a {@code
+ *       ConfigWatcher} at all and with what interval; the watcher that would apply a later change is
+ *       itself built once, so a running watcher keeps polling on its original enabled flag and
+ *       interval regardless of what a reload changes it to, the same shape as {@code lifecycle} —
+ *       not cold, because no already-open resource goes inconsistent with the new value, the watcher
+ *       (if any) simply keeps its old settings), adding or removing a profile (a new backend needs its own launcher,
  *       which is constructed once), <em>and an existing profile's launch settings</em> —
  *       {@code model}, {@code baseUrl}, {@code argv}, {@code env}, {@code mcpUrl},
  *       {@code exhaustedPattern} (CB-578 stage A — compiled once into {@code Fleetd.main}'s
  *       pattern map at startup), {@code errorPattern} (fleetd #201 Unit 5 — compiled once into
- *       {@code Fleetd.main}'s backend-error pattern map at startup, the same way), and the rest.
+ *       {@code Fleetd.main}'s backend-error pattern map at startup, the same way),
+ *       {@code ideProjectDir} / {@code ideOpenCommand} / {@code autoCompactWindow} (fleetd #323
+ *       instance 1 — all three are read at spawn off the same frozen profile map and were missing
+ *       from {@link #sameLaunchSettings}), and the rest of {@link #sameLaunchSettings}.
  *       {@code credentialId} (CB-578 stage B) is NOT on
  *       this list — it is read live off the config supplier at every quarantine check and
  *       exhaustion event, exactly like {@code weight} / {@code maxLoad}, so it is hot instead.
  *       {@code HerdrPeerLauncher} takes {@code Map.copyOf(profiles)} at construction and resolves
  *       each spawn out of that copy, so those never reach a launch until the daemon restarts. A
  *       reload logs these rather than pretending they applied.</li>
+ *   <li><strong>Split</strong> (fleetd #330) — read <em>both</em> ways at different sites, so the
+ *       key does not fit any class above as a whole: {@code health:} and {@code coordinator:}.
+ *       Each is read off the startup snapshot to build a long-lived object, and read live off
+ *       {@link #get()} at a different, unrelated site — so half of a reload's effect already
+ *       applies while the other half waits for a restart, and a bare "config reloaded" would
+ *       under-claim by exactly that half.
+ *       <ul>
+ *         <li>{@code health:} — the monitor itself ({@code enabled}, {@code intervalSeconds},
+ *             {@code workingSuspectAfterSeconds}) is built once at {@code Fleetd.java:556-563}
+ *             and never rebuilt, so a changed value needs a restart to actually start, stop, or
+ *             retime it. The coverage string {@code fleet_profiles} reports
+ *             ({@code Fleetd.java:648-650}) is read live off {@link #get()} on every call, so it
+ *             already reflects the new value.</li>
+ *         <li>{@code coordinator:} — the {@code LeadMailbox} connection ({@code uri},
+ *             {@code uriEnv}, {@code selfId}, {@code prefetch}) is opened once at
+ *             {@code Fleetd.java:502} and never reopened, so a changed value needs a restart —
+ *             {@code selfId} in particular names this daemon's own AMQP inbox queue, and a peer
+ *             lead that learned the old name would not discover a new one on its own. The broker
+ *             URI env-var <em>name</em> that {@code MemberEnvAllowList} keeps out of a member's
+ *             environment is read live off {@link #get()} on every spawn
+ *             ({@code HerdrPeerLauncher.java:1530}), so it already applies.</li>
+ *       </ul>
+ *       A split change is still accepted — {@link Outcome#applied()} stays {@code true}, the same
+ *       as a deferred change — because the live half genuinely took effect; refusing the whole
+ *       reload would leave the operator worse off than today. {@link Outcome#split()} names the
+ *       key and says which half is which each time, rather than trying to score "how changed" a
+ *       mixed key is or handle "both halves changed in one reload" as a special case.</li>
  *   <li><strong>Cold</strong> — cannot change at all under a running daemon: {@code bind:},
  *       {@code herdrSocket:}, {@code broker:} and {@code auth:}. The socket is bound, the broker
  *       connection is open, and the auth mode decides who may reach the port that is already
  *       listening.</li>
  * </ul>
  *
+ * <p><strong>The denominator, measured on 2026-09-04 (fleetd #330).</strong> {@code FleetConfig} has
+ * 22 top-level record components. Two of them are named nowhere in this file, and the reason is the
+ * same for both: {@code memberCredentials} and {@code memberLoginShell} are <strong>hot</strong> and
+ * correctly absent — both are read live off {@code config.get()} at spawn time
+ * ({@code Fleetd.java:198, 205, 729} and {@code HerdrPeerLauncher#configuredMemberLoginShell}), so a
+ * reload takes effect on the next spawn with no entry needed here.
+ * {@code health} and {@code coordinator} used to be a third kind — <strong>undecided</strong>, not
+ * hot — until fleetd #330 added the <strong>split</strong> class above and gave them a home. A
+ * reload touching either used to report a bare "config reloaded", which under-claimed; now it names
+ * the key and says which half is which.
+ * <p>The point of writing the count down: "not mentioned in this file" looks identical for a key
+ * that is correctly hot and for a key nobody triaged. Twice now — {@code worktreeGroup} (#323) and
+ * {@code primary}/{@code configReload} (#326) — the second kind hid among the first. A top-level
+ * coverage checker in the {@link ConfigRefProfileCoverageTest} shape (one level up, over
+ * {@code FleetConfig} itself rather than {@code FleetConfig.Profile}) proves this file's four
+ * classes exhaust the record's components — see {@code ConfigRefTopLevelCoverageTest}.
+ *
  * <p><strong>A cold change refuses the whole reload.</strong> Not the hot half applied and the cold
  * half warned about: that would leave the running daemon in a state matching no file on disk, which
  * is the worst thing a reload can do to an operator debugging one. Refusing keeps the invariant that
  * the live config is always some version of the file, and the message names the keys that must
- * change through a restart.
+ * change through a restart. A split change does <em>not</em> refuse, for a different reason than a
+ * deferred change does not: its live half genuinely took effect, so refusing would throw that away
+ * and leave the operator worse off than the partial-but-honest report {@link Outcome#split()} gives.
  *
  * <p>A reload that fails to parse or fails validation is also refused, and the previous config keeps
  * running. A config file being edited is normally read once mid-save; degrading a working daemon
@@ -71,10 +137,24 @@ public final class ConfigRef implements Supplier<FleetConfig> {
 
     private static final Logger log = LoggerFactory.getLogger(ConfigRef.class);
 
-    /** Keys that cannot change under a running daemon — see the class doc. */
-    private static final Set<String> COLD_KEYS =
+    /**
+     * Keys that cannot change under a running daemon — see the class doc.
+     *
+     * <p>Package-private (not {@code private}) so {@code ConfigRefTopLevelCoverageTest} can fold it
+     * into the top-level triage it checks, the same way it reads {@link #SPLIT_KEYS}.
+     */
+    static final Set<String> COLD_KEYS =
              Set.of("bind", "herdrSocket", "memberHerdrSocket", "broker", "auth");
 
+    /**
+     * Keys read BOTH off the startup snapshot and live off {@link #get()} at different sites, so
+     * neither the hot, deferred nor cold class fits them as a whole — see the class doc's Split
+     * bullet (fleetd #330). A changed split key is accepted ({@link Outcome#applied()} stays
+     * {@code true}) and reported by name, with a message naming which half is live and which needs
+     * a restart.
+     */
+    static final Set<String> SPLIT_KEYS = Set.of("health", "coordinator");
+
     private final Path path;
     private final AtomicReference<FleetConfig> current;
 
@@ -102,25 +182,37 @@ public final class ConfigRef implements Supplier<FleetConfig> {
     /**
      * What a reload attempt did.
      *
+     * <p>{@code split} is a separate field from {@code deferred} rather than a differently-worded
+     * entry inside it, because the two carry different guarantees for any caller that branches on
+     * them rather than just printing {@link #summary()}: every {@code deferred} entry means "this
+     * key's whole change waits for a restart", while every {@code split} entry means "part of this
+     * key's change already applied, and the message says which part" — collapsing them would force
+     * a caller to re-parse the message to tell those apart. See the class doc's Split bullet
+     * (fleetd #330) for why the key needs this at all.
+     *
      * @param applied  true when the new config is now live
      * @param coldKeys cold keys whose value changed, which is why an unapplied reload was refused
-     * @param deferred keys that changed and were accepted, but whose effect waits for a restart
+     * @param deferred keys that changed and were accepted, but whose effect waits entirely on a
+     *                 restart
+     * @param split    split keys that changed and were accepted, each named with which half of it
+     *                 is already live and which half waits for a restart
      * @param error    the parse or validation failure that refused the reload, else {@code null}
      */
     public record Outcome(boolean applied, List<String> coldKeys, List<String> deferred,
-                          String error) {
+                          List<String> split, String error) {
 
         public Outcome {
             coldKeys = List.copyOf(coldKeys);
             deferred = List.copyOf(deferred);
+            split = List.copyOf(split);
         }
 
         static Outcome refusedCold(List<String> keys) {
-            return new Outcome(false, keys, List.of(), null);
+            return new Outcome(false, keys, List.of(), List.of(), null);
         }
 
         static Outcome failed(String error) {
-            return new Outcome(false, List.of(), List.of(), error);
+            return new Outcome(false, List.of(), List.of(), List.of(), error);
         }
 
         /** A one-line summary for the operator — the reason, not just the verdict. */
@@ -132,11 +224,18 @@ public final class ConfigRef implements Supplier<FleetConfig> {
                 return "config reload refused — these keys cannot change under a running daemon: "
                         + String.join(", ", coldKeys) + ". Restart fleetd to apply them.";
             }
-            if (!deferred.isEmpty()) {
-                return "config reloaded; these changes need a restart to take effect: "
-                        + String.join(", ", deferred);
+            if (deferred.isEmpty() && split.isEmpty()) {
+                return "config reloaded";
             }
-            return "config reloaded";
+            StringBuilder out = new StringBuilder("config reloaded");
+            if (!deferred.isEmpty()) {
+                out.append("; these changes need a restart to take effect: ")
+                        .append(String.join(", ", deferred));
+            }
+            if (!split.isEmpty()) {
+                out.append("; partially live — ").append(String.join(" | ", split));
+            }
+            return out.toString();
         }
     }
 
@@ -177,8 +276,9 @@ public final class ConfigRef implements Supplier<FleetConfig> {
         }
 
         List<String> deferred = changedDeferredKeys(old, fresh);
+        List<String> split = changedSplitKeys(old, fresh);
         current.set(fresh);
-        Outcome out = new Outcome(true, List.of(), deferred, null);
+        Outcome out = new Outcome(true, List.of(), deferred, split, null);
         log.info(out.summary());
         return out;
     }
@@ -221,6 +321,29 @@ public final class ConfigRef implements Supplier<FleetConfig> {
         if (!Objects.equals(old.worktreeRoot(), fresh.worktreeRoot())) {
             changed.add("worktreeRoot");
         }
+        // Baked into the same GitWorktrees as worktreeRoot (Fleetd.java:251) and never rebuilt
+        // either — see the class doc. Missing this check was fleetd #323 instance 2: a reload
+        // that changed only worktreeGroup reported "config reloaded" with nothing deferred, and
+        // newly provisioned worktrees kept the old sharing behaviour.
+        if (!Objects.equals(old.worktreeGroup(), fresh.worktreeGroup())) {
+            changed.add("worktreeGroup");
+        }
+        // fleetd #326: Fleetd.java:506, 519, 520 read cfg.primary() only off the startup snapshot
+        // (PrimaryRegistry's pinned terminal, ReplyPushLoop's reminder cap and backoff) — neither is
+        // rebuilt on reload, so a changed value needs a restart. Note what it does NOT mean:
+        // primary.terminal is deprecated (CB-532), identity comes from leaders:/leadScan:, so a lead
+        // using those is unaffected by this pin either way. See the class doc for the exact scope.
+        if (!Objects.equals(old.primary(), fresh.primary())) {
+            changed.add("primary");
+        }
+        // fleetd #326: Fleetd.java:679-680 read cfg.configReload() only at startup to decide whether
+        // to build a ConfigWatcher at all and with what interval — the watcher that would apply a
+        // later change is itself built once, so a running watcher keeps its original enabled flag and
+        // interval regardless of what a reload changes it to. Not cold: no already-open resource goes
+        // inconsistent with the new value, a watcher (if any) simply keeps polling on the old settings.
+        if (!Objects.equals(old.configReload(), fresh.configReload())) {
+            changed.add("configReload");
+        }
         if (!Objects.equals(old.spawnReadyTimeoutMs(), fresh.spawnReadyTimeoutMs())
                 || !Objects.equals(old.spawnReadyPollMs(), fresh.spawnReadyPollMs())) {
             changed.add("spawnReady*");
@@ -265,14 +388,60 @@ public final class ConfigRef implements Supplier<FleetConfig> {
     }
 
     /**
-     * Whether two versions of a profile would launch a peer identically. Compares every component
-     * the launcher reads at spawn; {@code weight}, {@code maxLoad} and {@code credentialId} are
-     * excluded because those are read live (by the placement policy and, for credentialId, by
-     * {@code CompositePeerLauncher}/the CB-578 stage B exhaustion sink) and really do take effect on
-     * the next spawn.
+     * Split keys whose value differs between the running config and the candidate — see the class
+     * doc's Split bullet (fleetd #330). Unlike {@link #changedDeferredKeys}, this does not try to
+     * tell which sub-field moved: any change to {@code health:} or {@code coordinator:} gets the
+     * same fixed message, because the message already names both halves every time, so there is no
+     * "which half changed" question left for the caller to answer.
      */
-    private static boolean sameLaunchSettings(FleetConfig.Profile a, FleetConfig.Profile b) {
-        return Objects.equals(a.baseUrl(), b.baseUrl())
+    private static List<String> changedSplitKeys(FleetConfig old, FleetConfig fresh) {
+        List<String> changed = new ArrayList<>();
+        if (!Objects.equals(old.health(), fresh.health())) {
+            changed.add("health: the monitor itself (enabled, interval, workingSuspectAfter) is "
+                    + "frozen at startup and needs a restart; the coverage status fleet_profiles "
+                    + "reports is read live and already applied");
+        }
+        if (!Objects.equals(old.coordinator(), fresh.coordinator())) {
+            changed.add("coordinator: the LeadMailbox connection (uri, uriEnv, selfId, prefetch) is "
+                    + "opened once and needs a restart; the broker URI env-var name kept out of a "
+                    + "member's environment is read live on every spawn and already applied");
+        }
+        // Kept in step with SPLIT_KEYS the same way changedColdKeys is kept in step with COLD_KEYS —
+        // every message here must be traceable to one of the two split keys the class doc documents.
+        assert changed.stream().allMatch(m -> SPLIT_KEYS.stream().anyMatch(k -> m.startsWith(k + ":")))
+                : "a split entry was reported that does not start with a SPLIT_KEYS name: " + changed;
+        return changed;
+    }
+
+    /**
+     * {@link FleetConfig.Profile} record components deliberately left out of
+     * {@link #sameLaunchSettings} because they are read <em>live</em>, not baked in at spawn — see
+     * the class doc's <em>Hot</em> bullet. {@code weight} and {@code maxLoad} are read live by the
+     * placement policy on every spawn; {@code credentialId} is read live by
+     * {@code CompositePeerLauncher} and the CB-578 stage B exhaustion sink. Nothing else is
+     * excluded — see {@code sameLaunchSettingsComparesEveryProfileComponentOrExcludesIt} in
+     * {@code ConfigRefProfileCoverageTest}, which enumerates every {@code Profile} record component
+     * by reflection and fails the build if one is neither compared below nor named here.
+     */
+    static final Set<String> LAUNCH_SETTINGS_EXCLUDED = Set.of("weight", "maxLoad", "credentialId");
+
+    /**
+     * Whether two versions of a profile would launch a peer identically.
+     *
+     * <p>This must compare every {@link FleetConfig.Profile} record component except the three in
+     * {@link #LAUNCH_SETTINGS_EXCLUDED}. That is not a claim this javadoc can make good on by
+     * itself — a javadoc saying "compares every component" is exactly what fleetd #323 found to be
+     * false for three fields (and a sibling method's field list, for a fourth). The actual
+     * guarantee comes from {@code ConfigRefProfileCoverageTest}: it enumerates every record
+     * component of {@code FleetConfig.Profile} by reflection, mutates each one not in
+     * {@code LAUNCH_SETTINGS_EXCLUDED} on a base profile, and asserts this method reports a
+     * difference — so a new component that is neither compared here nor added to
+     * {@code LAUNCH_SETTINGS_EXCLUDED} (with a reason) fails that test by name, rather than
+     * silently reporting "config reloaded" for a value the daemon never picked up.
+     */
+    static boolean sameLaunchSettings(FleetConfig.Profile a, FleetConfig.Profile b) {
+        return Objects.equals(a.profile(), b.profile())
+                && Objects.equals(a.baseUrl(), b.baseUrl())
                 && Objects.equals(a.model(), b.model())
                 && Objects.equals(a.configDir(), b.configDir())
                 && Objects.equals(a.tokenEnv(), b.tokenEnv())
@@ -298,6 +467,16 @@ public final class ConfigRef implements Supplier<FleetConfig> {
                 // fleetd #201 Unit 5: errorPattern is compiled once into Fleetd.main's backend-error
                 // pattern map at startup (see BackendErrorPatternLookup wiring), the same way
                 // exhaustedPattern is — a reload never re-reads it either.
-                && Objects.equals(a.errorPattern(), b.errorPattern());
+                && Objects.equals(a.errorPattern(), b.errorPattern())
+                // fleetd #323 instance 1: ideProjectDir and ideOpenCommand are read at spawn off the
+                // same frozen profile map as ideMcpUrl above (ClaudeCodeLauncher.java:267/269,
+                // OpenCodeLauncher.java:474/480/486) and were missing from this comparison.
+                && Objects.equals(a.ideProjectDir(), b.ideProjectDir())
+                && Objects.equals(a.ideOpenCommand(), b.ideOpenCommand())
+                // fleetd #323 instance 1: autoCompactWindow is read at spawn the same way
+                // (ClaudeCodeLauncher.java:926, OpenCodeLauncher.java:650). Comparing it here only
+                // makes the reload REPORT that a restart is needed — it deliberately does not make
+                // autoCompactWindow take effect live, which is a separate, larger change.
+                && Objects.equals(a.autoCompactWindow(), b.autoCompactWindow());
     }
 }

fleetd/src/main/java/dev/ltms/fleet/inject/Injector.java

@@ -157,6 +157,7 @@ public final class Injector {
         boolean awaitingCompletion; // a delivered message's turn is not yet known-complete
         boolean turnObserved;      // saw a real `working` sample since that delivery (turn ran)
         int unknownSinceTurn;      // consecutive `unknown` samples while a delegation is outstanding (CB-109)
+        int unknownSincePostTurn;  // the same, for the post-turn housekeeping phase (fleetd #306)
         int notReadySincePoll;     // consecutive injectable samples a queued message waited on the readiness gate (CB-114)
         boolean postTurnPending;   // completion observed; adapter housekeeping has not started yet
         boolean awaitingPostTurnPickup;
@@ -217,10 +218,12 @@ public final class Injector {
                 t.awaitingPickup = false;
                 t.injectableSincePickup = 0;
                 t.unknownSinceTurn = 0;
+                t.unknownSincePostTurn = 0;
                 t.notReadySincePoll = 0;
                 if (t.awaitingCompletion) t.turnObserved = true;
             } else if (status.injectable()) { // IDLE or BLOCKED
                 t.unknownSinceTurn = 0;
+                t.unknownSincePostTurn = 0;
                 if (t.awaitingPostTurnPickup) {
                     if (++t.injectableSincePostTurnPickup >= PICKUP_GRACE_POLLS) {
                         t.awaitingPostTurnPickup = false;
@@ -315,6 +318,24 @@ public final class Injector {
                     t.unknownSinceTurn = 0;
                     turnFailed = true;
                 }
+                // fleetd #306: the same escape for the post-turn housekeeping phase. Four latches
+                // gate delivery (awaitingCompletion, postTurnPending, awaitingPostTurnPickup,
+                // postTurnObserved) and only the first had a way out of a sustained unknown streak —
+                // a gate that closed one direction only. The other two below are released here as
+                // well; postTurnPending needs no escape because it is cleared unconditionally on the
+                // line after the listener call that sets it.
+                //
+                // This does NOT set turnFailed. The delegated turn already completed and its waiter
+                // already resolved — what is outstanding is adapter housekeeping (the `/clear`).
+                // Reporting a turn failure here would drive SessionManager.onFailed on a session
+                // that genuinely finished its work, which is a worse lie than the wedge.
+                if ((t.awaitingPostTurnPickup || t.postTurnObserved)
+                        && ++t.unknownSincePostTurn >= TURN_STALL_GRACE_POLLS) {
+                    t.awaitingPostTurnPickup = false;
+                    t.postTurnObserved = false;
+                    t.injectableSincePostTurnPickup = 0;
+                    t.unknownSincePostTurn = 0;
+                }
             }
 
             // Reclaim the entry once the worker is fully quiescent (nothing queued, no pickup or

fleetd/src/main/java/dev/ltms/fleet/mcp/ConnectionIdentity.java

@@ -36,6 +36,25 @@ public final class ConnectionIdentity {
      * primary / an off-host client) and its {@code pid} (or {@code -1} if not resolvable).
      */
     public record Caller(String terminal, long pid) {
+
+        /**
+         * Whether the OS peer-PID lookup actually succeeded — {@code false} means {@code pid} is
+         * the {@code -1} sentinel, not a real process id, so this caller's identity could not be
+         * established at all. That is a different fact from a real pid that simply owns no worker
+         * pane (the primary's own connection): the primary is {@code resolved()} and has a
+         * {@code null terminal}; an unresolvable caller is {@code !resolved()} and also has a
+         * {@code null terminal}. The two look identical through {@link #terminal} alone, which is
+         * exactly how fleetd #317 happened — a failed {@code lsof} lookup and a genuine primary both
+         * fell through to {@code Principal.primary(...)}.
+         *
+         * <p>Centralised here, next to the sentinel it tests, for the same reason
+         * {@link ConnectionIdentity#isLoopback} is centralised rather than left for each caller to
+         * reimplement: a raw {@code pid > 0} check duplicated at every call site is precisely the
+         * "one rule, two copies" shape that let #305 drift.
+         */
+        public boolean resolved() {
+            return pid > 0;
+        }
     }
 
     /** Resolve the caller's terminal and PID from one peer-PID lookup. */
@@ -60,7 +79,44 @@ public final class ConnectionIdentity {
         return pid > 0 ? cwds.cwdForPid(pid) : null;
     }
 
-    private static boolean isLoopback(String addr) {
-        return "127.0.0.1".equals(addr) || "::1".equals(addr) || "0:0:0:0:0:0:0:1".equals(addr);
+    /**
+     * Whether {@code addr} is a same-host address, and therefore one whose peer PID is worth
+     * looking up. <strong>This is the one definition of loopback in the daemon</strong> —
+     * {@code CallerResolver} calls it rather than keeping its own, because the two used to differ
+     * and that difference was a privilege escalation (fleetd #305).
+     *
+     * <p>The whole of {@code 127.0.0.0/8} counts, not just {@code 127.0.0.1}. On Linux every
+     * address in that range is bound to {@code lo} by default, so a process can connect to
+     * {@code 127.0.0.1:8765} with a source address of {@code 127.0.0.2} — measured on the Linux
+     * fleet host, where binding that source succeeds.
+     *
+     * <p><strong>What excluding an address costs, stated as it is today.</strong> This paragraph
+     * used to say that narrowing this range turned a worker into the lead, and that widening the
+     * check was what closed the hole. That was true only while there were <em>two</em> definitions
+     * that disagreed: {@code ConnectionIdentity} skipped the identity lookup for {@code 127.0.0.2}
+     * while {@code CallerResolver} read the same address as loopback and granted the primary role.
+     * #305 removed the second copy, and with one shared definition the old sentence no longer holds.
+     *
+     * <p>Measured on 2026-09-04 by narrowing this method back to exactly {@code 127.0.0.1} and
+     * running {@code CallerResolverTest} and {@code ConnectionIdentityTest}: a caller from
+     * {@code 127.0.0.2} then resolves to {@code ANONYMOUS}, not {@code PRIMARY} — for a worker
+     * ({@code aWorkerOnAnyLoopbackSourceAddressIsStillAWorkerNotThePrimary}) and for a non-worker
+     * ({@code aNonWorkerOnAnyLoopbackSourceAddressIsStillThePrimary}) alike. Excluding an address
+     * now <em>refuses</em> its caller; it does not promote one.
+     *
+     * <p>So keep the whole range, but for the plain reason: a genuine worker or primary that
+     * connects from {@code 127.0.0.2} must be identifiable at all, and narrowing this predicate
+     * locks it out. That is an outage, and an outage is the direction to fail in — which is exactly
+     * why the range must not be narrowed casually and also why doing so is no longer a security
+     * hole. This predicate still does not decide whether a caller is trusted; it decides whether the
+     * caller's identity is <em>resolved at all</em>. What makes an unresolved caller safe is
+     * {@link Caller#resolved()} (#317), not this method.
+     */
+    public static boolean isLoopback(String addr) {
+        if (addr == null) {
+            return false;
+        }
+        String a = addr.startsWith("::ffff:") ? addr.substring(7) : addr; // IPv4-mapped IPv6
+        return a.startsWith("127.") || "::1".equals(a) || "0:0:0:0:0:0:0:1".equals(a);
     }
 }

fleetd/src/main/java/dev/ltms/fleet/mcp/FleetMcp.java

@@ -22,6 +22,7 @@ import dev.ltms.fleet.placement.BackendQuarantine;
 import dev.ltms.fleet.placement.PlacementException;
 import dev.ltms.fleet.session.SessionManager;
 import dev.ltms.fleet.session.MemberSession;
+import dev.ltms.fleet.session.ShuttingDownException;
 import dev.ltms.fleet.session.WorktreeRequest;
 import dev.ltms.fleet.peer.MemberRole;
 import dev.ltms.fleet.peer.PeerLauncher;
@@ -813,7 +814,12 @@ public final class FleetMcp {
             return error("fleet_reply is for workers only — could not identify the calling worker "
                     + "from the connection");
         }
-        if (content == null) {
+        // fleetd #302: isBlank, not == null, to match fleet_send's own guard above. MessageService
+        // .reply now REJECTS blank content, and this handler is a bare BiFunction with no try/catch
+        // around it — so a whitespace-only fleet_reply would leave here as an uncaught
+        // IllegalArgumentException instead of this clean tool error. Null and whitespace are the
+        // same mistake by the caller and must get the same answer.
+        if (isBlank(content)) {
             return error("content is required");
         }
         messages.reply(callerTerminal, content);
@@ -957,6 +963,10 @@ public final class FleetMcp {
             return text(json(memberView(member)));
         } catch (GuardException e) {
             return error("subscription boundary: " + e.getMessage());
+        } catch (ShuttingDownException e) {
+            // fleetd #308: the daemon's shutdown drain has already started — refuse loudly rather
+            // than register a session drainAll will never see again.
+            return error("shutting down: " + e.getMessage());
         } catch (PlacementException e) {
             // CB-599: no candidate had capacity (maxLoad, quarantine, or all-exhausted) — distinct
             // from "profile does not exist" below.

fleetd/src/main/java/dev/ltms/fleet/mcp/LsofPeerPidLookup.java

@@ -41,6 +41,14 @@ public final class LsofPeerPidLookup implements PeerPidLookup {
             if (!p.waitFor(2, TimeUnit.SECONDS)) {
                 p.destroyForcibly();
             }
+            if (found < 0) {
+                // fleetd #317: this is the silent path — lsof ran clean and simply reported no
+                // matching process (e.g. queried before the OS socket table settles). Previously
+                // this logged nothing at all, which is exactly why the escalation went unnoticed;
+                // the exception path below already logs. A caller now refused because of this is
+                // still refused (never promoted) — this line only makes the refusal diagnosable.
+                log.debug("lsof peer-pid lookup for port {} found no matching process", port);
+            }
             return found;
         } catch (Exception e) {
             log.debug("lsof peer-pid lookup for port {} failed: {}", port, e.getMessage());

fleetd/src/main/java/dev/ltms/fleet/member/CompositePeerLauncher.java

@@ -385,9 +385,12 @@ public final class CompositePeerLauncher implements PeerLauncher {
             }
         }
 
+        // unreachable.size() counts DISTINCT profiles, not attempts (a HashSet dedupes a profile
+        // added twice) — say "distinct" so the count matches the sentence and the profile list that
+        // follows, rather than reading as a count of attempts made (fleetd #315).
         throw new PeerUnreachableException(
                 "no reachable worker profile available after trying " + unreachable.size()
-                        + " candidate(s): " + String.join(", ", unreachable));
+                        + " distinct candidate(s): " + String.join(", ", unreachable));
     }
 
     /**

fleetd/src/main/java/dev/ltms/fleet/msg/AmqpReplyInbox.java

@@ -22,6 +22,7 @@ import java.util.concurrent.ConcurrentSkipListMap;
 import java.util.concurrent.ExecutionException;
 import java.util.concurrent.TimeUnit;
 import java.util.concurrent.TimeoutException;
+import java.util.concurrent.atomic.AtomicReference;
 
 /**
  * AMQP-backed {@link ReplyInbox} (CB-307 Stage 2): genuine cross-restart durability behind the same
@@ -93,8 +94,23 @@ public final class AmqpReplyInbox implements ReplyInbox, AutoCloseable {
     private final Channel channel;
     /** All channel operations (publish/declare/ack/cancel) serialize on this — a Channel is not thread-safe. */
     private final Object channelLock = new Object();
-    /** target → (msgId → held delivery). Per-target map is guarded by synchronizing on itself. */
+    /**
+     * target → (msgId → held delivery). Per-target map is guarded by synchronizing on itself.
+     *
+     * <p><strong>CB-318 tombstone.</strong> The value {@link #RELEASED} is a reserved sentinel: it
+     * marks a target whose {@link #release} has already run, so {@link #deliverCallback} can tell a
+     * delivery landing after release() apart from a fresh target it has never seen. See both methods'
+     * javadoc for why a plain {@code held.remove(target)} is not enough.
+     */
     private final ConcurrentHashMap<String, LinkedHashMap<String, Held>> held = new ConcurrentHashMap<>();
+
+    /**
+     * CB-318 sentinel stored in {@link #held} for a target whose {@link #release} has already run.
+     * Never mutated — every read site compares it by reference ({@code ==}) before touching it as a
+     * map, because it is a single object shared across every released target and calling a mutator on
+     * it would corrupt state for all of them.
+     */
+    private static final LinkedHashMap<String, Held> RELEASED = new LinkedHashMap<>();
     /** Targets whose queue is declared and consumer is running, mapped to their broker consumer tag. */
     private final ConcurrentHashMap<String, String> consumerTags = new ConcurrentHashMap<>();
 
@@ -201,6 +217,12 @@ public final class AmqpReplyInbox implements ReplyInbox, AutoCloseable {
                 channel.queueDeclare(queue, true, false, false, null); // durable, non-exclusive, keep on idle
                 String tag = channel.basicConsume(queue, false, deliverCallback(target), _ -> { });
                 consumerTags.put(target, tag);
+                // CB-318: drop a stale RELEASED tombstone from a prior ownership of this same target
+                // string, so a delivery under this fresh consumer is held normally instead of being
+                // nacked forever by deliverCallback's RELEASED check. Safe to do here, still under
+                // channelLock: no delivery for the consumer tag just registered above can reach
+                // deliverCallback before this basicConsume call returns.
+                held.remove(target, RELEASED);
                 log.debug("AMQP inbox owns queue {} for target {}", queue, target);
             } catch (IOException e) {
                 throw new IllegalStateException("cannot own queue " + queue, e);
@@ -243,6 +265,32 @@ public final class AmqpReplyInbox implements ReplyInbox, AutoCloseable {
      * later connection drop even though this release did not manage to requeue it immediately. A
      * failed {@code basicCancel} still throws, unchanged from before this fix — that failure means
      * the consumer may still be attached, so best-effort requeue is not attempted underneath it.
+     *
+     * <p><strong>CB-318: {@code held.remove(target)} alone leaves a second window open.</strong> The
+     * bullet above already explains why cancelling first does not save a tag from going stale — but
+     * that only accounts for a delivery landing before this method starts touching {@link #held}.
+     * {@code basicCancel} stops <em>new</em> dispatches; it does not flush one already handed to the
+     * consumer work pool. So a delivery can still land on that pool's thread and reach
+     * {@link #deliverCallback} at any point during, or after, this method's body — and a plain
+     * {@code held.remove(target)} does nothing to stop it: {@code deliverCallback}'s
+     * {@code computeIfAbsent} finds the key gone and happily creates a brand-new map under it, which
+     * this method — already past its {@code remove} — never looks at again. That entry then sits
+     * delivered-but-unacked on {@link #channel} until the whole inbox closes: never requeued, never
+     * redelivered, and {@link #peek} is never called again for a target nothing owns any more.
+     *
+     * <p>The fix is {@link #held}{@code .compute(target, ...)} instead of {@code remove}: it takes
+     * whatever was held (to nack, same as before) and, in the same atomic step, leaves the
+     * {@link #RELEASED} tombstone behind instead of an absent key. {@code computeIfAbsent} and
+     * {@code compute} calls for the same key are mutually exclusive in {@link ConcurrentHashMap} —
+     * whichever of this call and a concurrent {@code deliverCallback} runs first is fully visible to
+     * the other, with no gap between them. So a delivery that loses the race sees a real map here and
+     * gets nacked by the loop below, same as always; a delivery that wins the race (runs first) is
+     * itself nacked by that same loop, once it settles into {@code held}. A delivery that arrives once
+     * this method has stored {@link #RELEASED} finds it via {@code computeIfAbsent} and refuses itself
+     * — see {@link #deliverCallback}. Either way nothing is silently retained forever, satisfying the
+     * ticket's invariant against dropping a message. This closes the window rather than merely
+     * narrowing it — correctness does not depend on how much time elapses between the swap and this
+     * method returning.
      */
     @Override
     public void release(String target) {
@@ -255,8 +303,13 @@ public final class AmqpReplyInbox implements ReplyInbox, AutoCloseable {
                     throw new IllegalStateException("cannot cancel consumer for " + target, e);
                 }
             }
-            var perTarget = held.remove(target);
-            if (perTarget != null) {
+            AtomicReference<LinkedHashMap<String, Held>> previouslyHeld = new AtomicReference<>();
+            held.compute(target, (_, v) -> {
+                previouslyHeld.set(v);
+                return RELEASED;
+            });
+            var perTarget = previouslyHeld.get();
+            if (perTarget != null && perTarget != RELEASED) {
                 synchronized (perTarget) {
                     for (Held h : perTarget.values()) {
                         try {
@@ -326,7 +379,7 @@ public final class AmqpReplyInbox implements ReplyInbox, AutoCloseable {
     @Override
     public List<InboxMessage> peek(String target) {
         var perTarget = held.get(target);
-        if (perTarget == null) {
+        if (perTarget == null || perTarget == RELEASED) {
             return List.of();
         }
         synchronized (perTarget) {
@@ -337,7 +390,7 @@ public final class AmqpReplyInbox implements ReplyInbox, AutoCloseable {
     @Override
     public void ack(String target, String msgId) {
         var perTarget = held.get(target);
-        if (perTarget == null) {
+        if (perTarget == null || perTarget == RELEASED) {
             return;
         }
         Held h;
@@ -370,6 +423,20 @@ public final class AmqpReplyInbox implements ReplyInbox, AutoCloseable {
             }
             String content = new String(delivery.getBody(), StandardCharsets.UTF_8);
             var perTarget = held.computeIfAbsent(target, _ -> new LinkedHashMap<>());
+            if (perTarget == RELEASED) {
+                // CB-318: release() already ran for this target and left the RELEASED tombstone in
+                // held (see release()'s javadoc) — computeIfAbsent() is guaranteed to see it rather
+                // than recreate a fresh map, because ConcurrentHashMap serializes compute/
+                // computeIfAbsent calls for the same key against each other. Refuse the delivery
+                // instead of holding it somewhere release() will never look at again: requeue it, the
+                // same way release() nacks its own held entries, so a later owner (or a connection
+                // drop) can still recover it. This does not need channelLock across a broker round
+                // trip — basicNack, like the duplicate-ack case just below, does not wait for one.
+                synchronized (channelLock) {
+                    channel.basicNack(tag, false, true);
+                }
+                return;
+            }
             boolean duplicate;
             synchronized (perTarget) {
                 if (perTarget.containsKey(msgId)) {

fleetd/src/main/java/dev/ltms/fleet/msg/MessageService.java

@@ -187,6 +187,24 @@ public final class MessageService {
         private volatile Long completedNanos;
         private volatile Reply question;
         private volatile String turnId;
+        /**
+         * Set when this task's {@code fleet_ask} lapsed with no answer (fleetd #307):
+         * {@link #clearAsyncQuestion} then forgets {@link #turnId} (nulls it and drops the task from
+         * {@code asyncTasksByTurn}) so {@link #hasAsyncQuestion} stops reporting the target BUSY — a
+         * later {@code fleet_send} to it must be accepted, not refused. But the worker's turn is
+         * still genuinely live: it resumed on its own and will eventually call its real
+         * {@code fleet_reply}. Losing {@link #turnId} loses {@link #askAnsweredAsyncTasks}' only
+         * signal that such a reply belongs to this task, so that reply used to fall straight to the
+         * inbox and strand — {@code fleet_poll} stayed {@code PENDING} forever, later force-failed by
+         * {@link #abandon} with the misleading "session released before it replied". This flag is a
+         * second, independent signal that survives the forgetting: {@link #askAnsweredAsyncTasks}
+         * accepts it in place of a live {@link #turnId}, without ever re-adding the task to
+         * {@code asyncTasksByTurn} (so the BUSY release is untouched). Cleared implicitly once
+         * {@link #future} resolves — every match in {@link #askAnsweredAsyncTasks} already requires
+         * {@code !future.isDone()}, so a task that recovered (or was later failed by
+         * {@link #abandon}) can never match again regardless of this flag's value.
+         */
+        private volatile boolean askTimedOut;
 
         private Task(String ticket, String target, LongSupplier nowNanos) {
             this.ticket = ticket;
@@ -395,18 +413,16 @@ public final class MessageService {
      * {@link Rendezvous#resolveQuestion} must keep today's {@code NO_WAITER} behaviour — questions
      * are interactive and must never be queued.
      *
-     * <p><strong>Ambiguous match also falls to the inbox.</strong> {@link #askAnsweredAsyncTasks}
-     * cannot actually return more than one entry today (see its own javadoc for why — in short,
-     * {@link #hasAsyncQuestion} keeps a target BUSY, so no second task can reach this state, for as
-     * long as an earlier one's {@code turnId} is still stamped). That is an emergent guarantee from
-     * two other facts, not one this method enforces, so this branch stays in as defence in depth
-     * rather than being removed as dead code: if it ever weakens, returning whichever candidate a
-     * {@code ConcurrentHashMap} iteration reaches first would let a genuine reply complete the
-     * <em>wrong</em> ticket — silently handing the lead something that reads like a correct answer to
-     * a delegation the worker never touched, which is worse than a failure because the lead acts on
-     * it. When more than one candidate exists, guessing is not safe: fall back to the inbox exactly
-     * as the zero-candidate case does, and let {@link #abandon} apply the eventual recovery
-     * deterministically instead.
+     * <p><strong>Ambiguous match also falls to the inbox.</strong> {@link #askAnsweredAsyncTasks} can
+     * return more than one entry — a reachable state, not a hypothetical one (see its own javadoc:
+     * an {@code fleet_ask} that lapsed with no answer, fleetd #307, frees the target for a completely fresh
+     * delegation, which can itself go on to ask-and-lapse before the first worker's real reply
+     * arrives). Returning whichever candidate a {@code ConcurrentHashMap} iteration reaches first
+     * would let a genuine reply complete the <em>wrong</em> ticket — silently handing the lead
+     * something that reads like a correct answer to a delegation the worker never touched, which is
+     * worse than a failure because the lead acts on it. When more than one candidate exists, guessing
+     * is not safe: fall back to the inbox exactly as the zero-candidate case does, and let
+     * {@link #abandon} apply the eventual recovery deterministically instead.
      *
      * <p><strong>{@code content} is required (fleetd #302).</strong> Both doors that reach this
      * method must reject a missing/blank reply the same way, so the check lives here rather than in
@@ -434,15 +450,18 @@ public final class MessageService {
             count(FleetMetrics.REPLIES, "path", "rendezvous");
             return true;                       // a live send took it — unchanged fast path
         }
-        // #137: no live rendezvous waiter, but this may be the worker's real fleet_reply resuming a
-        // turn that {@link #answer} already gave up waiting on. answer()'s own bounded wait (the
-        // primary's fleet_send{turnId} call, capped well under a minute) can time out and close its
-        // waiter long before the worker — now actually resuming real work — finishes and replies. That
-        // reply used to have nowhere to land but the session inbox, leaving the async ticket's future
-        // unresolved forever: fleet_poll{ticket} stayed PENDING until fleet_stop's abandon() forced it
-        // FAILED with a misleading "session released before it replied" reason, even though the reply
-        // had, in fact, arrived. Completing the matching ticket directly here means fleet_poll{ticket}
-        // sees the real reply instead.
+        // #137/fleetd #307: no live rendezvous waiter, but this may be the worker's real fleet_reply resuming
+        // a turn that either answer() (#137) or ask() (fleetd #307) already gave up waiting on:
+        //  - answer()'s own bounded wait (the primary's fleet_send{turnId} call, capped well under a
+        //    minute) can time out and close its waiter long before the worker — now actually resuming
+        //    real work — finishes and replies.
+        //  - ask()'s own wait for the primary can time out first, with the worker resuming on its own
+        //    and finishing unanswered.
+        // Either way that reply used to have nowhere to land but the session inbox, leaving the async
+        // ticket's future unresolved forever: fleet_poll{ticket} stayed PENDING until fleet_stop's
+        // abandon() forced it FAILED with a misleading "session released before it replied" reason,
+        // even though the reply had, in fact, arrived. Completing the matching ticket directly here
+        // means fleet_poll{ticket} sees the real reply instead.
         List<Task> candidates = askAnsweredAsyncTasks(session);
         if (candidates.size() == 1) {
             Task orphan = candidates.get(0);
@@ -473,34 +492,42 @@ public final class MessageService {
     }
 
     /**
-     * Every still-open async task on {@code target} whose {@code fleet_ask} was already answered —
-     * its {@link Task#turnId} is stamped but its {@link Task#question} was cleared by {@link #answer}
-     * — yet whose future is not resolved yet (#137). Empty if no such task exists, including the
-     * common case where {@code target}'s worker never used {@code fleet_ask} at all (a task that was
-     * never asked has {@code turnId == null}, so it can never match here and only ever completes
-     * through the ordinary rendezvous fast path in {@link #reply}).
+     * Every still-open async task on {@code target} whose worker is genuinely expected to send a
+     * real {@code fleet_reply} next with nothing left registered to catch it: either its
+     * {@code fleet_ask} was already answered — {@link Task#turnId} is stamped but {@link
+     * Task#question} was cleared by {@link #answer} — or its {@code fleet_ask} lapsed unanswered and
+     * {@link Task#askTimedOut} marks that (fleetd #307; {@link Task#turnId} is {@code null} by then, forgotten
+     * so the target is not left BUSY — see {@link Task#askTimedOut}'s own javadoc). Either way the
+     * task's future is not resolved yet. Empty if no such task exists, including the common case
+     * where {@code target}'s worker never used {@code fleet_ask} at all (a task that was never asked
+     * has both {@code turnId == null} and {@code askTimedOut == false}, so it can never match here and
+     * only ever completes through the ordinary rendezvous fast path in {@link #reply}).
      *
-     * <p><strong>Returns at most one entry today — verified, not assumed.</strong> {@link #send}
-     * refuses to open a waiter on {@code target} while {@link #hasAsyncQuestion} is true, and that
-     * check matches ANY task whose {@code turnId} is still stamped in {@code asyncTasksByTurn} —
-     * not only while its question is still open. {@link #answer} deliberately leaves that stamp in
-     * place ({@code clearAsyncQuestion(turnId, false)}) until the resumed turn's own future actually
-     * resolves, at which point {@link #finishAsyncTask} both removes the stamp AND completes that
-     * task's future in the same call. So a second task can never reach "{@code turnId} stamped, future
-     * still open" — the exact pair this method matches on — while a first one already holds it: by
-     * the time the stamp is gone, so is the eligibility. This is an emergent property of those two
-     * facts holding together, not something this method (or its callers) enforces on its own — flip
-     * {@code forgetTurn} to {@code true} in that one {@link #answer} call and it silently stops being
-     * true, with nothing left to fail loudly. The callers below still handle "more than one" as
-     * defence in depth against exactly that, not because they exercise it today: {@link #reply}
-     * treats it as unresolvable and falls back to the inbox; {@link #abandon} would pick the oldest
-     * deterministically (its own {@code matching} list has no such guarantee — see its javadoc).
+     * <p><strong>Can return more than one entry — reachable, not just defence in depth.</strong>
+     * {@link #send} refuses to open a waiter on {@code target} while {@link #hasAsyncQuestion} is
+     * true, and that check matches ANY task whose {@code turnId} is still stamped in
+     * {@code asyncTasksByTurn}. While a task's {@code turnId} stays stamped — {@link #answer} leaves
+     * it in place ({@code clearAsyncQuestion(turnId, false)}) until {@link #finishAsyncTask} removes
+     * the stamp and completes the future in the same call — no second task on the same target can
+     * reach an eligible state, because {@link #send} would refuse it as BUSY first. That single-task
+     * guarantee holds only for the {@code turnId}-stamped half of this method's match: an
+     * {@link Task#askTimedOut} task is, by construction, no longer stamped in {@code asyncTasksByTurn}
+     * (that is the whole point of forgetting {@code turnId} in {@link #clearAsyncQuestion}), so the
+     * target is free the moment one ask lapses. A fresh, independent {@code sendAsync} to the same
+     * target can then be dispatched, itself pause on {@code fleet_ask}, and itself time out — landing
+     * a second {@code askTimedOut} task on the very target the first one is still waiting to answer
+     * for. Two (or more) genuinely open tasks on one target is therefore a real, reachable state
+     * today, not a hypothetical: {@link #reply} treats it as unresolvable and falls back to the
+     * inbox rather than guess which task a reply belongs to (guessing wrong would hand the lead a
+     * plausible-looking answer to a delegation the worker never touched — worse than a failure,
+     * because the lead acts on it); {@link #abandon} instead picks the oldest deterministically (its
+     * own {@code matching} list has a different, wider match — see its javadoc).
      */
     private List<Task> askAnsweredAsyncTasks(String target) {
         List<Task> candidates = new ArrayList<>();
         for (Task task : tasks.values()) {
-            if (target.equals(task.target) && task.question == null && task.turnId != null
-                    && !task.future.isDone()) {
+            if (target.equals(task.target) && task.question == null && !task.future.isDone()
+                    && (task.turnId != null || task.askTimedOut)) {
                 candidates.add(task);
             }
         }
@@ -898,6 +925,12 @@ public final class MessageService {
             return new AskResult(AskOutcome.ANSWERED, answer);
         } catch (TimeoutException e) {
             log.debug("fleet_ask from {} went unanswered in {}ms", workerSession, timeoutMillis);
+            // fleetd #307: mark the task BEFORE clearAsyncQuestion(forgetTurn=true) below drops it out of
+            // asyncTasksByTurn and nulls its turnId — that forgetting is deliberate and stays (it is
+            // what keeps the target from staying BUSY forever), but it would otherwise also erase
+            // askAnsweredAsyncTasks' only signal that the worker's eventual real fleet_reply still
+            // belongs to this task, stranding it in the inbox with a false "never replied" verdict.
+            markAskTimedOut(ticket.turnId());
             clearAsyncQuestion(ticket.turnId(), true);
             return new AskResult(AskOutcome.TIMED_OUT, null);
         } catch (ExecutionException e) {
@@ -1162,6 +1195,23 @@ public final class MessageService {
         return task;
     }
 
+    /**
+     * Mark {@code turnId}'s task as having a {@code fleet_ask} that lapsed with no answer (fleetd #307), so
+     * {@link #askAnsweredAsyncTasks} still recognizes the worker's eventual real {@code fleet_reply}
+     * as belonging to it after {@link #clearAsyncQuestion}'s {@code forgetTurn=true} erases
+     * {@link Task#turnId} — see {@link Task#askTimedOut}. Must be called before that forgetting, while
+     * {@code turnId} can still resolve the task in {@code asyncTasksByTurn}; a lookup afterward would
+     * find nothing. Only when it matches the task's current turn — same guard as
+     * {@link #clearAsyncQuestion} — so a chained second {@code fleet_ask} (#282) that already moved
+     * the task to a fresh {@code turnId} cannot mark it for a turn that is no longer its own.
+     */
+    private void markAskTimedOut(String turnId) {
+        Task task = asyncTasksByTurn.get(turnId);
+        if (task != null && turnId.equals(task.turnId)) {
+            task.askTimedOut = true;
+        }
+    }
+
     /** Clear an answered or lapsed question, but only when it matches the ticket's current turn. */
     private void clearAsyncQuestion(String turnId, boolean forgetTurn) {
         // CB-582: tell the push loop first — like ticketCollected, a removal for a turnId it never
@@ -1180,14 +1230,61 @@ public final class MessageService {
         }
     }
 
-    /** Complete and detach an async ticket after its worker's actual terminal reply. */
+    /**
+     * Complete and detach an async ticket after its worker's actual terminal reply.
+     *
+     * <p><strong>fleetd #324.</strong> {@code task.turnId} is read into {@code turnId} exactly once.
+     * It used to be read twice — once for the null check, once as the removal key — and {@code
+     * volatile} makes each of those reads individually fresh but does not make the pair atomic.
+     * {@link #answer} calls this while holding {@code sessionLocks} for the target; {@link #ask}'s
+     * own timeout path calls {@link #clearAsyncQuestion} (which nulls {@link Task#turnId}) under no
+     * lock at all. When that unlocked null-out landed between the two reads here, the second read saw
+     * {@code null} and {@code asyncTasksByTurn.remove(null, task)} threw {@code NullPointerException}
+     * on the lead's own {@code answer()} call — even though {@code task.future.complete(result)} on
+     * the line above had already run, so the answer was in fact delivered. Capturing the field once
+     * removes the torn read; see the ticket for why the wider asymmetry between the locked and
+     * unlocked sides is not fixed by this alone.
+     */
     private void finishAsyncTask(Task task, Reply result) {
         task.future.complete(result);
-        if (task.turnId != null) {
-            asyncTasksByTurn.remove(task.turnId, task);
+        String turnId = task.turnId;
+        if (turnId != null) {
+            if (finishAsyncTaskRaceHook != null) {
+                // Test-only (fleetd #324): see the field's own javadoc.
+                finishAsyncTaskRaceHook.run();
+            }
+            asyncTasksByTurn.remove(turnId, task);
         }
     }
 
+    /**
+     * Null in production; test seam for fleetd #324 — invoked from {@link #finishAsyncTask(Task,
+     * Reply)} right after {@code task.turnId}'s null-check passes and before the (now-local) value is
+     * used for the removal. A test installs this to force, deterministically, the exact interleaving
+     * that a real race between this method and {@link #ask}'s unlocked timeout cleanup can otherwise
+     * only produce by chance: firing it here reproduces "the field went null between the check and the
+     * use" against the pre-fix code, and demonstrates the fix tolerates it (the captured local is used
+     * unconditionally, so a hook that nulls the field afterward cannot affect this call).
+     */
+    private volatile Runnable finishAsyncTaskRaceHook;
+
+    /**
+     * Test-only (fleetd #324): install {@link #finishAsyncTaskRaceHook}. Package-private so the test,
+     * in the same package, can reach it without widening any production API.
+     */
+    void setFinishAsyncTaskRaceHookForTest(Runnable hook) {
+        this.finishAsyncTaskRaceHook = hook;
+    }
+
+    /**
+     * Test-only (fleetd #324): run the exact production cleanup {@link #ask}'s own timeout path runs
+     * unlocked — {@link #clearAsyncQuestion(String, boolean)} with {@code forgetTurn=true} — so a test
+     * can reproduce that specific mutation instead of hand-rolling an approximation of it.
+     */
+    void forgetTurnForTest(String turnId) {
+        clearAsyncQuestion(turnId, true);
+    }
+
     /** Complete the async ticket correlated to a specific answered turn. */
     private void finishAsyncTask(String turnId, Reply result) {
         Task task = asyncTasksByTurn.get(turnId);

fleetd/src/main/java/dev/ltms/fleet/placement/FixedPlacementPolicy.java

@@ -5,10 +5,14 @@ import java.util.List;
 
 /**
  * Backward-compatible placement: an unqualified spawn always resolves to the configured default
- * profile, exactly as {@code CompositePeerLauncher} did before CB-518. This ignores caps and
- * reachability so that a pre-existing config behaves identically after upgrade.
+ * profile, exactly as {@code CompositePeerLauncher} did before CB-518. This ignores caps
+ * ({@code maxLoad}) so that a pre-existing config behaves identically after upgrade — capacity
+ * gating for automatic placement is deliberately out of scope for {@code fixed}, exactly as it
+ * always has been. Reachability is a narrower exception (fleetd #315, below): a profile is never
+ * checked for reachability up front, only skipped once it has already failed in <em>this same</em>
+ * spawn call's retry loop — see the unreachable case below.
  *
- * <p>Three exceptions walk past the default instead of returning it unconditionally:
+ * <p>Four exceptions walk past the default instead of returning it unconditionally:
  * <ul>
  *   <li>Quarantine (CB-578 stage B): a quarantined default is a credential that just refused on
  *       a usage limit, not a transient capacity or reachability concern.
@@ -16,13 +20,21 @@ import java.util.List;
  *       ({@code BackendOutagePolicy}) — a separate, shorter-lived source from quarantine. When a
  *       profile is both quarantined and cooling off, only the quarantine reason is reported
  *       (exhaustion takes priority), matching {@code CompositePeerLauncher}'s explicit-spawn order.
+ *   <li>Unreachable (fleetd #315): {@code CompositePeerLauncher.spawn} retries a failed candidate
+ *       on the next one and rebuilds the {@link PlacementContext} so {@code ctx.unreachable()}
+ *       names every profile that already failed with {@code PeerUnreachableException} in this same
+ *       call. Without this check {@code select} kept handing back the same dead default forever —
+ *       the retry loop's own comment says "so the policy excludes this profile", and this is what
+ *       makes that true for {@code fixed} too, matching {@code weighted}/{@code round-robin}
+ *       (both filter on {@code ctx.unreachable()} via {@link PlacementPolicyUtil#available}).
  *   <li>Weight 0 (CB-554): {@code fixed} is still automatic selection, so a profile the operator
  *       marked "never auto-select me" ({@code weight <= 0}) must be skipped here exactly as
  *       {@code weighted}/{@code round-robin} skip it — an explicit {@code fleet_spawn} naming
  *       the profile is unaffected, only this automatic fallback walk.
  * </ul>
- * A fleet where nothing is ever quarantined, cooling off, or weight-0 never exercises any of these
- * paths, so today's behaviour is unchanged.
+ * A fleet where nothing is ever quarantined, cooling off, unreachable, or weight-0 never exercises
+ * any of these paths, so today's behaviour is unchanged — in particular, the very first selection
+ * of a spawn call always sees an empty {@code unreachable} set, so the first choice is untouched.
  */
 final class FixedPlacementPolicy implements PlacementPolicy {
 
@@ -30,12 +42,12 @@ final class FixedPlacementPolicy implements PlacementPolicy {
     public PlacementCandidate select(PlacementContext ctx) {
         String d = ctx.defaultProfile();
         if (d != null && !d.isBlank() && !ctx.quarantined().contains(d) && !ctx.coolingOff().contains(d)
-                && !weightExcluded(ctx, d)) {
+                && !ctx.unreachable().contains(d) && !weightExcluded(ctx, d)) {
             return new PlacementCandidate(d, null, 1.0f, null);
         }
         for (PlacementCandidate c : ctx.candidates()) {
             if (!ctx.quarantined().contains(c.profile()) && !ctx.coolingOff().contains(c.profile())
-                    && !c.excluded()) {
+                    && !ctx.unreachable().contains(c.profile()) && !c.excluded()) {
                 return new PlacementCandidate(c.profile(), null, c.weight(), c.maxLoad());
             }
         }
@@ -44,8 +56,9 @@ final class FixedPlacementPolicy implements PlacementPolicy {
             // Exhaustion quarantine takes priority: reported only when quarantine is absent, so the
             // message never claims "cooling off" for a profile that is really backend-exhausted.
             boolean dCoolingOff = !dQuarantined && ctx.coolingOff().contains(d);
+            boolean dUnreachable = ctx.unreachable().contains(d);
             boolean dWeightExcluded = weightExcluded(ctx, d);
-            if (dQuarantined || dCoolingOff || dWeightExcluded) {
+            if (dQuarantined || dCoolingOff || dUnreachable || dWeightExcluded) {
                 List<String> reasons = new ArrayList<>();
                 if (dQuarantined) {
                     reasons.add("is quarantined (backend exhausted)");
@@ -53,6 +66,9 @@ final class FixedPlacementPolicy implements PlacementPolicy {
                 if (dCoolingOff) {
                     reasons.add("is cooling off after repeated backend errors");
                 }
+                if (dUnreachable) {
+                    reasons.add("is unreachable");
+                }
                 if (dWeightExcluded) {
                     reasons.add("has weight 0 (excluded from automatic selection)");
                 }
@@ -62,7 +78,7 @@ final class FixedPlacementPolicy implements PlacementPolicy {
         }
         if (!ctx.candidates().isEmpty()) {
             throw new PlacementException("all worker profiles are excluded from automatic "
-                    + "selection (quarantined, cooling off, or weight-0)");
+                    + "selection (quarantined, cooling off, unreachable, or weight-0)");
         }
         throw new PlacementException("no worker profiles configured");
     }

fleetd/src/main/java/dev/ltms/fleet/rest/FleetApp.java

@@ -19,6 +19,7 @@ import dev.ltms.fleet.peer.PeerUnreachableException;
 import dev.ltms.fleet.placement.PlacementException;
 import dev.ltms.fleet.msg.MessageService;
 import dev.ltms.fleet.session.SessionManager;
+import dev.ltms.fleet.session.ShuttingDownException;
 import dev.ltms.fleet.peer.MemberRole;
 import dev.ltms.fleet.session.MemberSession;
 import dev.ltms.fleet.session.WorktreeRequest;
@@ -501,6 +502,11 @@ public final class FleetApp {
             ctx.status(201).json(view(member));
         } catch (GuardException e) {
             ctx.status(403).json(Map.of("error", "subscription_boundary", "detail", e.getMessage()));
+        } catch (ShuttingDownException e) {
+            // fleetd #308: the daemon's shutdown drain has already started — 503, not a bare 500,
+            // so this reads the same as PlacementException below: valid request, refused because
+            // of a transient daemon state rather than a bad argument.
+            ctx.status(503).json(Map.of("error", "shutting_down", "detail", e.getMessage()));
         } catch (PlacementException e) {
             // CB-599: no candidate had capacity (maxLoad, quarantine, or all-exhausted) — a benign,
             // likely-transient refusal, distinct from "profile does not exist" below. 503: the
@@ -510,6 +516,12 @@ public final class FleetApp {
             ctx.status(400).json(Map.of("error", "unknown_profile", "detail", e.getMessage()));
         } catch (PeerUnreachableException e) {
             ctx.status(502).json(Map.of("error", "spawn_timeout", "detail", e.getMessage()));
+        } catch (HerdrException e) {
+            // fleetd #304: not every herdr failure on the spawn path is a readiness timeout, so
+            // PeerUnreachableException above does not cover this. Without this catch the exception
+            // escapes to Javalin's default 500, while fleet_spawn reports the same failure as a
+            // clean named error (FleetMcp.spawn) — the #297 one-door-guarded shape.
+            herdrError(ctx, e);
         }
     }
 
@@ -530,13 +542,29 @@ public final class FleetApp {
         return (s == null || s.isBlank()) ? null : s;
     }
 
-    /** Tear a worker down by pane id. */
+    /**
+     * Tear a worker down by pane id.
+     *
+     * <p>fleetd #304: the {@code HerdrException} catch is not cosmetic. {@code release} deregisters
+     * the session, notifies the release listener and preserves a dirty worktree <em>before</em> it
+     * calls {@code launcher.stop}, so a throw from that stop arrives after the teardown the caller
+     * asked for has already happened. Letting it escape gave Javalin's default 500, which tells the
+     * caller to retry — and the retry finds nothing in the registry, reaches the same stop, and
+     * throws again, so it can never succeed. {@code herdrError} instead answers 404 ("the pane is
+     * gone, stop retrying") or 502 ("herdr is upstream and broken, a retry may help"), matching what
+     * {@code fleet_stop} reports for the same failure.
+     */
     private void stopMember(Context ctx) {
         String paneId = ctx.pathParam("paneId");
         if (!allow(ctx, routeAction("DELETE /members/{paneId}"), paneId)) {
             return;
         }
-        sessions.release(paneId);
+        try {
+            sessions.release(paneId);
+        } catch (HerdrException e) {
+            herdrError(ctx, e);
+            return;
+        }
         ctx.status(204);
     }
 

fleetd/src/main/java/dev/ltms/fleet/session/GitWorktrees.java

@@ -93,6 +93,9 @@ public final class GitWorktrees implements Worktrees {
     /** OS group name for {@link #shareWithGroup} (fleetd #185 stage 3); {@code null} ⇒ feature off. */
     private final String group;
     private final Consumer<String> afterWorktreeAdded;
+    /** How the initial {@code git worktree add} command runs. Package-private test seam for an
+     *  interrupted command after Git has made worktree state. */
+    private final Function<String[], String> worktreeAddRunner;
     /** How {@link #shareWithGroup}'s processes (git config / chgrp / chmod / find) actually run.
      *  Defaults to the real {@link #exec(String...)}. Package-private test seam so a unit test can
      *  prove "no group configured ⇒ zero processes spawned" and inspect exactly what a configured
@@ -132,7 +135,13 @@ public final class GitWorktrees implements Worktrees {
 
     /** Test seam combining a configurable {@code group} with {@link #afterWorktreeAdded}. */
     GitWorktrees(String configuredRoot, String group, Consumer<String> afterWorktreeAdded) {
-        this(configuredRoot, group, afterWorktreeAdded, null);
+        this(configuredRoot, group, afterWorktreeAdded, null, null);
+    }
+
+    /** Test seam for changing how {@link #shareWithGroup}'s processes run. */
+    GitWorktrees(String configuredRoot, String group, Consumer<String> afterWorktreeAdded,
+                  Function<String[], String> shareGroupRunner) {
+        this(configuredRoot, group, afterWorktreeAdded, shareGroupRunner, null);
     }
 
     /**
@@ -141,13 +150,15 @@ public final class GitWorktrees implements Worktrees {
      * exactly what commands a configured group runs, without a real second OS user/group.
      *
      * @param shareGroupRunner {@code null} ⇒ the real {@link #exec(String...)}.
+     * @param worktreeAddRunner {@code null} ⇒ the real {@link #exec(String...)}.
      */
     GitWorktrees(String configuredRoot, String group, Consumer<String> afterWorktreeAdded,
-                 Function<String[], String> shareGroupRunner) {
+                  Function<String[], String> shareGroupRunner, Function<String[], String> worktreeAddRunner) {
         this.configuredRoot = configuredRoot;
         this.group = (group == null || group.isBlank()) ? null : group;
         this.afterWorktreeAdded = afterWorktreeAdded == null ? _ -> {} : afterWorktreeAdded;
         this.shareGroupRunner = shareGroupRunner != null ? shareGroupRunner : this::exec;
+        this.worktreeAddRunner = worktreeAddRunner != null ? worktreeAddRunner : this::exec;
     }
 
     @Override
@@ -166,8 +177,8 @@ public final class GitWorktrees implements Worktrees {
         String wt = path.toAbsolutePath().toString();
         log.info("adding worktree branch={} path={} base={}", branch, wt, base);
         removeUserInfoFromHttpsOrigin(repoRoot);
-        exec("git", "-C", repoRoot, "worktree", "add", wt, "-b", branch, base);
         try {
+            worktreeAddRunner.apply(new String[] {"git", "-C", repoRoot, "worktree", "add", wt, "-b", branch, base});
             afterWorktreeAdded.accept(wt);
             requireCredentialFreeHttpsOrigin(wt);
             configureEnvironmentCredentialHelper(repoRoot, wt);
@@ -181,10 +192,11 @@ public final class GitWorktrees implements Worktrees {
     }
 
     /**
-     * {@code add()} has already created the worktree and its branch by the time any step from
-     * {@link #afterWorktreeAdded} through {@link #isolateToolSurface} can throw — including
+     * {@code git worktree add} may have created the worktree and its branch by the time it, or any
+     * later step through {@link #isolateToolSurface}, throws. This includes
      * {@link #requireCredentialFreeHttpsOrigin}, an intended security refusal, not only an IO
-     * accident. Without this, {@code add()} never returns, so its caller
+     * accident. A Git-reported {@code worktree add} failure usually creates nothing, but an
+     * interrupted command can leave partial state. Without cleanup, {@code add()} never returns, so its caller
      * ({@code SessionManager#acquireWithWorktree}) never receives a path to register or clean up:
      * its local {@code path} stays null, the {@code if (path != null)} guard in its own catch block
      * never runs, and the worktree directory and branch leak on disk forever with nothing tracking
@@ -204,6 +216,10 @@ public final class GitWorktrees implements Worktrees {
      * used in {@code SessionManager#acquireWithWorktree}'s own catch block.
      */
     private void cleanupAfterAddFailure(String repoRoot, String worktreePath, String branch, RuntimeException original) {
+        if (!Files.exists(Path.of(worktreePath))) {
+            log.debug("provisioning failed before worktree {} existed; nothing to clean up", worktreePath);
+            return;
+        }
         log.warn("provisioning failed for branch={} path={}: {} — cleaning up before rethrowing",
                 branch, worktreePath, original.getMessage());
         try {

fleetd/src/main/java/dev/ltms/fleet/session/SessionManager.java

@@ -21,6 +21,7 @@ import java.util.Optional;
 import java.util.Set;
 import java.util.concurrent.ConcurrentHashMap;
 import java.util.concurrent.TimeUnit;
+import java.util.concurrent.atomic.AtomicBoolean;
 import java.util.concurrent.atomic.AtomicLong;
 import java.util.function.Consumer;
 import java.util.function.LongSupplier;
@@ -61,6 +62,8 @@ public final class SessionManager implements TurnListener {
     private final LongSupplier nowNanos;
     private final int contextCap;
     private final boolean clearAfterTurn;
+    /** Null in production; test seam for the interval before an idle session's conditional release. */
+    private final Consumer<MemberSession> beforeIdleRelease;
     private volatile MemberLifecycle memberLifecycle = MemberLifecycle.NONE;
     /**
      * CB-586: the repo root the fleet actually works in, remembered the first time a worktree
@@ -71,6 +74,16 @@ public final class SessionManager implements TurnListener {
      */
     private volatile String fleetRepoRoot;
 
+    /**
+     * fleetd #308: flips true the instant {@link #drainAll} starts, before its registry snapshot
+     * is even taken — so a spawn already in flight sees the refusal as early as a plain flag can
+     * make it. This alone cannot close the race completely: a caller that read {@code false} just
+     * before the flip can still land in the registry after the snapshot. {@link #drainAll}'s
+     * post-loop sweep is what catches that straggler; the two mechanisms are deliberately paired,
+     * see {@link #drainAll}'s javadoc.
+     */
+    private final AtomicBoolean draining = new AtomicBoolean(false);
+
     /** CB-520: notified with a terminalId on every acquire; no-op until wired. */
     private final List<Consumer<String>> acquireListeners = new java.util.concurrent.CopyOnWriteArrayList<>();
     /** CB-516: notified with a {@link ReleaseDetail} on every release; no-op until wired. */
@@ -102,13 +115,23 @@ public final class SessionManager implements TurnListener {
     }
 
     public SessionManager(PeerLauncher launcher, Worktrees worktrees, LongSupplier nowNanos,
-                          int contextCap, boolean clearAfterTurn) {
+                           int contextCap, boolean clearAfterTurn) {
+        this(launcher, worktrees, nowNanos, contextCap, clearAfterTurn, null);
+    }
+
+    /**
+     * Package-private constructor for a deterministic reap/delivery race test. Production callers
+     * use the constructor above, whose null hook adds no callback or lock to an ordinary reap.
+     */
+    SessionManager(PeerLauncher launcher, Worktrees worktrees, LongSupplier nowNanos,
+                   int contextCap, boolean clearAfterTurn, Consumer<MemberSession> beforeIdleRelease) {
         this.launcher = launcher;
         this.worktrees = worktrees;
         this.presence = new PresenceFleet(this);
         this.nowNanos = nowNanos;
         this.contextCap = contextCap;
         this.clearAfterTurn = clearAfterTurn;
+        this.beforeIdleRelease = beforeIdleRelease;
     }
 
     /**
@@ -181,6 +204,14 @@ public final class SessionManager implements TurnListener {
     public MemberSession acquire(String profile, MemberRole role, String requestedCwd, String callerCwd,
                                  String ownerTerminal, WorktreeRequest wt,
                                  String sessionName, String resumeSessionId) {
+        // fleetd #308: refuse before anything else runs — no slot reservation, no launcher spawn —
+        // so a caller learns the daemon is going down instead of getting a session drainAll will
+        // never see again. Checked here because every other acquire(...) overload delegates to
+        // this one, so this is the single point every spawn path passes through.
+        if (draining.get()) {
+            throw new ShuttingDownException("fleetd is shutting down; refusing to spawn a session "
+                    + "the shutdown drain would never see");
+        }
         MemberRole memberRole = (role == null) ? MemberRole.DEV : role;
         requireResumeCapability(profile, resumeSessionId);
         // CB-619 / fleetd #123: an explicit profile bypasses placement (CompositePeerLauncher only
@@ -272,10 +303,32 @@ public final class SessionManager implements TurnListener {
      */
     private void release(String paneId, ReleaseCause cause) {
         MemberSession removed = registry.remove(paneId);
+        releaseRemoved(paneId, removed, handles.remove(paneId), cause);
+    }
+
+    /**
+     * Tear a session down only while {@code expected} is still its registry value. A lifecycle
+     * transition replaces the immutable record, so this prevents a reap based on an old READY or
+     * DONE record from stopping a worker that delivery has made BUSY.
+     */
+    private boolean releaseIfCurrent(MemberSession expected, ReleaseCause cause) {
+        if (!registry.remove(expected.paneId(), expected)) {
+            // A lifecycle transition replaced the record between the caller's check and this remove.
+            // Log it: this race is by definition unobservable otherwise, and a reaper that silently
+            // declines to reap is the hardest kind of behaviour to diagnose after the fact.
+            log.debug("skipping reap of pane={}: its registry record changed after the idle check "
+                    + "(most likely a delivery made it BUSY)", expected.paneId());
+            return false;
+        }
+        releaseRemoved(expected.paneId(), expected, handles.remove(expected.paneId()), cause);
+        return true;
+    }
+
+    private void releaseRemoved(String paneId, MemberSession removed, PeerHandle removedHandle,
+                                ReleaseCause cause) {
         // fleetd #209: remove right alongside the registry entry so a released session's handle is
         // never leaked — but keep the local reference below, so the id can still be resolved for
         // the ReleaseDetail this teardown notifies with.
-        PeerHandle removedHandle = handles.remove(paneId);
         boolean preserveWorktree = cause == ReleaseCause.SHUTDOWN;
         String snapshotRef = null;
         if (removed != null) {
@@ -333,6 +386,31 @@ public final class SessionManager implements TurnListener {
         // from the registry with no pane stop is an orphaned pane — a live terminal burning a fleet
         // slot that no longer appears in the roster and can never be reclaimed.
         launcher.stop(paneId);
+        if (removed != null && !preserveWorktree && removed.worktree() != null) {
+            // fleetd #316: the `dirty` read above ran while the worker could still write to this
+            // worktree, so a stale `false` must not be trusted to authorise the --force removal
+            // below. Re-read the worktree's state one more time, right here — immediately before
+            // the one step that would destroy it, and only on the path that is actually about to
+            // do that (invariant 4: no second unconditional `git status` on a release that already
+            // decided to preserve). By now `launcher.stop` has returned, so this read reflects
+            // whatever the worker managed to write up to and including its teardown, not whatever
+            // it had written at release-start time.
+            if (dirtyImmediatelyBeforeRemoval(removed)) {
+                preserveWorktree = true;
+                // The pre-stop snapshot above never ran for this session (the pre-stop read said
+                // clean), so this is the only chance to get the newly-discovered work into
+                // refs/wip/* rather than leaving the on-disk preserve as the sole copy. Best-effort,
+                // like every other snapshot attempt — trySnapshot logs and swallows its own failure.
+                String lateSnapshotRef = trySnapshot(removed, cause);
+                log.warn("release {} preserves worktree {} for pane={} terminal={}: it reported "
+                        + "clean before the pane stopped but dirty immediately before removal — the "
+                        + "worker wrote to it during teardown, and --force removing it now would "
+                        + "have destroyed that work{}",
+                        cause, removed.worktree(), paneId, removed.terminalId(),
+                        lateSnapshotRef == null ? "" : " (snapshotted to refs/wip/" + removed.branch()
+                                + " commit=" + lateSnapshotRef + ")");
+            }
+        }
         if (removed != null && !preserveWorktree && removed.worktree() != null) {
             // fleetd #283: this is the one cleanup step in this method that used to be bare. By the
             // time it runs, the registry entry, the retained handle, and the pane are all already
@@ -351,6 +429,24 @@ public final class SessionManager implements TurnListener {
         }
     }
 
+    /**
+     * fleetd #316: the read that actually authorises {@code worktrees.remove}, taken with the
+     * worker's pane already stopped. Fails toward preserving (returns {@code true}) on any
+     * exception — the same rule the pre-stop check applies (CB-581): once we can no longer tell
+     * whether the worktree is dirty, preserving costs disk while deleting on a guess can destroy
+     * work that has no other copy.
+     */
+    private boolean dirtyImmediatelyBeforeRemoval(MemberSession removed) {
+        try {
+            return worktrees.hasUncommitted(removed.worktree());
+        } catch (RuntimeException e) {
+            log.warn("release could not re-check worktree {} for pane={} terminal={} immediately "
+                    + "before removal; preserving it rather than risk destroying unsaved work: {}",
+                    removed.worktree(), removed.paneId(), removed.terminalId(), e.toString());
+            return true;
+        }
+    }
+
     /**
      * Best-effort snapshot of a dirty worktree into {@code refs/wip/<branch>} (CB-578 stage C). A
      * failure here must never escalate: the caller has already decided to preserve the worktree
@@ -846,14 +942,18 @@ public final class SessionManager implements TurnListener {
             }
             long idleNanos = now - s.lastActivityAtNanos();
             if (idleNanos > idleTtlNanos) {
-                log.debug("reaping idle session terminal={} pane={}: idle {}s exceeds the {}s ttl",
-                        s.terminalId(), s.paneId(), TimeUnit.NANOSECONDS.toSeconds(idleNanos),
-                        TimeUnit.NANOSECONDS.toSeconds(idleTtlNanos));
                 // CB-581: one session that fails to release must not abort the whole reaping pass —
                 // match drainAll's per-session try/catch so the rest of the roster still gets reaped.
                 try {
-                    release(s.paneId());
-                    reaped++;
+                    if (beforeIdleRelease != null) {
+                        beforeIdleRelease.accept(s);
+                    }
+                    if (releaseIfCurrent(s, ReleaseCause.COMPLETED)) {
+                        log.debug("reaping idle session terminal={} pane={}: idle {}s exceeds the {}s ttl",
+                                s.terminalId(), s.paneId(), TimeUnit.NANOSECONDS.toSeconds(idleNanos),
+                                TimeUnit.NANOSECONDS.toSeconds(idleTtlNanos));
+                        reaped++;
+                    }
                 } catch (RuntimeException e) {
                     log.warn("reap failed for pane={} terminal={} worktree={}; continuing with "
                             + "remaining sessions", s.paneId(), s.terminalId(), s.worktree(), e);
@@ -884,10 +984,40 @@ public final class SessionManager implements TurnListener {
      * a reason to delete a worker's only copy of its uncommitted work. A session still {@code BUSY}
      * when the timeout expired is abandoned mid-turn and logged loudly so an operator can find its
      * kept worktree.
+     *
+     * <p>fleetd #308: {@code roster()} is a one-shot snapshot (see its javadoc), and nothing used
+     * to stop a new session from registering after it was taken — {@link #acquire} stayed open for
+     * as long as this drain waited on a {@code BUSY} session, up to the whole {@code timeoutNanos}
+     * budget. Two things close that window, deliberately paired because neither alone is complete:
+     * {@link #draining} is flipped true before the snapshot is even taken, so {@link #acquire}
+     * refuses (invariant 3: loudly, via {@link ShuttingDownException}) as much of the window as a
+     * plain flag can close; and the sweep below re-reads the registry once the initial snapshot has
+     * fully drained and drains whatever a straggler — a caller that read the flag as {@code false}
+     * a moment before it flipped — still managed to register. The sweep shares the same
+     * {@code deadline} rather than getting its own: {@code timeoutNanos} is a budget for the WHOLE
+     * drain (see above), and a straggler must not buy the drain more time than the flag it lost the
+     * race against would have. In the ordinary case the sweep finds nothing and costs one empty
+     * {@link #roster()} call.
      */
     void drainAll(long timeoutNanos) {
         long deadline = System.nanoTime() + timeoutNanos;
-        for (MemberSession s : roster()) {
+        draining.set(true);
+        drainSnapshot(roster(), deadline);
+        List<MemberSession> stragglers = roster();
+        if (!stragglers.isEmpty()) {
+            log.warn("drain sweep found {} session(s) registered after the drain snapshot was "
+                    + "taken (raced past the shutdown guard); draining them too", stragglers.size());
+            drainSnapshot(stragglers, deadline);
+        }
+    }
+
+    /**
+     * Drain exactly the sessions in {@code snapshot}, waiting out a {@code BUSY} one against the
+     * shared whole-drain {@code deadline} before releasing it. Shared by {@link #drainAll}'s main
+     * pass and its post-loop straggler sweep (fleetd #308) so both honor the same one budget.
+     */
+    private void drainSnapshot(List<MemberSession> snapshot, long deadline) {
+        for (MemberSession s : snapshot) {
             try {
                 if (s.state() == MemberSession.State.BUSY) {
                     while (System.nanoTime() < deadline) {

fleetd/src/main/java/dev/ltms/fleet/session/ShuttingDownException.java

@@ -0,0 +1,18 @@
+package dev.ltms.fleet.session;
+
+/**
+ * Thrown by {@link SessionManager#acquire} when a spawn is requested after the daemon's shutdown
+ * drain has already begun (fleetd #308).
+ *
+ * <p>{@link SessionManager#drainAll} snapshots the registry once and tears down exactly what is
+ * in that snapshot. A session registered after the snapshot is invisible to the drain loop: its
+ * pane is left running and its worktree is never preserved, and nothing else ever reclaims
+ * either — the daemon's in-memory registry dies with the process. Refusing the spawn here,
+ * loudly, is what stops that session from ever being created in the first place, rather than
+ * silently handing the caller a session the daemon can no longer manage.
+ */
+public final class ShuttingDownException extends RuntimeException {
+    public ShuttingDownException(String message) {
+        super(message);
+    }
+}

fleetd/src/test/java/dev/ltms/fleet/auth/CallerResolverTest.java

@@ -103,6 +103,54 @@ class CallerResolverTest {
         assertEquals(Role.PRIMARY, p.role(), "the historical behaviour, now an explicit choice");
     }
 
+    // ── fleetd #317: an unresolvable caller must never be promoted to the primary ──────────────────
+    // #305 closed the trigger where a resolved pid matched no pane *and* had no ancestry walk to
+    // save it. This is the other trigger PaneLocator's javadoc names: the pid never resolves at
+    // all — LsofPeerPidLookup returns -1 on any failure, including (silently) "lsof found no
+    // match" — so there is no candidate pid for an ancestry walk to even attempt.
+
+    /**
+     * The failing-without-the-fix case. Before #317's fix, {@code c.terminal() == null} was the
+     * only test in the loopback-trust fallback, and an unresolved pid produces exactly that same
+     * {@code null} terminal as a genuine primary — so this caller was handed
+     * {@code Principal.primary(...)}, a real worker's failed lookup becoming indistinguishable from
+     * the lead.
+     */
+    @Test
+    void aFailedPeerPidLookupIsRefusedNotPromotedToPrimary() {
+        ConnectionIdentity unresolved = new ConnectionIdentity(new PaneLocator(herdr), _ -> -1);
+        Principal p = new CallerResolver(unresolved).resolve("127.0.0.1", 55555, null);
+
+        assertEquals(Role.ANONYMOUS, p.role(),
+                "an unresolvable caller must never be silently promoted to the primary");
+    }
+
+    /**
+     * The companion invariant #317 must not break: a caller whose lookup genuinely succeeded, and
+     * who simply owns no herdr pane — the real primary's own connection — is still the primary.
+     * This is {@link #loopbackTrustTreatsANonWorkerLoopbackCallerAsThePrimary} pinned again here,
+     * named for #317 and placed next to the test it must be distinguished from: same {@code null}
+     * terminal, opposite verdict, because {@code Caller.resolved()} tells them apart.
+     */
+    @Test
+    void aRealPidThatOwnsNoPaneIsStillThePrimaryNotRefused() {
+        Principal p = new CallerResolver(nonWorkerIdentity()).resolve("127.0.0.1", 55555, null);
+
+        assertEquals(Role.PRIMARY, p.role());
+    }
+
+    /** #317 point 4: token mode never consults {@code c.pid()}, so a failed lookup must not change it. */
+    @Test
+    void tokenModeIsUndisturbedByAnUnresolvedLookup() {
+        ConnectionIdentity unresolved = new ConnectionIdentity(new PaneLocator(herdr), _ -> -1);
+        CallerResolver r = new CallerResolver(unresolved, true, "s3cret");
+
+        assertEquals(Role.ANONYMOUS, r.resolve("127.0.0.1", 55555, null).role(),
+                "no credential is still just ANONYMOUS, as before #317 — unchanged by the lookup failing");
+        assertEquals(Role.PRIMARY, r.resolve("127.0.0.1", 55555, "Bearer s3cret").role(),
+                "a valid token still authenticates the primary even though the peer-pid lookup failed");
+    }
+
     @Test
     void tokenModeRefusesANonWorkerCallerThatPresentsNoToken() {
         Principal p = new CallerResolver(nonWorkerIdentity(), true, "s3cret")
@@ -413,4 +461,29 @@ class CallerResolverTest {
         assertThrows(IllegalArgumentException.class, () -> new CallerResolver(id, true, null));
         assertThrows(IllegalArgumentException.class, () -> new CallerResolver(id, true, "  "));
     }
+    @Test
+    void aWorkerOnAnyLoopbackSourceAddressIsStillAWorkerNotThePrimary() {
+        // fleetd #305: the escalation. ConnectionIdentity used to accept only 127.0.0.1, so a
+        // worker connecting from 127.0.0.2 resolved to no terminal, and this resolver's own
+        // (wider) loopback check then made it the PRIMARY — granting spawn, stop, send and drain.
+        // Measured on the Linux fleet host: binding a source of 127.0.0.2 succeeds there, so the
+        // path is real and not theoretical.
+        CallerResolver r = new CallerResolver(workerIdentity(), false, null);
+        for (String src : new String[]{"127.0.0.1", "127.0.0.2", "127.1.2.3", "::ffff:127.0.0.2"}) {
+            Principal p = r.resolve(src, 55555, null);
+            assertEquals(Role.WORKER, p.role(), "a worker must stay a worker from source " + src);
+            assertEquals("term_a", p.terminal(), "worker terminal from source " + src);
+        }
+    }
+
+    @Test
+    void aNonWorkerOnAnyLoopbackSourceAddressIsStillThePrimary() {
+        // The other direction of the same fix: widening the identity check must not demote a
+        // legitimate same-host primary that happens to connect from another 127.* address.
+        CallerResolver r = new CallerResolver(nonWorkerIdentity(), false, null);
+        for (String src : new String[]{"127.0.0.1", "127.0.0.2", "::ffff:127.0.0.1"}) {
+            assertEquals(Role.PRIMARY, r.resolve(src, 55555, null).role(), "source " + src);
+        }
+    }
+
 }

fleetd/src/test/java/dev/ltms/fleet/config/ConfigRefProfileCoverageTest.java

@@ -0,0 +1,202 @@
+package dev.ltms.fleet.config;
+
+import org.junit.jupiter.api.Test;
+
+import java.lang.reflect.Constructor;
+import java.lang.reflect.RecordComponent;
+import java.util.Arrays;
+import java.util.LinkedHashMap;
+import java.util.List;
+import java.util.Map;
+import java.util.Set;
+import java.util.TreeSet;
+
+import static org.junit.jupiter.api.Assertions.assertEquals;
+import static org.junit.jupiter.api.Assertions.assertTrue;
+
+/**
+ * fleetd #323: {@code ConfigRef.sameLaunchSettings} javadoc used to claim it "compares every
+ * component the launcher reads at spawn". It did not — {@code ideProjectDir}, {@code
+ * ideOpenCommand} and {@code autoCompactWindow} were all baked in at daemon startup (see
+ * {@code ClaudeCodeLauncher}/{@code OpenCodeLauncher}) and missing from the comparison, so a reload
+ * that changed only one of them reported "config reloaded" and the running daemon kept the old
+ * value.
+ *
+ * <p>This class is the mechanism the issue asked for: it enumerates every record component of
+ * {@link FleetConfig.Profile} by reflection and proves — by actually mutating a base profile one
+ * field at a time and calling the real method — that each component is either compared by
+ * {@link ConfigRef#sameLaunchSettings} or named in {@link ConfigRef#LAUNCH_SETTINGS_EXCLUDED} with
+ * a reason. A new profile field that is neither fails this test by name, not a hand-maintained list
+ * going stale.
+ */
+class ConfigRefProfileCoverageTest {
+
+    private static final RecordComponent[] COMPONENTS = FleetConfig.Profile.class.getRecordComponents();
+
+    /**
+     * One valid, non-blank value per record component — "the a value". None of these trip any
+     * defaulting/normalization in {@code Profile}'s compact constructor (see {@code
+     * FleetConfig.java}), so what goes in is what {@code sameLaunchSettings} sees back out.
+     */
+    private static final Map<String, Object> BASE = baseValues();
+
+    /** The same shape, each value distinct from {@link #BASE} — "the b value". */
+    private static final Map<String, Object> ALT = altValues();
+
+    private static Map<String, Object> baseValues() {
+        Map<String, Object> v = new LinkedHashMap<>();
+        v.put("profile", "sonnet");
+        v.put("baseUrl", "http://gx00.gw:8000");
+        v.put("model", "sonnet");
+        v.put("configDir", "/config/a");
+        v.put("tokenEnv", "TOKEN_A");
+        v.put("argv", List.of("claude", "--flag-a"));
+        v.put("placement", "tab");
+        v.put("workspace", "workspace-a");
+        v.put("tabLabel", "label-a");
+        v.put("mcpUrl", "http://mcp-a");
+        v.put("cwd", "/cwd/a");
+        v.put("parityOverlay", List.of(".env", ".env.a"));
+        v.put("gitTokenEnv", "GIT_TOKEN_A");
+        v.put("gitHostEnv", "GITEA_HOST_A");
+        v.put("kind", "claude-code");
+        v.put("env", Map.of("K", "A"));
+        v.put("weight", 1.0f);
+        v.put("maxLoad", 5);
+        v.put("subscription", Boolean.TRUE);
+        v.put("exhaustedPattern", "usage limit a");
+        v.put("credentialId", "cred-a");
+        v.put("ideMcpUrl", "http://ide-mcp-a");
+        v.put("ideProjectDir", "modules/a");
+        v.put("ideOpenCommand", "open-cmd-a {dir}");
+        v.put("autoCompactWindow", 150000);
+        v.put("errorPattern", "error a");
+        assertNamesMatchComponents(v);
+        return v;
+    }
+
+    private static Map<String, Object> altValues() {
+        Map<String, Object> v = new LinkedHashMap<>();
+        v.put("profile", "sonnet-b");
+        v.put("baseUrl", "http://gx01.gw:8000");
+        v.put("model", "haiku");
+        v.put("configDir", "/config/b");
+        v.put("tokenEnv", "TOKEN_B");
+        v.put("argv", List.of("claude", "--flag-b"));
+        v.put("placement", "weighted");
+        v.put("workspace", "workspace-b");
+        v.put("tabLabel", "label-b");
+        v.put("mcpUrl", "http://mcp-b");
+        v.put("cwd", "/cwd/b");
+        v.put("parityOverlay", List.of(".env", ".env.b"));
+        v.put("gitTokenEnv", "GIT_TOKEN_B");
+        v.put("gitHostEnv", "GITEA_HOST_B");
+        v.put("kind", "opencode");
+        v.put("env", Map.of("K", "B"));
+        v.put("weight", 2.0f);
+        v.put("maxLoad", 9);
+        v.put("subscription", Boolean.FALSE);
+        v.put("exhaustedPattern", "usage limit b");
+        v.put("credentialId", "cred-b");
+        v.put("ideMcpUrl", "http://ide-mcp-b");
+        v.put("ideProjectDir", "modules/b");
+        v.put("ideOpenCommand", "open-cmd-b {dir}");
+        v.put("autoCompactWindow", 250000);
+        v.put("errorPattern", "error b");
+        assertNamesMatchComponents(v);
+        return v;
+    }
+
+    private static void assertNamesMatchComponents(Map<String, Object> values) {
+        Set<String> componentNames = new TreeSet<>();
+        for (RecordComponent rc : COMPONENTS) {
+            componentNames.add(rc.getName());
+        }
+        assertEquals(componentNames, new TreeSet<>(values.keySet()),
+                "this test's value map has drifted from FleetConfig.Profile's actual components — "
+                        + "update BASE/ALT alongside the record");
+    }
+
+    private static FleetConfig.Profile profileOf(Map<String, Object> values) throws ReflectiveOperationException {
+        Class<?>[] types = Arrays.stream(COMPONENTS).map(RecordComponent::getType).toArray(Class<?>[]::new);
+        Object[] args = Arrays.stream(COMPONENTS)
+                .map(rc -> values.get(rc.getName()))
+                .toArray();
+        Constructor<FleetConfig.Profile> ctor = FleetConfig.Profile.class.getDeclaredConstructor(types);
+        return ctor.newInstance(args);
+    }
+
+    /** {@code BASE} with exactly one named component swapped for its {@code ALT} value. */
+    private static FleetConfig.Profile mutate(String componentName) throws ReflectiveOperationException {
+        Map<String, Object> values = new LinkedHashMap<>(BASE);
+        values.put(componentName, ALT.get(componentName));
+        return profileOf(values);
+    }
+
+    /**
+     * The mechanism fleetd #323 asked for: enumerate {@link FleetConfig.Profile}'s record
+     * components, mutate each non-excluded one, and prove {@code sameLaunchSettings} actually
+     * notices — not just that some hand-maintained list claims it does. Prints the denominator
+     * (total / compared / excluded) the issue required: a checker that cannot state its own
+     * denominator is the failure this repo keeps hitting.
+     */
+    @Test
+    void sameLaunchSettingsComparesEveryProfileComponentOrExcludesIt() throws ReflectiveOperationException {
+        int total = COMPONENTS.length;
+        Set<String> excluded = ConfigRef.LAUNCH_SETTINGS_EXCLUDED;
+
+        Set<String> allNames = new TreeSet<>();
+        for (RecordComponent rc : COMPONENTS) {
+            allNames.add(rc.getName());
+        }
+        assertTrue(allNames.containsAll(excluded),
+                "ConfigRef.LAUNCH_SETTINGS_EXCLUDED names a component that does not exist on "
+                        + "FleetConfig.Profile — check for a typo: " + excluded);
+
+        // The exclusion set is this mechanism's own escape hatch, so it has to be pinned too.
+        // Found by mutation while verifying fleetd #323: moving autoCompactWindow and
+        // ideOpenCommand OUT of the comparison and INTO the exclusion set left the whole suite
+        // green — the loop below simply skips them, and the denominator assertion still balances.
+        // That is exactly the lazy move a failing coverage test invites, and it silently restores
+        // the #323 bug. Only ideProjectDir and worktreeGroup were saved by a behavioural test in
+        // ConfigRefTest; the other two had none. So: growing this set now requires editing this
+        // line as well, which is a visible, deliberate diff rather than a quiet one.
+        assertEquals(Set.of("weight", "maxLoad", "credentialId"), excluded,
+                "ConfigRef.LAUNCH_SETTINGS_EXCLUDED changed. A component belongs in it ONLY if it "
+                        + "is read live off the config supplier, not baked into a launcher at "
+                        + "startup. If you are adding one to silence this test, that is fleetd #323 "
+                        + "happening again: compare it in sameLaunchSettings instead. If it really "
+                        + "is read live, name where it is read and update this assertion.");
+
+        FleetConfig.Profile base = profileOf(BASE);
+        List<String> uncovered = new java.util.ArrayList<>();
+        int compared = 0;
+        for (RecordComponent rc : COMPONENTS) {
+            String name = rc.getName();
+            if (excluded.contains(name)) {
+                continue;
+            }
+            FleetConfig.Profile mutated = mutate(name);
+            if (ConfigRef.sameLaunchSettings(base, mutated)) {
+                uncovered.add(name);
+            } else {
+                compared++;
+            }
+        }
+
+        System.out.printf(
+                "ConfigRef.sameLaunchSettings coverage — %d Profile components total, %d compared, "
+                        + "%d excluded (%s)%n",
+                total, compared, excluded.size(), excluded);
+
+        assertEquals(List.of(), uncovered,
+                "these FleetConfig.Profile components changed but ConfigRef.sameLaunchSettings "
+                        + "reported no difference — add each one to the comparison (it is read at "
+                        + "spawn and baked in until a restart) or to ConfigRef.LAUNCH_SETTINGS_EXCLUDED "
+                        + "with a reason it is genuinely read live: " + uncovered);
+        assertEquals(total, compared + excluded.size(),
+                "every FleetConfig.Profile record component must be either compared or excluded — "
+                        + total + " components, " + compared + " compared, " + excluded.size()
+                        + " excluded");
+    }
+}

fleetd/src/test/java/dev/ltms/fleet/config/ConfigRefTest.java

@@ -434,6 +434,291 @@ class ConfigRefTest {
         assertEquals("provider 5xx", ref.get().profiles().get("sonnet").errorPattern());
     }
 
+    /**
+     * fleetd #323 instance 1: {@code ideProjectDir} is read at spawn off the frozen profile map
+     * (see {@code ClaudeCodeLauncher}/{@code OpenCodeLauncher}) exactly like {@code model}, but was
+     * missing from {@code sameLaunchSettings} — a reload changing only this field used to report a
+     * bare "config reloaded" and the running daemon kept launching with the old value.
+     */
+    @Test
+    void changingAProfilesIdeProjectDirIsReportedAsDeferred(@TempDir Path dir) throws Exception {
+        Path f = dir.resolve("fleetd.yaml");
+        Files.writeString(f, """
+                bind:
+                  host: 127.0.0.1
+                  port: 8765
+                herdrSocket: ~/.config/herdr/herdr.sock
+                profiles:
+                  sonnet:
+                    baseUrl: http://gx00.gw:8000
+                    model: sonnet
+                    ideProjectDir: fleetd
+                guard:
+                  offSubscriptionHosts:
+                    - gx00.gw
+                """);
+        ConfigRef ref = refFor(f);
+
+        Files.writeString(f, """
+                bind:
+                  host: 127.0.0.1
+                  port: 8765
+                herdrSocket: ~/.config/herdr/herdr.sock
+                profiles:
+                  sonnet:
+                    baseUrl: http://gx00.gw:8000
+                    model: sonnet
+                    ideProjectDir: fleetd-renamed
+                guard:
+                  offSubscriptionHosts:
+                    - gx00.gw
+                """);
+        ConfigRef.Outcome out = ref.reload();
+
+        assertTrue(out.applied());
+        assertEquals(1, out.deferred().size(), out.deferred().toString());
+        assertTrue(out.deferred().getFirst().contains("sonnet"), out.deferred().toString());
+        assertTrue(out.deferred().getFirst().contains("launch settings"), out.deferred().toString());
+        // The snapshot still carries the new value — a restart is what makes it take effect.
+        assertEquals("fleetd-renamed", ref.get().profiles().get("sonnet").ideProjectDir());
+    }
+
+    /**
+     * fleetd #323 instance 2: {@code worktreeGroup} is baked into the same {@code GitWorktrees}
+     * as {@code worktreeRoot} (Fleetd.java:251) and never rebuilt, but only {@code worktreeRoot}
+     * was on {@code changedDeferredKeys} — a reload changing only the group reported a bare
+     * "config reloaded" and newly provisioned worktrees kept the old sharing behaviour.
+     */
+    @Test
+    void changingWorktreeGroupIsReportedAsDeferred(@TempDir Path dir) throws Exception {
+        Path f = dir.resolve("fleetd.yaml");
+        Files.writeString(f, yaml("worktreeGroup: devgroup\n"));
+        ConfigRef ref = refFor(f);
+
+        Files.writeString(f, yaml("worktreeGroup: devgroup2\n"));
+        ConfigRef.Outcome out = ref.reload();
+
+        assertTrue(out.applied());
+        assertEquals(java.util.List.of("worktreeGroup"), out.deferred());
+        assertTrue(out.summary().contains("needs a restart") || out.summary().contains("need a restart"),
+                out.summary());
+        // The snapshot still carries the new value — a restart is what makes it take effect.
+        assertEquals("devgroup2", ref.get().worktreeGroup());
+    }
+
+    /**
+     * fleetd #326: {@code primary} is read only off the startup snapshot — {@code Fleetd.java:506,
+     * 519, 520} feed {@code PrimaryRegistry} and {@code ReplyPushLoop} at construction and neither is
+     * rebuilt on reload — but it was missing from {@link ConfigRef#changedDeferredKeys}, so a reload
+     * that only changed the pinned primary terminal reported a bare "config reloaded" while a lead
+     * whose tab no longer matched stayed demoted to worker.
+     */
+    @Test
+    void changingPrimaryIsReportedAsDeferred(@TempDir Path dir) throws Exception {
+        Path f = dir.resolve("fleetd.yaml");
+        Files.writeString(f, yaml("""
+                primary:
+                  terminal: term-a
+                """));
+        ConfigRef ref = refFor(f);
+
+        Files.writeString(f, yaml("""
+                primary:
+                  terminal: term-b
+                """));
+        ConfigRef.Outcome out = ref.reload();
+
+        assertTrue(out.applied());
+        assertEquals(java.util.List.of("primary"), out.deferred());
+        assertTrue(out.summary().contains("need") && out.summary().contains("restart"), out.summary());
+        // The snapshot still carries the new value — a restart is what makes it take effect.
+        assertEquals("term-b", ref.get().primary().terminal());
+    }
+
+    /**
+     * fleetd #326: {@code configReload} itself is read only at startup ({@code Fleetd.java:679-680})
+     * to decide whether to build a {@code ConfigWatcher} at all, and with what interval — the watcher
+     * that would apply a later change is itself built once, so it is deferred rather than cold (see
+     * {@link ConfigRef}'s class doc: cold means an already-open resource would go inconsistent with
+     * the new value, and there is no such resource here — a running watcher just keeps polling on its
+     * original enabled/interval until a restart, exactly like {@code lifecycle} or {@code guard}).
+     * Before this fix, turning reload off (or changing its interval) through a reload reported a bare
+     * "config reloaded" — the obvious joke the issue names.
+     */
+    @Test
+    void changingConfigReloadIsReportedAsDeferred(@TempDir Path dir) throws Exception {
+        Path f = dir.resolve("fleetd.yaml");
+        Files.writeString(f, yaml("""
+                configReload:
+                  enabled: true
+                  intervalSeconds: 10
+                """));
+        ConfigRef ref = refFor(f);
+
+        Files.writeString(f, yaml("""
+                configReload:
+                  enabled: false
+                  intervalSeconds: 30
+                """));
+        ConfigRef.Outcome out = ref.reload();
+
+        assertTrue(out.applied());
+        assertEquals(java.util.List.of("configReload"), out.deferred());
+        assertTrue(out.summary().contains("need") && out.summary().contains("restart"), out.summary());
+        // The snapshot still carries the new value — a restart is what makes it take effect.
+        assertFalse(ref.get().configReload().isEnabled());
+        assertEquals(30, ref.get().configReload().intervalSeconds());
+    }
+
+    /**
+     * fleetd #330: {@code health:} is read both ways — {@code Fleetd.java:556-563} builds the
+     * monitor off the startup snapshot and never rebuilds it, but {@code Fleetd.java:648-650} reads
+     * {@code config.get().health()} live on every {@code fleet_profiles} call. A changed value is
+     * neither purely hot nor purely deferred, so it gets its own {@code split} report naming both
+     * halves rather than a bare "config reloaded" (which would hide the frozen half) or a plain
+     * {@code deferred} entry (which would hide that the coverage string already applied).
+     */
+    @Test
+    void changingHealthIsReportedAsSplit(@TempDir Path dir) throws Exception {
+        Path f = dir.resolve("fleetd.yaml");
+        Files.writeString(f, yaml("""
+                health:
+                  enabled: true
+                  intervalSeconds: 30
+                """));
+        ConfigRef ref = refFor(f);
+
+        Files.writeString(f, yaml("""
+                health:
+                  enabled: true
+                  intervalSeconds: 90
+                """));
+        ConfigRef.Outcome out = ref.reload();
+
+        assertTrue(out.applied());
+        assertTrue(out.deferred().isEmpty(), out.deferred().toString());
+        assertEquals(1, out.split().size(), out.split().toString());
+        assertTrue(out.split().getFirst().startsWith("health:"), out.split().toString());
+        assertTrue(out.split().getFirst().contains("restart"), out.split().toString());
+        assertTrue(out.split().getFirst().contains("live"), out.split().toString());
+        assertTrue(out.summary().contains("partially live"), out.summary());
+        // The snapshot still carries the new value — the monitor itself is what waits for a restart.
+        assertEquals(90, ref.get().health().intervalSeconds());
+    }
+
+    /**
+     * fleetd #330: {@code coordinator:} is the other split key — {@code Fleetd.java:502} opens the
+     * {@code LeadMailbox} off the startup snapshot and never reopens it, but
+     * {@code HerdrPeerLauncher.java:1530} reads {@code config.get().coordinator()} live on every
+     * spawn to keep the broker URI env-var name out of a member's environment.
+     */
+    @Test
+    void changingCoordinatorIsReportedAsSplit(@TempDir Path dir) throws Exception {
+        Path f = dir.resolve("fleetd.yaml");
+        Files.writeString(f, yaml("""
+                coordinator:
+                  selfId: mac-a
+                """));
+        ConfigRef ref = refFor(f);
+
+        Files.writeString(f, yaml("""
+                coordinator:
+                  selfId: mac-b
+                """));
+        ConfigRef.Outcome out = ref.reload();
+
+        assertTrue(out.applied());
+        assertTrue(out.deferred().isEmpty(), out.deferred().toString());
+        assertEquals(1, out.split().size(), out.split().toString());
+        assertTrue(out.split().getFirst().startsWith("coordinator:"), out.split().toString());
+        assertTrue(out.split().getFirst().contains("restart"), out.split().toString());
+        assertTrue(out.split().getFirst().contains("live"), out.split().toString());
+        // The snapshot still carries the new value — the LeadMailbox connection is what waits for a
+        // restart; selfId names this daemon's own inbox queue and a peer cannot discover a rename.
+        assertEquals("mac-b", ref.get().coordinator().selfId());
+    }
+
+    /**
+     * A split change must not refuse the reload (invariant 2 of fleetd #330) and
+     * {@code Outcome.applied()} must stay {@code true} (invariant 3) — unlike a cold change, the
+     * live half of a split key genuinely took effect, so refusing would throw that away.
+     */
+    @Test
+    void aSplitChangeDoesNotRefuseTheReload(@TempDir Path dir) throws Exception {
+        Path f = dir.resolve("fleetd.yaml");
+        Files.writeString(f, yaml("coordinator:\n  selfId: mac-a\n"));
+        ConfigRef ref = refFor(f);
+
+        Files.writeString(f, yaml("coordinator:\n  selfId: mac-b\n"));
+        ConfigRef.Outcome out = ref.reload();
+
+        assertTrue(out.applied());
+        assertTrue(out.error() == null);
+        assertTrue(out.coldKeys().isEmpty());
+    }
+
+    /**
+     * Both split keys can change in one reload — the report names both, and a caller reading
+     * {@code split} does not have to guess which half of which key already applied.
+     */
+    @Test
+    void changingBothSplitKeysReportsBoth(@TempDir Path dir) throws Exception {
+        Path f = dir.resolve("fleetd.yaml");
+        Files.writeString(f, yaml("""
+                health:
+                  enabled: true
+                coordinator:
+                  selfId: mac-a
+                """));
+        ConfigRef ref = refFor(f);
+
+        Files.writeString(f, yaml("""
+                health:
+                  enabled: false
+                coordinator:
+                  selfId: mac-b
+                """));
+        ConfigRef.Outcome out = ref.reload();
+
+        assertTrue(out.applied());
+        assertEquals(2, out.split().size(), out.split().toString());
+        assertTrue(out.split().stream().anyMatch(s -> s.startsWith("health:")), out.split().toString());
+        assertTrue(out.split().stream().anyMatch(s -> s.startsWith("coordinator:")), out.split().toString());
+    }
+
+    /**
+     * A split key and a deferred key changing in the same reload must both show up, each in its own
+     * list — proving the two fields do not step on each other and {@link ConfigRef.Outcome#summary()}
+     * reports both halves of the message.
+     */
+    @Test
+    void aSplitChangeAndADeferredChangeCoexist(@TempDir Path dir) throws Exception {
+        Path f = dir.resolve("fleetd.yaml");
+        Files.writeString(f, yaml("""
+                coordinator:
+                  selfId: mac-a
+                lifecycle:
+                  drainTimeoutSeconds: 30
+                """));
+        ConfigRef ref = refFor(f);
+
+        Files.writeString(f, yaml("""
+                coordinator:
+                  selfId: mac-b
+                lifecycle:
+                  drainTimeoutSeconds: 60
+                """));
+        ConfigRef.Outcome out = ref.reload();
+
+        assertTrue(out.applied());
+        assertEquals(java.util.List.of("lifecycle"), out.deferred());
+        assertEquals(1, out.split().size(), out.split().toString());
+        assertTrue(out.split().getFirst().startsWith("coordinator:"), out.split().toString());
+        assertTrue(out.summary().contains("need a restart") || out.summary().contains("needs a restart"),
+                out.summary());
+        assertTrue(out.summary().contains("partially live"), out.summary());
+    }
+
     @Test
     void aFixedRefHasNoFileAndRefusesToReload() {
         FleetConfig cfg = new FleetConfig(null, null, null, null, null, null,

fleetd/src/test/java/dev/ltms/fleet/config/ConfigRefTopLevelCoverageTest.java

@@ -0,0 +1,167 @@
+package dev.ltms.fleet.config;
+
+import org.junit.jupiter.api.Test;
+
+import java.lang.reflect.RecordComponent;
+import java.util.LinkedHashSet;
+import java.util.List;
+import java.util.Set;
+import java.util.TreeSet;
+
+import static org.junit.jupiter.api.Assertions.assertEquals;
+import static org.junit.jupiter.api.Assertions.assertTrue;
+
+/**
+ * fleetd #330: a new top-level {@link FleetConfig} record component must be triaged into a reload
+ * class before it ships, or it repeats fleetd #323 ({@code worktreeGroup} missing from {@code
+ * ConfigRef.changedDeferredKeys}) and fleetd #326 ({@code primary}/{@code configReload} missing the
+ * same way) — a key silently absent from {@link ConfigRef}'s reload machinery, so a reload changing
+ * only that key reports a bare "config reloaded" for a change the running daemon never picked up.
+ *
+ * <p>This is the top-level counterpart of {@link ConfigRefProfileCoverageTest}: instead of
+ * enumerating {@link FleetConfig.Profile}'s record components, it enumerates {@link FleetConfig}'s
+ * own — {@code bind}, {@code health}, {@code memberCredentials}, and so on — and requires each to
+ * fall into exactly one of four homes: {@link ConfigRef#COLD_KEYS}, {@link #DEFERRED_TOP_LEVEL_KEYS}
+ * (compared in {@code ConfigRef.changedDeferredKeys}), {@link ConfigRef#SPLIT_KEYS}, or
+ * {@link #HOT_EXCLUDED_TOP_LEVEL_KEYS} (the escape hatch: read live off the config supplier, so no
+ * reload bookkeeping is needed for it at all).
+ *
+ * <h2>What this checker can and cannot prove</h2>
+ * It proves the record's <em>shape</em> is fully triaged: every one of {@code FleetConfig}'s
+ * components sits in exactly one of the four sets, none sits in two, and the escape hatch
+ * ({@link #HOT_EXCLUDED_TOP_LEVEL_KEYS}) cannot silently grow without a visible diff to this file.
+ * That is what "a new component cannot be added without someone triaging it" means in practice.
+ *
+ * <p>It CANNOT prove that any of the citations are <em>true</em>. "Compared in {@code
+ * changedDeferredKeys}" and "read live off {@code config.get()}" are facts about {@code
+ * ConfigRef.java}, {@code Fleetd.java} and {@code HerdrPeerLauncher.java} that a reflection-only
+ * test over {@code FleetConfig}'s shape has no way to inspect — this test would pass identically
+ * whether or not the cited line still does what the comment next to it says. Trust the citation
+ * because a person read the source (the exact call sites are named next to each set below), not
+ * because this test is green. {@link ConfigRefTest} is what behaviourally proves the deferred and
+ * split keys it covers actually get reported; {@link ConfigRefProfileCoverageTest} does the same,
+ * behaviourally, for {@code FleetConfig.Profile}'s own fields.
+ */
+class ConfigRefTopLevelCoverageTest {
+
+    private static final RecordComponent[] COMPONENTS = FleetConfig.class.getRecordComponents();
+
+    /**
+     * Top-level components whose change {@code ConfigRef.changedDeferredKeys} reads and reports on
+     * — verified by reading that method as of fleetd #330, not derived from this test.
+     * {@code spawnReadyTimeoutMs}/{@code spawnReadyPollMs} are compared together and reported under
+     * one combined label ({@code "spawnReady*"}); {@code profiles} is compared twice over — once for
+     * added/removed profile names, once for an existing profile's launch settings — and that second
+     * comparison excludes {@code weight}/{@code maxLoad}/{@code credentialId} as hot sub-fields,
+     * which is what {@link ConfigRefProfileCoverageTest} exists to keep honest at the sub-field
+     * level. {@code profiles} itself still belongs here, not in the hot-exclusion set below: most of
+     * a profile's fields are NOT read live, so citing "read live off the config supplier" for the
+     * whole top-level key would be false.
+     */
+    private static final Set<String> DEFERRED_TOP_LEVEL_KEYS = Set.of(
+            "guard", "worktreeRoot", "worktreeGroup", "primary", "configReload",
+            "leadHeartbeat", "lifecycle", "spawnReadyTimeoutMs", "spawnReadyPollMs",
+            "quarantineCooldownSeconds", "profiles");
+
+    /**
+     * The escape hatch: top-level components with no reload bookkeeping at all, because every read
+     * of them goes live through {@link ConfigRef#get()} rather than off a startup snapshot. A
+     * component belongs here ONLY if that is true — never because adding it here makes this test
+     * pass. fleetd #323 is the cautionary tale for exactly this pattern: the identical hatch on
+     * {@code ConfigRef.LAUNCH_SETTINGS_EXCLUDED} let two profile fields be silently re-broken with
+     * the whole suite green, and it was only caught by mutating the checker itself (see this class's
+     * own mutation test below, and {@code ConfigRefProfileCoverageTest}'s equivalent).
+     *
+     * <ul>
+     *   <li>{@code placement} — read live by the placement policy on every spawn (class doc, Hot
+     *       bullet; {@code ConfigRefTest.aConsumerHoldingTheRefSeesTheNewValue} proves it
+     *       behaviourally).</li>
+     *   <li>{@code fleet} — role pools, {@code charters} and {@code tabLabel} are read live through
+     *       the supplier on {@code CompositePeerLauncher} (class doc, Hot bullet). The one
+     *       documented exception, {@code fleet.leaders}, is read only at startup and genuinely needs
+     *       a restart — a real gap in {@code changedDeferredKeys}, but one the class doc already
+     *       carries and that fleetd #330 explicitly did not re-open (its "seven readers" fact-find
+     *       named {@code health}/{@code coordinator} as the complete set of split-shaped keys, not
+     *       {@code fleet}). Reported as a caveat, not fixed here.</li>
+     *   <li>{@code memberCredentials} — read live at {@code Fleetd.java:198, 205, 729}.</li>
+     *   <li>{@code memberLoginShell} — read live at
+     *       {@code HerdrPeerLauncher#configuredMemberLoginShell}.</li>
+     * </ul>
+     */
+    private static final Set<String> HOT_EXCLUDED_TOP_LEVEL_KEYS =
+            Set.of("placement", "fleet", "memberCredentials", "memberLoginShell");
+
+    @Test
+    void everyTopLevelComponentIsAccountedForInExactlyOneClass() {
+        Set<String> allNames = new TreeSet<>();
+        for (RecordComponent rc : COMPONENTS) {
+            allNames.add(rc.getName());
+        }
+
+        Set<String> cold = ConfigRef.COLD_KEYS;
+        Set<String> split = ConfigRef.SPLIT_KEYS;
+        Set<String> deferred = DEFERRED_TOP_LEVEL_KEYS;
+        Set<String> hot = HOT_EXCLUDED_TOP_LEVEL_KEYS;
+
+        // Typo guard on each set — the same check ConfigRefProfileCoverageTest runs on
+        // LAUNCH_SETTINGS_EXCLUDED. A name that does not exist on FleetConfig is a silent no-op.
+        assertTrue(allNames.containsAll(cold),
+                "ConfigRef.COLD_KEYS names a component that does not exist on FleetConfig: " + cold);
+        assertTrue(allNames.containsAll(split),
+                "ConfigRef.SPLIT_KEYS names a component that does not exist on FleetConfig: " + split);
+        assertTrue(allNames.containsAll(deferred),
+                "DEFERRED_TOP_LEVEL_KEYS names a component that does not exist on FleetConfig: " + deferred);
+        assertTrue(allNames.containsAll(hot),
+                "HOT_EXCLUDED_TOP_LEVEL_KEYS names a component that does not exist on FleetConfig: " + hot);
+
+        // The escape hatch is pinned. Growing it requires editing this line — a visible, deliberate
+        // diff, not a quiet one. See the field javadoc above for what "belongs here" actually means.
+        assertEquals(Set.of("placement", "fleet", "memberCredentials", "memberLoginShell"), hot,
+                "HOT_EXCLUDED_TOP_LEVEL_KEYS changed. A component belongs here ONLY if it is read "
+                        + "live off the config supplier, never because adding it makes this test "
+                        + "pass. If you are adding one to silence this test, that is fleetd #323 "
+                        + "happening again: account for it in ConfigRef.changedDeferredKeys (or "
+                        + "COLD_KEYS/SPLIT_KEYS) instead. If it really is read live, name where and "
+                        + "update this assertion and the field javadoc together.");
+
+        // No component may sit in two buckets at once — the denominator check below could not catch
+        // that on its own (two buckets double-booking one key still sums to the right total if
+        // another key is simultaneously missing), so check every pair directly and name the culprit.
+        record Bucket(String name, Set<String> keys) {}
+        List<Bucket> buckets = List.of(
+                new Bucket("COLD_KEYS", cold), new Bucket("SPLIT_KEYS", split),
+                new Bucket("DEFERRED_TOP_LEVEL_KEYS", deferred), new Bucket("HOT_EXCLUDED_TOP_LEVEL_KEYS", hot));
+        for (int i = 0; i < buckets.size(); i++) {
+            for (int j = i + 1; j < buckets.size(); j++) {
+                Set<String> overlap = new LinkedHashSet<>(buckets.get(i).keys());
+                overlap.retainAll(buckets.get(j).keys());
+                assertEquals(Set.of(), overlap, "a component is in both " + buckets.get(i).name()
+                        + " and " + buckets.get(j).name() + ": " + overlap);
+            }
+        }
+
+        Set<String> union = new TreeSet<>();
+        union.addAll(cold);
+        union.addAll(split);
+        union.addAll(deferred);
+        union.addAll(hot);
+
+        System.out.printf(
+                "FleetConfig top-level coverage — %d components total: %d cold %s, %d deferred %s, "
+                        + "%d split %s, %d hot-excluded %s%n",
+                allNames.size(), cold.size(), cold, deferred.size(), deferred, split.size(), split,
+                hot.size(), hot);
+
+        Set<String> missing = new TreeSet<>(allNames);
+        missing.removeAll(union);
+        assertEquals(Set.of(), missing,
+                "these FleetConfig components are in none of COLD_KEYS, DEFERRED_TOP_LEVEL_KEYS, "
+                        + "SPLIT_KEYS or HOT_EXCLUDED_TOP_LEVEL_KEYS — triage each one into whichever "
+                        + "actually describes it: " + missing);
+        assertEquals(allNames.size(), cold.size() + deferred.size() + split.size() + hot.size(),
+                "counts don't sum to the component total even though every component was found in "
+                        + "the union — " + allNames.size() + " components, " + cold.size()
+                        + " cold + " + deferred.size() + " deferred + " + split.size() + " split + "
+                        + hot.size() + " hot-excluded");
+    }
+}

fleetd/src/test/java/dev/ltms/fleet/inject/InjectorTest.java

@@ -297,6 +297,68 @@ class InjectorTest {
         assertTrue(inj.activeTargets().isEmpty(), "the wedged target is reclaimed, not polled forever");
     }
 
+    /** A listener whose post-turn housekeeping always starts, as SessionManager's does with clearAfterTurn on. */
+    private static final class PostTurnListener implements TurnListener {
+        @Override public void onTurnComplete(String target) { }
+        @Override public boolean hasPostTurnAction(String target) { return true; }
+        @Override public boolean onTurnCompleteWithPostAction(String target) { return true; }
+    }
+
+    @Test
+    void aWorkerThatWedgesInUnknownAwaitingPostTurnPickupIsReleased() {
+        // fleetd #306: the post-turn phase had no way out of a sustained unknown streak, so the
+        // pickup latch stayed set, the target was polled forever, and every later message to it was
+        // blocked by the delivery gate — while the session still looked healthy.
+        Captor cap = new Captor();
+        Injector inj = new Injector(new AgentControl(herdr), new PostTurnListener());
+        inj.enqueue(T, "task", TestTurnTokens.inert(T));
+
+        inj.onStatus(T, AgentStatus.IDLE);    // deliver
+        inj.onStatus(T, AgentStatus.WORKING); // turn starts
+        inj.onStatus(T, AgentStatus.IDLE);    // turn completes; housekeeping dispatched
+        for (int i = 0; i < STALL_SAMPLES; i++) inj.onStatus(T, AgentStatus.UNKNOWN); // then wedges
+
+        assertTrue(inj.activeTargets().isEmpty(),
+                "a target wedged awaiting post-turn pickup must be reclaimed, not polled forever");
+        assertEquals(List.of(), cap.failed,
+                "the delegated turn already completed — a stuck /clear must not be reported as a failed turn");
+    }
+
+    @Test
+    void aWorkerThatWedgesInUnknownAfterPickingUpTheResetIsReleased() {
+        // The sibling latch. postTurnObserved is set when the reset is seen picked up (WORKING) and
+        // is cleared only on a later injectable sample, so a wedge right after pickup sticks too.
+        Captor cap = new Captor();
+        Injector inj = new Injector(new AgentControl(herdr), new PostTurnListener());
+        inj.enqueue(T, "task", TestTurnTokens.inert(T));
+
+        inj.onStatus(T, AgentStatus.IDLE);
+        inj.onStatus(T, AgentStatus.WORKING);
+        inj.onStatus(T, AgentStatus.IDLE);    // turn complete; reset dispatched
+        inj.onStatus(T, AgentStatus.WORKING); // reset picked up -> postTurnObserved
+        for (int i = 0; i < STALL_SAMPLES; i++) inj.onStatus(T, AgentStatus.UNKNOWN);
+
+        assertTrue(inj.activeTargets().isEmpty(), "a wedge after reset pickup must also be reclaimed");
+        assertEquals(List.of(), cap.failed, "still not a turn failure");
+    }
+
+    @Test
+    void aBriefUnknownDuringPostTurnHousekeepingDoesNotDropTheLatch() {
+        // The other direction: the escape must not fire on a glitch, or the queued next delegation
+        // would overtake housekeeping that is still running.
+        Injector inj = new Injector(new AgentControl(herdr), new PostTurnListener());
+        inj.enqueue(T, "first", TestTurnTokens.inert(T));
+        inj.enqueue(T, "second", TestTurnTokens.inert(T));
+
+        inj.onStatus(T, AgentStatus.IDLE);
+        inj.onStatus(T, AgentStatus.WORKING);
+        inj.onStatus(T, AgentStatus.IDLE);    // first completes; reset dispatched
+        for (int i = 0; i < 10; i++) inj.onStatus(T, AgentStatus.UNKNOWN); // well under the grace
+
+        assertFalse(inj.activeTargets().isEmpty(), "a brief glitch must not release the post-turn latch");
+        assertEquals(List.of("first"), sent(), "the queued delegation must not overtake housekeeping");
+    }
+
     @Test
     void aTransientUnknownGlitchNeitherFailsNorBlocksCompletion() {
         Captor cap = new Captor();

fleetd/src/test/java/dev/ltms/fleet/mcp/ConnectionIdentityTest.java

@@ -20,6 +20,17 @@ class ConnectionIdentityTest {
         assertEquals("term_a", with(_ -> FakeHerdr.WORKER_PID).callerTerminal("127.0.0.1", 55555));
     }
 
+    @Test
+    void resolvesWorkerFromAnyLoopbackSourceAddressNotJust127001() {
+        // fleetd #305. On Linux the whole 127.0.0.0/8 is bound to lo, so a worker can connect with
+        // a source address of 127.0.0.2. If identity resolution skips that address the caller has
+        // no terminal, and a caller with no terminal is the primary under loopback-trust — so this
+        // must resolve the worker, not null.
+        assertEquals("term_a", with(_ -> FakeHerdr.WORKER_PID).callerTerminal("127.0.0.2", 55555));
+        assertEquals("term_a", with(_ -> FakeHerdr.WORKER_PID).callerTerminal("127.1.2.3", 55555));
+        assertEquals("term_a", with(_ -> FakeHerdr.WORKER_PID).callerTerminal("::ffff:127.0.0.2", 55555));
+    }
+
     @Test
     void nullForOffHostCaller() {
         // A non-loopback peer can't be an on-host worker → treat as primary/unknown.
@@ -32,6 +43,22 @@ class ConnectionIdentityTest {
         assertNull(with(_ -> 999_999).callerTerminal("127.0.0.1", 55555));
     }
 
+    @Test
+    void callerIsUnresolvedWhenThePeerPidLookupFails() {
+        // fleetd #317: LsofPeerPidLookup returns -1 on any failure — a fork error, or (silently)
+        // simply no matching lsof line. Caller.resolved() is the one place that sentinel is tested.
+        ConnectionIdentity.Caller c = with(_ -> -1).resolve("127.0.0.1", 55555);
+        assertFalse(c.resolved(), "a -1 pid means the lookup failed, not that this pid owns no pane");
+    }
+
+    @Test
+    void callerIsResolvedWhenThePidIsRealEvenThoughItOwnsNoPane() {
+        // The primary's own connection: a real, lsof-found pid that just isn't a worker pane. This
+        // must read as "resolved" — the distinction #317 turns on.
+        ConnectionIdentity.Caller c = with(_ -> 999_999).resolve("127.0.0.1", 55555);
+        assertTrue(c.resolved());
+    }
+
     @Test
     void resolvesTheCallersPidAndCwd() {
         // CB-112: the primary maps to no pane, but its PID and cwd are still readable.

fleetd/src/test/java/dev/ltms/fleet/mcp/FleetMcpTest.java

@@ -189,7 +189,7 @@ class FleetMcpTest {
     }
 
     @Test
-    void unansweredAsyncAskReturnsTheTicketToPending() throws Exception {
+    void unansweredAsyncAskReturnsTheTicketToPendingThenAWorkersLateReplyStillCompletesIt() throws Exception {
         McpSchema.CallToolResult accepted = FleetMcp.sendAsync(messages, "term_a", "do it", null, Set.of());
         String ticket = textOf(accepted).substring(textOf(accepted).indexOf("ticket=") + "ticket=".length()).trim();
 
@@ -203,8 +203,15 @@ class FleetMcpTest {
         assertTrue(textOf(ask).contains("no answer"), textOf(ask));
         assertTrue(textOf(FleetMcp.poll(messages, ticket, null)).startsWith("[pending"));
 
+        // fleetd #307: the worker resumed on its own after the primary never answered, and its real
+        // fleet_reply must complete its OWN async ticket — not strand in the inbox with
+        // fleet_poll{ticket} stuck PENDING forever and later force-failed with a false "session
+        // released before it replied" reason. This used to land in the inbox instead (see the old
+        // assertion this replaced: messages.drainReplies("term_a").getFirst()...) — that was the bug.
         FleetMcp.reply(messages, "term_a", "finished after timeout");
-        assertEquals("finished after timeout", messages.drainReplies("term_a").getFirst().content());
+        assertEquals("finished after timeout", textOf(FleetMcp.poll(messages, ticket, null)));
+        assertTrue(messages.drainReplies("term_a").isEmpty(),
+                "the reply completed its own ticket directly and never touched the inbox");
     }
 
     @Test
@@ -326,6 +333,26 @@ class FleetMcpTest {
         assertEquals("orphan", drained.getFirst().content());
     }
 
+    @Test
+    void replyWithBlankContentIsACleanToolErrorNotAnUncaughtException() {
+        // fleetd #302: MessageService.reply now REJECTS blank content by throwing. fleet_reply's
+        // handler is a bare BiFunction with no try/catch around it, so if this guard only checked
+        // `== null` (as it did), a whitespace-only reply would leave the handler as an uncaught
+        // IllegalArgumentException instead of a tool error the caller can read. Null and whitespace
+        // are the same caller mistake and must get the same answer — the sibling fleet_send guard
+        // has always used isBlank for exactly this reason.
+        for (String blank : new String[] {null, "", "   ", "\n\t"}) {
+            McpSchema.CallToolResult res = assertDoesNotThrow(
+                    () -> FleetMcp.reply(messages, "term_a", blank),
+                    "blank content must be refused as a tool error, never thrown out of the handler");
+            assertEquals(Boolean.TRUE, res.isError(), "blank content is an error result");
+            assertTrue(textOf(res).contains("content is required"),
+                    "the error names the missing argument: " + textOf(res));
+        }
+        assertEquals(0, messages.drainReplies("term_a").size(),
+                "a refused reply must not reach the inbox");
+    }
+
     @Test
     void bridgePollWithTargetDrainsReplies() {
         // A reply with no open send queues it in the inbox.

fleetd/src/test/java/dev/ltms/fleet/member/CompositePeerLauncherTest.java

@@ -615,6 +615,67 @@ class CompositePeerLauncherTest {
         assertEquals(1, adapter.spawnCount("b"));
     }
 
+    /**
+     * fleetd #315: {@code CompositePeerLauncher.spawn} rebuilds the {@link PlacementContext} after
+     * every failed attempt "so the policy excludes this profile" (see the comment at the retry call
+     * site) — but {@code FixedPlacementPolicy} never read {@code ctx.unreachable()}, so under the
+     * default {@code fixed} placement every retry re-picked the same dead default and a second,
+     * healthy, configured profile was never tried. This is the same scenario as
+     * {@link #failoverRetriesNextCandidateWhenProfileIsUnreachable}, but pinned to {@code fixed()}
+     * instead of {@code weighted()} — the three existing failover tests all use {@code weighted()},
+     * which is exactly why nobody caught this: the retry loop's contract has no coverage under its
+     * own default policy.
+     */
+    @Test
+    void failoverRetriesNextCandidateUnderFixedPlacementWhenProfileIsUnreachable() {
+        FakeHerdr herdr = new FakeHerdr();
+        Map<String, FleetConfig.Profile> profiles = ordered(
+                "a", stubWorker("a"),
+                "b", stubWorker("b"));
+        StubLauncher adapter = new StubLauncher("claude", herdr, profiles, "a", Set.of("a"));
+        CompositePeerLauncher composite = new CompositePeerLauncher(
+                List.of(adapter), "a", profiles, PlacementPolicies.fixed(), _ -> 0);
+
+        PeerHandle h = composite.spawn(new SpawnRequest(null, null, null));
+        assertEquals("b", h.profile(),
+                "fixed placement must fail over from the unreachable default a to the healthy b");
+        assertEquals(1, adapter.spawnCount("a"), "a was tried once and failed");
+        assertEquals(1, adapter.spawnCount("b"), "b was tried once and succeeded");
+    }
+
+    /**
+     * fleetd #315: the same fix — {@code FixedPlacementPolicy} consulting {@code ctx.unreachable()}
+     * — also covers the wiring-bug branch in {@code CompositePeerLauncher.spawn}: a profile that
+     * placement is allowed to choose (it is in the configured candidate list) but that no delegate
+     * declares ({@code byProfile.get(chosen.profile()) == null}). That branch adds the profile to
+     * {@code unreachable} and {@code continue}s without ever calling a launcher, so before this fix
+     * {@code fixed} handed back the same adapterless profile on every remaining attempt too.
+     */
+    @Test
+    void failoverSkipsAConfiguredProfileNoAdapterDeclaresUnderFixedPlacement() {
+        FakeHerdr herdr = new FakeHerdr();
+        // Placement's candidate list has three profiles, in this order (LinkedHashMap preserves it,
+        // and the fixed default resolves to the first — see the `ordered` helper's own javadoc).
+        Map<String, FleetConfig.Profile> profiles = new LinkedHashMap<>();
+        profiles.put("c", stubWorker("c"));
+        profiles.put("a", stubWorker("a"));
+        profiles.put("b", stubWorker("b"));
+        // The adapter only declares a and b — c is a configured profile with no owning adapter,
+        // the "wiring bug" the comment in CompositePeerLauncher.spawn calls out.
+        Map<String, FleetConfig.Profile> adapterProfiles = new LinkedHashMap<>();
+        adapterProfiles.put("a", stubWorker("a"));
+        adapterProfiles.put("b", stubWorker("b"));
+        StubLauncher adapter = new StubLauncher("claude", herdr, adapterProfiles, "a", Set.of());
+        CompositePeerLauncher composite = new CompositePeerLauncher(
+                List.of(adapter), "a", profiles, PlacementPolicies.fixed(), _ -> 0);
+
+        PeerHandle h = composite.spawn(new SpawnRequest(null, null, null));
+        assertEquals("a", h.profile(),
+                "c has no adapter, so fixed placement must skip it and land on the next candidate, a");
+        assertEquals(0, adapter.spawnCount("c"), "c is never spawned — no adapter owns it");
+        assertEquals(1, adapter.spawnCount("a"));
+    }
+
     @Test
     void explicitSpawnAtMaxLoadThrowsPlacementExceptionNamingProfileLiveAndCap() {
         FakeHerdr herdr = new FakeHerdr();

fleetd/src/test/java/dev/ltms/fleet/msg/AmqpReplyInboxReleaseRaceTest.java

@@ -0,0 +1,264 @@
+package dev.ltms.fleet.msg;
+
+import com.rabbitmq.client.AMQP;
+import com.rabbitmq.client.Channel;
+import com.rabbitmq.client.Connection;
+import com.rabbitmq.client.DeliverCallback;
+import com.rabbitmq.client.Delivery;
+import com.rabbitmq.client.Envelope;
+import org.junit.jupiter.api.Test;
+import org.junit.jupiter.api.Timeout;
+
+import java.lang.reflect.InvocationHandler;
+import java.lang.reflect.Proxy;
+import java.nio.charset.StandardCharsets;
+import java.time.Duration;
+import java.util.List;
+import java.util.concurrent.CopyOnWriteArrayList;
+import java.util.concurrent.CountDownLatch;
+import java.util.concurrent.TimeUnit;
+import java.util.concurrent.atomic.AtomicInteger;
+import java.util.concurrent.atomic.AtomicReference;
+
+import static org.junit.jupiter.api.Assertions.assertEquals;
+import static org.junit.jupiter.api.Assertions.assertNull;
+import static org.junit.jupiter.api.Assertions.assertTrue;
+
+/**
+ * CB-318: a delivery landing on the consumer work-pool thread <em>after</em>
+ * {@link AmqpReplyInbox#release} has already swapped the target's {@code held} entry for its
+ * tombstone, but <em>before</em> {@code release()} itself returns, must be nacked-with-requeue —
+ * never silently retained in a fresh map {@code release()} has already stopped looking at.
+ *
+ * <p><strong>This forces the actual interleaving, not a sequence of calls.</strong> {@code release()}
+ * runs on its own thread and is made to block <em>inside</em> its nack loop, via a fake
+ * {@link Channel} whose {@code basicNack} blocks on its first invocation. That block is only
+ * reachable after {@code release()}'s {@code held.compute(...)} has already swapped in the
+ * {@code RELEASED} tombstone (the compute call happens strictly before the loop that calls
+ * {@code basicNack}), so observing it is direct, ordering-guaranteed proof that the tombstone is in
+ * place and {@code release()} has not yet returned — still holding {@code channelLock} — when a
+ * second thread fires {@code own()}'s captured {@link DeliverCallback} for a brand-new message on the
+ * same target. No mocking library is on the classpath, so the fake broker is a {@link Proxy}, the
+ * same pattern {@code AmqpReplyInboxRecoveryRaceTest} already uses.
+ *
+ * <p><strong>What this does and does not prove.</strong> It proves that a delivery whose
+ * {@code computeIfAbsent} call is ordered strictly after {@code release()}'s tombstone swap — while
+ * {@code release()} is still running — is nacked-with-requeue rather than silently parked forever.
+ * It does not drive a real broker: {@code basicNack} here is a recorded call on a fake channel, not a
+ * verified requeue-and-redeliver. That half of the contract (a nacked-with-requeue delivery really
+ * does come back to a later owner) is already covered against a real broker by
+ * {@code AmqpReplyInboxContractTest.releaseCancelsConsumerAndRequeuesHeldDeliveryForRecovery}, which
+ * this test does not duplicate.
+ */
+class AmqpReplyInboxReleaseRaceTest {
+
+    @Test
+    @Timeout(15)
+    void deliveryArrivingWhileReleaseIsStillRunningIsNackedNotStranded() throws Exception {
+        String target = "worker-release-race";
+
+        List<long[]> nacks = new CopyOnWriteArrayList<>();   // {deliveryTag, requeue(1/0)}
+        List<Long> acks = new CopyOnWriteArrayList<>();
+        AtomicReference<DeliverCallback> deliverCallback = new AtomicReference<>();
+        CountDownLatch nackStarted = new CountDownLatch(1);
+        CountDownLatch releaseMayFinishNack = new CountDownLatch(1);
+        AtomicInteger nackCallCount = new AtomicInteger();
+
+        Channel consumeChannel = fakeConsumeChannel(deliverCallback, nacks, acks, nackCallCount,
+                nackStarted, releaseMayFinishNack);
+        Channel publishChannel = fakeInertChannel();
+        Connection connection = fakeConnection(consumeChannel, publishChannel);
+
+        AmqpReplyInbox inbox = new AmqpReplyInbox(connection, AmqpReplyInbox.DEFAULT_PREFETCH);
+        inbox.own(target);
+        assertTrue(deliverCallback.get() != null, "own() must have registered a DeliverCallback");
+
+        // Seed one already-held delivery (m0) so release()'s nack loop has something to iterate, and
+        // therefore somewhere to block, before it can return.
+        deliverCallback.get().handle("ctag", delivery(1L, "m0", "first"));
+
+        AtomicReference<Throwable> releaseError = new AtomicReference<>();
+        Thread releaseThread = new Thread(() -> {
+            try {
+                inbox.release(target);
+            } catch (Throwable t) {
+                releaseError.set(t);
+            }
+        }, "release-under-test");
+        releaseThread.start();
+
+        // This latch only fires from inside the fake channel's basicNack — i.e. from inside
+        // release()'s nack loop, which release()'s code only reaches AFTER held.compute(...) has
+        // already swapped in RELEASED. Waiting for it is direct proof the swap has happened and
+        // release() has not yet returned (it is stuck mid-loop, still holding channelLock).
+        assertTrue(nackStarted.await(10, TimeUnit.SECONDS),
+                "release() never reached its nack loop — it may not have started");
+
+        // The exact interleaving CB-318 describes: a delivery for a NEW message on the same target
+        // lands on the "consumer work-pool thread" (this second thread) while release() is still
+        // running. With the pre-fix code (a bare held.remove(target)) this created a brand-new map
+        // under computeIfAbsent that release() — already past its remove — never looks at again.
+        AtomicReference<Throwable> deliveryError = new AtomicReference<>();
+        Thread deliveryThread = new Thread(() -> {
+            try {
+                deliverCallback.get().handle("ctag", delivery(2L, "m1", "second"));
+            } catch (Throwable t) {
+                deliveryError.set(t);
+            }
+        }, "concurrent-delivery");
+        deliveryThread.start();
+
+        // Head start for the delivery thread to reach (and, on the fixed code, block on)
+        // channelLock — release() still holds it at this point, so a correct fix cannot have
+        // resolved m1's nack yet. Purely in-memory work (computeIfAbsent, a reference compare)
+        // separates deliveryThread.start() from that block point, so 300ms is a large margin, not a
+        // tight timing assumption.
+        Thread.sleep(300);
+        assertEquals(1, nackCallCount.get(),
+                "the concurrent delivery must not resolve its nack before release() gives up "
+                        + "channelLock — if this is 2 already, the interleaving below is not being "
+                        + "tested, only a sequential call");
+
+        releaseMayFinishNack.countDown(); // let release() finish nacking m0 and return
+        assertTrue(releaseThread.join(Duration.ofSeconds(10)), "release() did not finish");
+        assertTrue(deliveryThread.join(Duration.ofSeconds(10)), "the concurrent delivery did not finish");
+
+        assertNull(releaseError.get(), "release() threw: " + releaseError.get());
+        assertNull(deliveryError.get(), "the concurrent delivery threw: " + deliveryError.get());
+
+        assertEquals(2, nacks.size(),
+                "both the pre-held m0 and the concurrently-arriving m1 must be nacked, got: "
+                        + nacks.stream().map(n -> "[tag=" + n[0] + " requeue=" + n[1] + "]").toList());
+        assertTrue(nacks.stream().allMatch(n -> n[1] == 1L),
+                "invariant 1 (never drop): every nack must set requeue=true");
+        assertTrue(nacks.stream().anyMatch(n -> n[0] == 1L), "m0's delivery tag must be nacked");
+        assertTrue(nacks.stream().anyMatch(n -> n[0] == 2L),
+                "m1 — delivered while release() was still running, after the tombstone swap — must be "
+                        + "nacked, not silently retained in a map release() will never look at again");
+        assertTrue(acks.isEmpty(), "invariant 1 (never drop): a held reply must never be basicAck'd");
+    }
+
+    private static Delivery delivery(long tag, String msgId, String body) {
+        Envelope envelope = new Envelope(tag, false, "", "irrelevant");
+        AMQP.BasicProperties props = new AMQP.BasicProperties.Builder().messageId(msgId).build();
+        return new Delivery(envelope, props, body.getBytes(StandardCharsets.UTF_8));
+    }
+
+    /** A {@link Proxy}-backed consume {@link Channel}: blocks the FIRST {@code basicNack} call on
+     *  {@code releaseMayFinishNack}, after signalling {@code nackStarted} — everything else records
+     *  the call and returns a harmless default, matching the style already used by
+     *  {@code AmqpReplyInboxRecoveryRaceTest}. */
+    private static Channel fakeConsumeChannel(AtomicReference<DeliverCallback> deliverCallback,
+                                               List<long[]> nacks, List<Long> acks,
+                                               AtomicInteger nackCallCount,
+                                               CountDownLatch nackStarted,
+                                               CountDownLatch releaseMayFinishNack) {
+        InvocationHandler handler = (proxy, method, args) -> {
+            String name = method.getName();
+            if (name.equals("basicConsume")) {
+                deliverCallback.set((DeliverCallback) args[2]);
+                return "ctag";
+            }
+            if (name.equals("basicNack")) {
+                long tag = (long) args[0];
+                boolean requeue = (boolean) args[2];
+                if (nackCallCount.incrementAndGet() == 1) {
+                    nackStarted.countDown();
+                    if (!releaseMayFinishNack.await(10, TimeUnit.SECONDS)) {
+                        throw new IllegalStateException("test never released the nack latch");
+                    }
+                }
+                nacks.add(new long[] {tag, requeue ? 1L : 0L});
+                return null;
+            }
+            if (name.equals("basicAck")) {
+                acks.add((long) args[0]);
+                return null;
+            }
+            if (name.equals("equals")) {
+                return proxy == args[0];
+            }
+            if (name.equals("hashCode")) {
+                return System.identityHashCode(proxy);
+            }
+            if (name.equals("toString")) {
+                return "FakeConsumeChannel";
+            }
+            return defaultValue(method.getReturnType());
+        };
+        return (Channel) Proxy.newProxyInstance(AmqpReplyInboxReleaseRaceTest.class.getClassLoader(),
+                new Class<?>[] {Channel.class}, handler);
+    }
+
+    /** A {@link Proxy}-backed {@link Channel} that answers every call with a harmless default — used
+     *  as the publish channel, which this test never actually publishes on. */
+    private static Channel fakeInertChannel() {
+        InvocationHandler handler = (proxy, method, args) -> {
+            String name = method.getName();
+            if (name.equals("equals")) {
+                return proxy == args[0];
+            }
+            if (name.equals("hashCode")) {
+                return System.identityHashCode(proxy);
+            }
+            if (name.equals("toString")) {
+                return "FakeInertChannel";
+            }
+            return defaultValue(method.getReturnType());
+        };
+        return (Channel) Proxy.newProxyInstance(AmqpReplyInboxReleaseRaceTest.class.getClassLoader(),
+                new Class<?>[] {Channel.class}, handler);
+    }
+
+    /** A {@link Proxy}-backed {@link Connection} handing out {@code first} then {@code second} from
+     *  successive {@code createChannel()} calls, matching {@link AmqpReplyInbox}'s constructor. */
+    private static Connection fakeConnection(Channel first, Channel second) {
+        AtomicInteger calls = new AtomicInteger();
+        InvocationHandler handler = (proxy, method, args) -> {
+            String name = method.getName();
+            if (name.equals("createChannel") && (args == null || args.length == 0)) {
+                return calls.getAndIncrement() == 0 ? first : second;
+            }
+            if (name.equals("equals")) {
+                return proxy == args[0];
+            }
+            if (name.equals("hashCode")) {
+                return System.identityHashCode(proxy);
+            }
+            if (name.equals("toString")) {
+                return "FakeConnection";
+            }
+            return defaultValue(method.getReturnType());
+        };
+        return (Connection) Proxy.newProxyInstance(AmqpReplyInboxReleaseRaceTest.class.getClassLoader(),
+                new Class<?>[] {Connection.class}, handler);
+    }
+
+    private static Object defaultValue(Class<?> type) {
+        if (!type.isPrimitive() || type == void.class) {
+            return null;
+        }
+        if (type == boolean.class) {
+            return Boolean.FALSE;
+        }
+        if (type == long.class) {
+            return 0L;
+        }
+        if (type == short.class) {
+            return (short) 0;
+        }
+        if (type == byte.class) {
+            return (byte) 0;
+        }
+        if (type == char.class) {
+            return (char) 0;
+        }
+        if (type == double.class) {
+            return 0.0d;
+        }
+        if (type == float.class) {
+            return 0.0f;
+        }
+        return 0;
+    }
+}

fleetd/src/test/java/dev/ltms/fleet/msg/MessageServiceTest.java

@@ -940,6 +940,54 @@ class MessageServiceTest {
         assertEquals("PR opened: https://example/pulls/42", view.reply());
     }
 
+    /**
+     * fleetd #324: {@code answer()} holds {@code sessionLocks} for the target and, once the worker's
+     * real terminal reply arrives, calls {@code finishAsyncTask}, which used to read the volatile
+     * {@code task.turnId} twice — once to check it is non-null, once as the key for
+     * {@code asyncTasksByTurn.remove}. {@code ask()}'s own timeout path mutates the same field with no
+     * lock at all. This test does not wait for a real race to land in that narrow window between the
+     * two reads — instead it drives the exact sequence the ticket describes (worker asks, primary
+     * answers, worker's real reply arrives) and, via a package-private test hook wired to fire at
+     * precisely that point, runs the identical production cleanup {@code ask()}'s timeout catch block
+     * runs ({@code clearAsyncQuestion(turnId, true)}) so the field goes {@code null} between the two
+     * reads deterministically rather than by chance.
+     *
+     * <p>What this proves: given that exact interleaving, {@code answer()} must not throw and the
+     * ticket must still resolve to the worker's real reply. What it does not prove: that the
+     * interleaving itself is reachable in production — that is established by reading the code (see
+     * the ticket), not by this test, since forcing it via a hook is not the same as two independent
+     * threads racing on their own schedules.
+     */
+    @Test
+    void finishAsyncTaskSurvivesTurnIdGoingNullBetweenItsTwoReads() 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);
+        String turnId = asking.turnId();
+
+        // Fire ask()'s own unlocked timeout cleanup at the moment finishAsyncTask has already checked
+        // task.turnId is non-null but has not yet used it — the exact torn-read window fleetd #324
+        // describes.
+        messages.setFinishAsyncTaskRaceHookForTest(() -> messages.forgetTurnForTest(turnId));
+
+        CompletableFuture<MessageService.Reply> answer =
+                CompletableFuture.supplyAsync(() -> messages.answer(turnId, "config.yaml", 5000));
+        assertEquals("config.yaml", ask.get(5, TimeUnit.SECONDS).answer());
+        awaitWaiting(); // answer() opened its own forward waiter for the resumed worker turn
+
+        assertTrue(messages.reply(T, "PR opened: https://example/pulls/42"));
+
+        assertEquals(MessageService.Outcome.REPLIED, answer.get(5, TimeUnit.SECONDS).outcome(),
+                "the lead's own answer() call must not throw because ask()'s timeout cleanup raced it");
+        MessageService.TaskView done = awaitTicketPhase(ticket, MessageService.Phase.DONE);
+        assertEquals("PR opened: https://example/pulls/42", done.reply(),
+                "the ticket must still resolve to the worker's real reply despite the forced race");
+    }
+
     @Test
     void unansweredAsyncQuestionReturnsTheTicketToPendingAndReleasesItsTarget() throws Exception {
         String ticket = messages.sendAsync(T, "task that asks");
@@ -957,6 +1005,76 @@ class MessageServiceTest {
         assertEquals(MessageService.Phase.DONE, awaitTicketPhase(next, MessageService.Phase.DONE).phase());
     }
 
+    /**
+     * fleetd #307: a worker's {@code fleet_ask} can time out because the primary never answers —
+     * distinct from {@link #aReplyAfterAnswerTimesOutStillCompletesTheAsyncTicket}, where the
+     * primary DID answer and only its own bounded wait for the resumed turn expired.
+     * {@code ask()}'s timeout path deliberately forgets the task's {@code turnId} (so
+     * {@code hasAsyncQuestion} stops reporting the target BUSY — see
+     * {@code unansweredAsyncQuestionReturnsTheTicketToPendingAndReleasesItsTarget} above), which used
+     * to also erase the one signal {@code askAnsweredAsyncTasks} needed to recognize the worker's
+     * eventual real {@code fleet_reply}. That reply then had nowhere to land but the inbox, and
+     * {@code fleet_poll{ticket}} stayed PENDING forever — later force-failed with the false reason
+     * "session released before it replied", even though the worker had, in fact, replied.
+     */
+    @Test
+    void aReplyAfterAnAskTimeoutStillCompletesTheAsyncTicket() throws Exception {
+        String ticket = messages.sendAsync(T, "task that asks then finishes alone");
+        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");
+
+        // The worker keeps working past the timeout and only now calls fleet_reply — with no live
+        // rendezvous waiter open (ask()'s timeout already closed it) and no new send() having
+        // reopened one for this target.
+        assertTrue(messages.reply(T, "PR opened: https://example/pulls/42"));
+
+        MessageService.TaskView done = awaitTicketPhase(ticket, MessageService.Phase.DONE);
+        assertEquals("PR opened: https://example/pulls/42", done.reply(),
+                "fleet_poll{ticket} must return the worker's real reply, not stay pending forever");
+        assertEquals("reply", done.replySource());
+        assertFalse(messages.hasStrandedReply(T),
+                "the reply completed its own ticket directly and never touched the inbox");
+    }
+
+    /**
+     * fleetd #307's ambiguity guard: an ask timeout frees its target ({@code hasAsyncQuestion}
+     * becomes false the instant it lapses — proven above), so a second, independent delegation can
+     * be dispatched to the same target and itself go on to ask-and-lapse before the first worker's
+     * real reply ever arrives. Two open tasks are then both eligible candidates on one target with
+     * no live waiter to disambiguate them. A reply arriving now must not guess which one it answers
+     * — guessing wrong would hand the lead a plausible-looking answer to a delegation the worker
+     * never touched, worse than a failure because the lead acts on it — so it must fall back to the
+     * inbox exactly as the zero-candidate case does.
+     */
+    @Test
+    void twoAskTimedOutTicketsOnOneTargetFallBackToTheInboxRatherThanGuess() throws Exception {
+        String ticket1 = messages.sendAsync(T, "first task that asks");
+        awaitWaiting();
+        injectDelivery();
+        assertEquals(MessageService.AskOutcome.TIMED_OUT, messages.ask(T, "Q1?", 200).outcome());
+
+        String ticket2 = messages.sendAsync(T, "second task that asks");
+        awaitWaiting();
+        injectDelivery();
+        assertEquals(MessageService.AskOutcome.TIMED_OUT, messages.ask(T, "Q2?", 200).outcome());
+
+        assertTrue(messages.reply(T, "which task does this answer?"));
+
+        assertEquals(MessageService.Phase.PENDING, messages.poll(ticket1).phase(),
+                "an ambiguous reply must not guess ticket1");
+        assertEquals(MessageService.Phase.PENDING, messages.poll(ticket2).phase(),
+                "an ambiguous reply must not guess ticket2");
+        assertTrue(messages.hasStrandedReply(T));
+        var drained = messages.drainReplies(T);
+        assertEquals(1, drained.size());
+        assertEquals("which task does this answer?", drained.get(0).content());
+    }
+
     @Test
     void asyncQuestionBelongsToTheTaskThatOwnsItsForwardWaiter() throws Exception {
         String first = messages.sendAsync(T, "first task");

fleetd/src/test/java/dev/ltms/fleet/rest/FleetAppTest.java

@@ -415,7 +415,11 @@ class FleetAppTest {
         FakeHerdr herdr = new FakeHerdr().agentNameTakenTimes(99);
         int port = start(herdr, "http://gx00.gw:8000", Set.of("gx00.gw"));
 
-        assertEquals(500, req(port, "POST", "/members").statusCode());
+        // fleetd #304: 502, not Javalin's default 500 — the herdr failure is named, and the body
+        // carries herdr's own message, matching what fleet_spawn reports for the same failure.
+        HttpResponse<String> res = req(port, "POST", "/members");
+        assertEquals(502, res.statusCode());
+        assertEquals("herdr_error", mapper.readTree(res.body()).get("error").asText());
         assertTrue(herdr.called("tab.create"), "a tab was created before the failed start");
         assertEquals("w9:t2", params(herdr, "tab.close").get("tab_id"), "orphaned tab must be closed");
     }
@@ -733,7 +737,12 @@ class FleetAppTest {
         int port = start(herdr, "http://gx00.gw:8000", Set.of("gx00.gw"));
 
         // A genuine teardown failure must surface, not be reported as a successful 204.
-        assertEquals(500, req(port, "DELETE", "/members/w9:pW").statusCode());
+        // fleetd #304: it surfaces as a named 502 rather than Javalin's default 500. The property
+        // this test guards is "not 204" and the herdr detail reaching the caller — a bare 500 gave
+        // the body "Server Error" and said nothing about herdr.
+        HttpResponse<String> res = req(port, "DELETE", "/members/w9:pW");
+        assertEquals(502, res.statusCode());
+        assertEquals("herdr_error", mapper.readTree(res.body()).get("error").asText());
         assertFalse(herdr.called("tab.close"), "tab is not removed when the pane close failed");
     }
 
@@ -746,4 +755,5 @@ class FleetAppTest {
         assertEquals(204, req(port, "DELETE", "/members/w9:pW").statusCode());
         assertTrue(herdr.called("tab.close"));
     }
+
 }

fleetd/src/test/java/dev/ltms/fleet/session/GitWorktreesTest.java

@@ -403,6 +403,52 @@ class GitWorktreesTest {
         assertTrue(heads.isBlank(), "the branch leaked after a post-creation step threw:\n" + heads);
     }
 
+    /**
+     * fleetd #309. The add runner is a narrow seam for the case where Git has created state but
+     * the caller then kills the process. The runner first performs the real add in this throwaway
+     * repo, then throws the same kind of exception that {@link GitWorktrees#exec} uses for a timeout.
+     * This proves the failure path cleans both real Git objects without waiting for a slow checkout.
+     */
+    @Test
+    void addCleansUpWhenTheWorktreeAddRunnerFailsAfterCreatingState(@TempDir Path tmp) throws Exception {
+        Path repo = initRepo(tmp.resolve("repo"));
+        String branch = "cb-309-timeout";
+        WorktreeException timeout = new WorktreeException("command timed out: synthetic git worktree add");
+        AtomicReference<String> createdPath = new AtomicReference<>();
+        GitWorktrees worktrees = new GitWorktrees(tmp.resolve("wts").toString(), null, null, null, command -> {
+            createdPath.set(command[5]);
+            try {
+                git(repo, "worktree", "add", command[5], "-b", command[7], command[8]);
+            } catch (Exception e) {
+                throw new AssertionError("test setup could not create the worktree", e);
+            }
+            throw timeout;
+        });
+
+        WorktreeException thrown = assertThrows(WorktreeException.class,
+                () -> worktrees.add(repo.toString(), branch, "HEAD"));
+
+        assertSame(timeout, thrown, "cleanup must not replace the add failure");
+        assertNotNull(createdPath.get(), "the add runner must receive the worktree path");
+        assertFalse(Files.exists(Path.of(createdPath.get())),
+                "the worktree directory leaked after the add runner failed");
+        assertFalse(refExists(repo, "refs/heads/" + branch), "the branch leaked after the add runner failed");
+    }
+
+    /** An ordinary Git refusal must not delete the existing branch or log a cleanup warning. */
+    @Test
+    void addFailureBeforeCreatingAWorktreeIsQuiet(@TempDir Path tmp) throws Exception {
+        Path repo = initRepo(tmp.resolve("repo"));
+        String branch = "already-exists";
+        git(repo, "branch", branch);
+
+        assertThrows(WorktreeException.class, () -> new GitWorktrees(tmp.resolve("wts").toString())
+                .add(repo.toString(), branch, "HEAD"));
+
+        assertTrue(refExists(repo, "refs/heads/" + branch), "the existing branch must remain");
+        assertTrue(capturedMessages().isEmpty(), "an ordinary Git refusal logged a warning: " + capturedMessages());
+    }
+
     // ---- CB-189: broader remote-URL coverage — every remote, both fetch and push URLs, any
     // non-SSH scheme. Reporting only, additive to the origin/https strip-and-refuse tests above. ----
 

fleetd/src/test/java/dev/ltms/fleet/session/SessionManagerTest.java

@@ -26,7 +26,12 @@ import org.slf4j.LoggerFactory;
 import java.util.List;
 import java.util.Map;
 import java.util.Set;
+import java.util.concurrent.CountDownLatch;
+import java.util.concurrent.ExecutorService;
+import java.util.concurrent.Executors;
+import java.util.concurrent.Future;
 import java.util.concurrent.TimeUnit;
+import java.util.concurrent.atomic.AtomicInteger;
 import java.util.function.LongSupplier;
 
 import static org.junit.jupiter.api.Assertions.*;
@@ -81,17 +86,50 @@ class SessionManagerTest {
         private volatile RuntimeException hasUncommittedFailure;
         private volatile RuntimeException snapshotFailure;
         private final java.util.concurrent.atomic.AtomicLong snapshotSeq = new java.util.concurrent.atomic.AtomicLong();
+        /** fleetd #316: successive {@code hasUncommitted} answers, one per call, last one sticky
+         *  once exhausted — models a worktree whose state changes between reads. Empty (the
+         *  default) falls back to the plain {@link #dirty} flag, so every existing test using this
+         *  fake keeps returning one fixed answer. */
+        private final List<Boolean> dirtySequence = new java.util.concurrent.CopyOnWriteArrayList<>();
+        private int failHasUncommittedOnCall = -1;
+        private RuntimeException hasUncommittedCallFailure;
+        private final java.util.concurrent.atomic.AtomicInteger hasUncommittedCalls =
+                new java.util.concurrent.atomic.AtomicInteger();
 
         RecordingWorktrees dirty(boolean dirty) {
             this.dirty = dirty;
             return this;
         }
 
+        /** fleetd #316: return {@code answers[0]} on the first {@code hasUncommitted} call,
+         *  {@code answers[1]} on the second, and so on; the last element repeats after that. */
+        RecordingWorktrees dirtySequence(boolean... answers) {
+            for (boolean a : answers) {
+                dirtySequence.add(a);
+            }
+            return this;
+        }
+
+        int hasUncommittedCallCount() {
+            return hasUncommittedCalls.get();
+        }
+
         RecordingWorktrees failHasUncommittedWith(RuntimeException e) {
             this.hasUncommittedFailure = e;
             return this;
         }
 
+        /**
+         * Throw from {@code hasUncommitted} on one specific call only, counting from 0. The
+         * whole-double {@link #failHasUncommittedWith} cannot express fleetd #316's fail-safe
+         * case, which needs the pre-stop read to succeed and only the late read to fail.
+         */
+        RecordingWorktrees failHasUncommittedOnCall(int call, RuntimeException e) {
+            this.failHasUncommittedOnCall = call;
+            this.hasUncommittedCallFailure = e;
+            return this;
+        }
+
         RecordingWorktrees failRemoveFor(String worktreePath) {
             failRemoveFor.add(worktreePath);
             return this;
@@ -122,9 +160,16 @@ class SessionManagerTest {
 
         @Override
         public boolean hasUncommitted(String worktreePath) {
+            int call = hasUncommittedCalls.getAndIncrement();
             if (hasUncommittedFailure != null) {
                 throw hasUncommittedFailure;
             }
+            if (call == failHasUncommittedOnCall) {
+                throw hasUncommittedCallFailure;
+            }
+            if (!dirtySequence.isEmpty()) {
+                return dirtySequence.get(Math.min(call, dirtySequence.size() - 1));
+            }
             return dirty;
         }
 
@@ -178,14 +223,19 @@ class SessionManagerTest {
     }
 
     private SessionManager sessionManager(FakeHerdr herdr, LongSupplier clock, int contextCap,
-                                          boolean clearAfterTurn) {
+                                           boolean clearAfterTurn) {
+        return sessionManager(herdr, clock, contextCap, clearAfterTurn, null);
+    }
+
+    private SessionManager sessionManager(FakeHerdr herdr, LongSupplier clock, int contextCap,
+                                           boolean clearAfterTurn, java.util.function.Consumer<MemberSession> hook) {
         FleetConfig.Profile cfg = new FleetConfig.Profile(
                 "ltms-local", "http://gx00.gw:8000", "coder", null, "FLEETD_WORKER_TOKEN",
                 List.of("ccs", "ltms-local"), "tab", "fleetd-workers",
                 "worker: {profile} #{n}", null, null, null);
         ClaudeCodeLauncher workers = new ClaudeCodeLauncher(new AgentControl(herdr), new WorkspaceControl(herdr),
                 new SubscriptionGuard(Set.of("gx00.gw")), Map.of(cfg.profile(), cfg), cfg.profile(), _ -> null);
-        return new SessionManager(workers, new GitWorktrees(), clock, contextCap, clearAfterTurn);
+        return new SessionManager(workers, new GitWorktrees(), clock, contextCap, clearAfterTurn, hook);
     }
 
     @Test
@@ -670,6 +720,24 @@ class SessionManagerTest {
                 "BUSY session remains");
     }
 
+    @Test
+    void reapIdleDoesNotReleaseSessionDeliveredAfterItsEligibilityCheck() {
+        long[] clock = {0};
+        FakeHerdr herdr = new FakeHerdr();
+        SessionManager[] manager = new SessionManager[1];
+        SessionManager sessions = sessionManager(herdr, () -> clock[0], 0, false,
+                session -> manager[0].onDelivered(session.terminalId(), TestTurnTokens.inert(session.terminalId())));
+        manager[0] = sessions;
+        MemberSession session = sessions.acquire("ltms-local", null, "/caller", "term_primary");
+        sessions.asPresence().markPresent(session.terminalId());
+
+        clock[0] = 11;
+        assertEquals(0, sessions.reapIdle(10), "delivery replaces the idle snapshot before release");
+        assertEquals(MemberSession.State.BUSY, sessions.get(session.paneId()).orElseThrow().state(),
+                "a just-delivered session stays registered and busy");
+        assertFalse(herdr.called("pane.close"), "the busy session pane is not stopped");
+    }
+
     @Test
     void doneSessionPastIdleTtlIsReaped() {
         long[] clock = {0};
@@ -824,6 +892,186 @@ class SessionManagerTest {
                 .count();
     }
 
+    // --- fleetd #308: a spawn accepted while the shutdown drain is running must not orphan ---
+
+    @Test
+    void acquireRefusesANewSpawnOnceDrainAllHasStarted() {
+        FakeHerdr herdr = new FakeHerdr();
+        SessionManager sessions = sessionManager(herdr);
+
+        sessions.drainAll(TimeUnit.MILLISECONDS.toNanos(50)); // empty roster — returns immediately,
+        // but the shutdown guard it flips must stay tripped for the life of the process.
+
+        ShuttingDownException e = assertThrows(ShuttingDownException.class,
+                () -> sessions.acquire("ltms-local", null, "/caller", "term_primary"),
+                "a spawn requested after the drain has begun must be refused loudly (invariant 3), "
+                        + "not silently registered into a registry the drain will never revisit");
+        assertNotNull(e.getMessage());
+        assertFalse(e.getMessage().isBlank(), "the refusal must say why, not just that it failed");
+        assertTrue(sessions.roster().isEmpty(), "the refused spawn must never reach the registry");
+    }
+
+    /**
+     * fleetd #308: the guard above closes most of the shutdown-race window, but it cannot close
+     * all of it — a caller that already passed the {@code draining} check before {@code drainAll}
+     * flips it can still be mid-{@code launcher.spawn()} (a real herdr round trip, not
+     * instantaneous) when {@code drainAll} takes its registry snapshot. This test forces exactly
+     * that interleaving with a launcher double that blocks the second {@code spawn()} call and the
+     * first {@code stop()} call until released, then proves the post-loop sweep in {@code
+     * drainAll} still finds and tears down the straggler that lands in the registry afterward.
+     */
+    @Test
+    void drainAllSweepsAStragglerThatRegisteredAfterTheInitialSnapshot() throws Exception {
+        FakeHerdr herdr = new FakeHerdr();
+        FleetConfig.Profile cfg = new FleetConfig.Profile(
+                "ltms-local", "http://gx00.gw:8000", "coder", null, "FLEETD_WORKER_TOKEN",
+                List.of("ccs", "ltms-local"), "tab", "fleetd-workers",
+                "worker: {profile} #{n}", null, null, null);
+        ClaudeCodeLauncher delegate = new ClaudeCodeLauncher(new AgentControl(herdr), new WorkspaceControl(herdr),
+                new SubscriptionGuard(Set.of("gx00.gw")), Map.of(cfg.profile(), cfg), cfg.profile(), _ -> null);
+        RaceLauncher race = new RaceLauncher(delegate);
+        SessionManager sessions = new SessionManager(race);
+
+        // Registered normally, before the drain starts — the first spawn call, never blocked.
+        MemberSession ready = sessions.acquire("ltms-local", "/ready", "/caller", "ownerR");
+
+        ExecutorService exec = Executors.newFixedThreadPool(2);
+        try {
+            // The straggler's acquire() reads `draining == false` (checked before this call ever
+            // touches the launcher) and then blocks inside its own spawn() — the second spawn call.
+            Future<MemberSession> straggler = exec.submit(() ->
+                    sessions.acquire("ltms-local", "/late", "/caller", "ownerLate"));
+
+            assertTrue(race.enteredSecondSpawn.await(5, TimeUnit.SECONDS),
+                    "the straggler must have passed the shutdown guard and reached spawn() before "
+                            + "drainAll ever runs");
+            assertEquals(1, sessions.roster().size(),
+                    "the straggler is still inside spawn() — not registered yet");
+
+            // drainAll flips `draining`, snapshots the registry (only `ready` is in it), and starts
+            // releasing that snapshot — its first release() call stops `ready`'s pane, which this
+            // launcher double blocks on so the interleaving below is deterministic, not a timing bet.
+            Future<?> drain = exec.submit(() -> sessions.drainAll(TimeUnit.SECONDS.toNanos(5)));
+
+            assertTrue(race.enteredFirstStop.await(5, TimeUnit.SECONDS),
+                    "drainAll must be stopping the ready session's pane — proof its initial "
+                            + "registry snapshot has already been taken");
+
+            // Only now does the straggler's spawn complete and register — strictly after the
+            // snapshot drainAll's main pass is working from.
+            race.releaseSecondSpawn.countDown();
+            MemberSession registered = straggler.get(5, TimeUnit.SECONDS);
+
+            // Let drainAll finish releasing `ready`; it then re-checks the registry and must find
+            // (and drain) the straggler that just landed in it.
+            race.releaseFirstStop.countDown();
+            drain.get(5, TimeUnit.SECONDS);
+
+            assertTrue(sessions.roster().isEmpty(),
+                    "the post-loop sweep must drain the straggler too, not just the initial snapshot");
+            assertNotNull(registered.paneId());
+            long paneCloseCalls = herdr.calls.stream().filter(c -> "pane.close".equals(c.method())).count();
+            assertEquals(2, paneCloseCalls,
+                    "both ready's pane AND the straggler's pane must actually be stopped — a pane "
+                            + "left running is exactly the orphan this ticket is about");
+        } finally {
+            exec.shutdownNow();
+        }
+    }
+
+    /**
+     * Delegates every call while blocking the SECOND {@code spawn()} call and the FIRST
+     * {@code stop()} call until the test releases them — used to force the fleetd #308 race
+     * deterministically instead of betting on real thread-scheduling timing.
+     */
+    private static final class RaceLauncher implements PeerLauncher {
+        private final PeerLauncher delegate;
+        private final AtomicInteger spawnCalls = new AtomicInteger();
+        private final AtomicInteger stopCalls = new AtomicInteger();
+        final CountDownLatch enteredSecondSpawn = new CountDownLatch(1);
+        final CountDownLatch releaseSecondSpawn = new CountDownLatch(1);
+        final CountDownLatch enteredFirstStop = new CountDownLatch(1);
+        final CountDownLatch releaseFirstStop = new CountDownLatch(1);
+
+        RaceLauncher(PeerLauncher delegate) {
+            this.delegate = delegate;
+        }
+
+        private static void awaitOrFail(CountDownLatch latch) {
+            try {
+                if (!latch.await(5, TimeUnit.SECONDS)) {
+                    throw new AssertionError("RaceLauncher latch timed out");
+                }
+            } catch (InterruptedException e) {
+                Thread.currentThread().interrupt();
+                throw new AssertionError("RaceLauncher latch interrupted", e);
+            }
+        }
+
+        @Override
+        public Set<Capability> capabilities() {
+            return delegate.capabilities();
+        }
+
+        @Override
+        public Set<Capability> capabilitiesFor(String profileName) {
+            return delegate.capabilitiesFor(profileName);
+        }
+
+        @Override
+        public PeerHandle spawn(SpawnRequest req) {
+            if (spawnCalls.incrementAndGet() == 2) {
+                enteredSecondSpawn.countDown();
+                awaitOrFail(releaseSecondSpawn);
+            }
+            return delegate.spawn(req);
+        }
+
+        @Override
+        public Set<String> profiles() {
+            return delegate.profiles();
+        }
+
+        @Override
+        public String defaultProfile() {
+            return delegate.defaultProfile();
+        }
+
+        @Override
+        public String effectiveCwd(SpawnRequest req) {
+            return delegate.effectiveCwd(req);
+        }
+
+        @Override
+        public List<String> parityOverlay(String profileName) {
+            return delegate.parityOverlay(profileName);
+        }
+
+        @Override
+        public List<?> list() {
+            return delegate.list();
+        }
+
+        @Override
+        public int reapOrphanWorkers() {
+            return delegate.reapOrphanWorkers();
+        }
+
+        @Override
+        public void stop(String id) {
+            if (stopCalls.incrementAndGet() == 1) {
+                enteredFirstStop.countDown();
+                awaitOrFail(releaseFirstStop);
+            }
+            delegate.stop(id);
+        }
+
+        @Override
+        public boolean clearContext(String id) {
+            return delegate.clearContext(id);
+        }
+    }
+
     private static List<String> promptTexts(FakeHerdr herdr) {
         return herdr.calls.stream()
                 .filter(c -> "agent.prompt".equals(c.method()))
@@ -999,6 +1247,104 @@ class SessionManagerTest {
                         + "dirty check threw");
     }
 
+    // --- fleetd #316: the dirty check must be re-taken after the worker is stopped, not trusted
+    // stale from before it ------------------------------------------------------------------------
+
+    @Test
+    void releaseDoesNotRemoveAWorktreeThatBecameDirtyBetweenTheFirstCheckAndRemoval() {
+        // Models the exact race #316 reports: hasUncommitted answers clean while the worker is
+        // still running (call 1), the worker then writes new work, and by the time release is
+        // about to force-remove the worktree a second read (call 2) would see it as dirty. Without
+        // the fix this test fails: release() never re-reads and force-removes the worktree anyway.
+        FakeHerdr herdr = new FakeHerdr();
+        RecordingWorktrees worktrees = new RecordingWorktrees().dirtySequence(false, true);
+        SessionManager sessions = sessionManager(herdr, worktrees);
+        MemberSession s = sessions.acquire("ltms-local", null, "/caller/proj", null,
+                new WorktreeRequest("cb-316a", null));
+
+        sessions.release(s.paneId());
+
+        assertTrue(worktrees.removeCalls().isEmpty(),
+                "a worktree that turned dirty between the pre-stop read and removal must be preserved");
+        assertEquals(2, worktrees.hasUncommittedCallCount(),
+                "the fix re-reads hasUncommitted exactly once more, immediately before removal");
+    }
+
+    @Test
+    void releasePreservesAWorktreeWhoseLateRecheckCannotBeRead() {
+        // fleetd #316 invariant 1, which no test pinned when the fix landed: the late re-check
+        // fails toward PRESERVING. Found by mutation — flipping dirtyImmediatelyBeforeRemoval's
+        // catch from `return true` to `return false` turned the guard into a cause of the very
+        // data loss it was added to stop, and the whole suite stayed green. The pre-stop read
+        // succeeds and says clean (call 0); the read that authorises the removal throws (call 1).
+        FakeHerdr herdr = new FakeHerdr();
+        RecordingWorktrees worktrees = new RecordingWorktrees()
+                .dirtySequence(false)
+                .failHasUncommittedOnCall(1, new WorktreeException("git status exited 128"));
+        SessionManager sessions = sessionManager(herdr, worktrees);
+        MemberSession s = sessions.acquire("ltms-local", null, "/caller/proj", null,
+                new WorktreeRequest("cb-316c", null));
+
+        sessions.release(s.paneId());
+
+        assertTrue(worktrees.removeCalls().isEmpty(),
+                "a worktree whose state cannot be read immediately before removal must be kept: "
+                        + "preserving costs disk, deleting on a guess destroys work with no other copy");
+        assertEquals(2, worktrees.hasUncommittedCallCount(),
+                "the late re-check still runs — it is the throwing call, not a skipped one");
+    }
+
+    @Test
+    void releaseSnapshotsWorkFoundOnlyByTheLateRecheck() {
+        // #316's second half: the pre-stop dirty=false means trySnapshot never ran for this
+        // session, so the late-discovered work would otherwise have no refs/wip/* copy at all —
+        // only the on-disk preserve. The re-check path must snapshot it too.
+        FakeHerdr herdr = new FakeHerdr();
+        RecordingWorktrees worktrees = new RecordingWorktrees().dirtySequence(false, true);
+        SessionManager sessions = sessionManager(herdr, worktrees);
+        MemberSession s = sessions.acquire("ltms-local", null, "/caller/proj", null,
+                new WorktreeRequest("cb-316b", null));
+
+        sessions.release(s.paneId());
+
+        assertEquals(java.util.List.of(s.worktree()), worktrees.snapshotCalls(),
+                "the newly-dirty worktree is snapshotted even though the pre-stop check saw it clean");
+    }
+
+    @Test
+    void releaseStillRemovesAWorktreeThatStaysCleanOnTheLateRecheck() {
+        // The ordinary, non-racing case: nothing else changes behaviour when the second read
+        // agrees with the first.
+        FakeHerdr herdr = new FakeHerdr();
+        RecordingWorktrees worktrees = new RecordingWorktrees().dirty(false);
+        SessionManager sessions = sessionManager(herdr, worktrees);
+        MemberSession s = sessions.acquire("ltms-local", null, "/caller/proj", null,
+                new WorktreeRequest("cb-316c", null));
+
+        sessions.release(s.paneId());
+
+        assertEquals(java.util.List.of(s.worktree()), worktrees.removeCalls(),
+                "a worktree that is still clean on the late recheck is removed as before");
+    }
+
+    @Test
+    void releaseNeverReChecksAWorktreeAlreadyPreservedByTheFirstDirtyCheck() {
+        // Invariant 4 from #316: no second unconditional git status. A release that already
+        // decided to preserve (the ordinary CB-576 dirty path) must not pay for a second read.
+        FakeHerdr herdr = new FakeHerdr();
+        RecordingWorktrees worktrees = new RecordingWorktrees().dirty(true);
+        SessionManager sessions = sessionManager(herdr, worktrees);
+        MemberSession s = sessions.acquire("ltms-local", null, "/caller/proj", null,
+                new WorktreeRequest("cb-316d", null));
+
+        sessions.release(s.paneId());
+
+        assertEquals(1, worktrees.hasUncommittedCallCount(),
+                "a release that already preserves on the first read must not re-check before "
+                        + "skipping the removal it was never going to do");
+        assertTrue(worktrees.removeCalls().isEmpty());
+    }
+
     /**
      * fleetd #283 defect 1 changed this test's own premise, so its assertions are updated along
      * with the production fix. Before #283, the middle session's worktree-removal failure escaped