fleetd #501: readiness-grace expiry logs measured elapsed time and the loop's own poll counter, never the configured budget

The line at Injector.java:383-387 printed two numbers that read as measurements
but were both compile-time constants: READINESS_GRACE_POLLS for the poll count
(the loop's own Target.notReadySincePoll counter was in scope at the same call
site), and READINESS_GRACE_POLLS * POLL_INTERVAL_MILLIS / 1000 for the elapsed
time — arithmetic on two constants, never a measurement, and wrong in the
direction that says everything ran on schedule.

Fix, copying the LongSupplier-clock shape LeadRollover already uses:
- print t.notReadySincePoll instead of the constant for the poll count (the two
  agree by construction on this branch, so no test can tell them apart — the
  comment says so honestly).
- add Target.notReadySinceMillis, stamped at the first non-ready sample and
  reset at all three sites notReadySincePoll already resets (:304, :355, :389
  pre-fix line numbers), to compute a real elapsed time at expiry.
- inject a LongSupplier nowMillis (defaulting to System::currentTimeMillis)
  through new package-private constructor overloads so a test can supply a
  clock whose advance does not track POLL_INTERVAL_MILLIS.

Tests use a ListAppender to assert on the log message contents, per
LeadRolloverTest's pattern. The elapsed-time test drives the loop with a stub
clock returning two literal, non-derived values so it can fail if the fix
regresses to the constant-arithmetic line — proved by mutation: reverting the
elapsed calculation to READINESS_GRACE_POLLS * POLL_INTERVAL_MILLIS turns that
one test red (1 failure); reverting the poll-count print to the constant is an
equivalent mutant (0 failures), because the counter and the constant are
identical at that exact call site by construction.

fleetd clean install: Tests run: 1683, Failures: 0, Errors: 0, Skipped: 0.

.claude/skills/handover/SKILL.md

@@ -141,6 +141,14 @@ refused.**
    own workspace before it hands it to you. The daemon and your pane can run in different
    directories, so a path you resolve yourself can point at a different file from the one the daemon
    will check.
+
+   **Check that the path is ignored by git before you write to it (#491).** A relative
+   `handoverPath` resolves inside YOUR workspace, which is usually a repository — and usually not
+   the `fleetd` one, so an ignore rule added to `fleetd` does not protect it. Run
+   `grep -n handover <your workspace>/.gitignore`. No output means the file you are about to write
+   will show up as untracked content in that repo. The file is a snapshot of live state and must
+   never be committed, so tell the operator rather than committing it or silently editing their
+   `.gitignore`.
 2. **Write the handover file at that path**, following sections 1–10 above.
 3. **Ask the operator, then `fleet_handover{action: "confirm", token, operatorConfirmed: true}`.**
 
@@ -164,10 +172,14 @@ fails.
   session.
 - **The roll can still refuse after `confirm` returns**, and by then there is no caller to tell.
   Those outcomes are logged only, as `lead-rollover:` lines in the daemon log.
-- **If the bootstrap prompt never lands, your context is gone and no fresh session starts.** This
-  has not yet been proven end-to-end (see fleetd #480). The recovery is the manual path: the file
-  is already written, so the operator starts a session and points it at the file. That is why you
-  write the file before you confirm, and never the other way round.
+- **The bootstrap prompt has never yet landed, and the fix is unproven (fleetd #489).** The first
+  real rollover, on 2026-09-12, joined `/clear` and the bootstrap text into one line and Claude Code
+  refused it as `Unknown command: /clearFresh`. The pane was never cleared and no context was lost,
+  so the failure was safe — the roll simply did nothing. PR #490 fixed the cause and is deployed,
+  but no roll has bootstrapped a fresh session end to end yet. **Assume it may still fail, and tell
+  the operator so before you confirm.** The recovery is the same either way: the file is already
+  written, so the operator starts a session and points it at the file. That is why you write the
+  file before you confirm, and never the other way round.
 
 ## Writing style
 

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

@@ -15,6 +15,7 @@ import java.util.Set;
 import java.util.concurrent.CompletableFuture;
 import java.util.concurrent.ConcurrentHashMap;
 import java.util.function.Consumer;
+import java.util.function.LongSupplier;
 import java.util.function.Predicate;
 import java.util.stream.Collectors;
 
@@ -94,6 +95,14 @@ public final class Injector {
     private final TurnListener turnListener;
     private final Predicate<String> ready; // CB-113: a target is deliverable only when available
     private final Consumer<String> forget; // CB-114: clear a gone worker's readiness/presence
+    /**
+     * Wall-clock source for the readiness-grace elapsed time logged in {@link #onStatus} (fleetd
+     * #501). Production constructors default this to {@code System::currentTimeMillis}; the
+     * package-private constructors below take it explicitly so a test can supply a stub whose
+     * advance does not track {@link #POLL_INTERVAL_MILLIS} — copying the shape {@code LeadRollover}
+     * already uses for the same purpose.
+     */
+    private final LongSupplier nowMillis;
     private final ConcurrentHashMap<String, Target> targets = new ConcurrentHashMap<>();
 
     /** Delivery only; completion signalling is a no-op and every target is treated as available. */
@@ -125,20 +134,34 @@ public final class Injector {
      */
     public Injector(AgentControl agents, TurnListener turnListener, Predicate<String> ready,
                     Consumer<String> forget) {
+        this(agents, turnListener, ready, forget, System::currentTimeMillis);
+    }
+
+    /** Full constructor — for tests: an injectable wall-clock supplier (fleetd #501). */
+    Injector(AgentControl agents, TurnListener turnListener, Predicate<String> ready,
+             Consumer<String> forget, LongSupplier nowMillis) {
         this.agents = agents;
         this.router = null;
         this.turnListener = turnListener;
         this.ready = ready;
         this.forget = forget;
+        this.nowMillis = nowMillis;
     }
 
     public Injector(HerdrRouter router, TurnListener turnListener, Predicate<String> ready,
                     Consumer<String> forget) {
+        this(router, turnListener, ready, forget, System::currentTimeMillis);
+    }
+
+    /** Full constructor — for tests: an injectable wall-clock supplier (fleetd #501). */
+    Injector(HerdrRouter router, TurnListener turnListener, Predicate<String> ready,
+             Consumer<String> forget, LongSupplier nowMillis) {
         this.agents = null;
         this.router = router;
         this.turnListener = turnListener;
         this.ready = ready;
         this.forget = forget;
+        this.nowMillis = nowMillis;
     }
 
     private AgentControl agentsFor(String target) {
@@ -209,6 +232,7 @@ public final class Injector {
         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)
+        long notReadySinceMillis;  // wall-clock time of the FIRST non-ready sample in the current notReadySincePoll streak (fleetd #501); reset alongside it
         boolean postTurnPending;   // completion observed; adapter housekeeping has not started yet
         boolean awaitingPostTurnPickup;
         boolean postTurnObserved;
@@ -302,6 +326,7 @@ public final class Injector {
                 t.unknownSinceTurn = 0;
                 t.unknownSincePostTurn = 0;
                 t.notReadySincePoll = 0;
+                t.notReadySinceMillis = 0;
                 if (t.awaitingCompletion) t.turnObserved = true;
             } else if (status.injectable()) { // IDLE or BLOCKED
                 t.unknownSinceTurn = 0;
@@ -353,6 +378,7 @@ public final class Injector {
                         Pending p = t.queue.peek();
                         if (p != null && ready.test(target)) {
                             t.notReadySincePoll = 0;
+                            t.notReadySinceMillis = 0;
                             try {
                                 agentsFor(target).send(target, p.text());
                                 t.queue.poll();
@@ -370,23 +396,52 @@ public final class Injector {
                                 sent = p;
                                 sendError = e;
                             }
-                        } else if (p != null && ++t.notReadySincePoll >= READINESS_GRACE_POLLS) {
-                            // The worker has been idle-but-not-ready for the whole grace: its Claude
-                            // never connected the bridge MCP (crashed during boot, or wedged on a
-                            // startup prompt). The readiness gate would hold this message forever, so
-                            // fail every queued message and release the target (CB-114) instead of
-                            // polling it indefinitely with the caller's future never completing.
-                            notReady = new ArrayList<>(t.queue);
-                            for (Pending pending : notReady) {
-                                pending.state = Pending.State.NOT_DELIVERED;
+                        } else if (p != null) {
+                            // fleetd #501: stamp the wall-clock time of the FIRST non-ready sample in
+                            // this streak, so the expiry log below can print how long the target
+                            // actually sat non-ready — not just how many polls that took.
+                            if (t.notReadySincePoll == 0) {
+                                t.notReadySinceMillis = nowMillis.getAsLong();
+                            }
+                            if (++t.notReadySincePoll >= READINESS_GRACE_POLLS) {
+                                // The worker has been idle-but-not-ready for the whole grace: its Claude
+                                // never connected the bridge MCP (crashed during boot, or wedged on a
+                                // startup prompt). The readiness gate would hold this message forever, so
+                                // fail every queued message and release the target (CB-114) instead of
+                                // polling it indefinitely with the caller's future never completing.
+                                notReady = new ArrayList<>(t.queue);
+                                for (Pending pending : notReady) {
+                                    pending.state = Pending.State.NOT_DELIVERED;
+                                }
+                                // fleetd #501: t.notReadySincePoll — the loop's own counter, already in
+                                // scope — is printed here instead of the READINESS_GRACE_POLLS constant.
+                                // On this branch the counter has JUST reached the threshold, so the two
+                                // agree by construction and no test can tell them apart. Printed anyway:
+                                // it gives this line one source of truth instead of two, so a later
+                                // change to the loop above cannot leave this message reporting a number
+                                // the loop no longer produces.
+                                //
+                                // elapsedMillis is a different case: it is NOT equal-by-construction to
+                                // the truth. notReadySincePoll only increments on a sample that reaches
+                                // this branch (p != null, not ready) — a poll that misses that condition
+                                // advances real time without advancing the counter — and this loop's real
+                                // period is not guaranteed to equal POLL_INTERVAL_MILLIS (load, or a host
+                                // sleep, can widen the real gap far past it). READINESS_GRACE_POLLS *
+                                // POLL_INTERVAL_MILLIS / 1000 is arithmetic on two constants, not a
+                                // measurement, so it stays here only as the labelled CONFIGURED budget,
+                                // never presented as elapsed time.
+                                long elapsedMillis = nowMillis.getAsLong() - t.notReadySinceMillis;
+                                log.warn("readiness grace for {} expired after {} polls (configured={} "
+                                                + "polls/{}s elapsed={}ms): target never became "
+                                                + "deliverable, so failing {} queued message(s) that "
+                                                + "never reached its pane",
+                                        target, t.notReadySincePoll, READINESS_GRACE_POLLS,
+                                        READINESS_GRACE_POLLS * POLL_INTERVAL_MILLIS / 1000, elapsedMillis,
+                                        notReady.size());
+                                t.queue.clear();
+                                t.notReadySincePoll = 0;
+                                t.notReadySinceMillis = 0;
                             }
-                            log.warn("readiness grace for {} expired after {} polls ({}s): target never "
-                                    + "became deliverable, so failing {} queued message(s) that never "
-                                    + "reached its pane",
-                                    target, READINESS_GRACE_POLLS,
-                                    READINESS_GRACE_POLLS * POLL_INTERVAL_MILLIS / 1000, notReady.size());
-                            t.queue.clear();
-                            t.notReadySincePoll = 0;
                         }
                     }
                 }

fleetd/src/main/java/dev/ltms/fleet/lead/LeadRollover.java

@@ -381,13 +381,18 @@ public final class LeadRollover {
      */
     private void runRollover(PendingRollover p, FleetConfig.LeadRollover cfg) {
         String lead = p.leadTerminal();
-        boolean turnSettled = waitUntilAtTurnBoundary(lead, cfg.turnSettleSeconds());
-        if (!turnSettled) {
-            log.warn("lead-rollover: pane {} did not reach a turn boundary (IDLE or DONE) within {}s "
-                            + "after confirm() — refusing to send /clear at all; the calling lead's "
-                            + "own turn is still live and clearing it now would destroy live context "
-                            + "(token={})",
-                    lead, cfg.turnSettleSeconds(), p.token());
+        long rollStartMillis = nowMillis.getAsLong();
+        TurnSettleResult turnResult = waitUntilAtTurnBoundary(lead, cfg.turnSettleSeconds());
+        if (!turnResult.settled()) {
+            // fleetd #494 follow-up: this line had the SAME defect as the /clear-timeout line below
+            // — cfg.turnSettleSeconds() is the CONFIGURED budget, not how long this wait actually
+            // ran. Print the measured elapsed time alongside it, labelled, exactly like the /clear
+            // path already does.
+            log.warn("lead-rollover: pane {} did not reach a turn boundary (IDLE or DONE) after "
+                            + "confirm() — refusing to send /clear at all; the calling lead's own "
+                            + "turn is still live and clearing it now would destroy live context "
+                            + "(token={}, configured={}s elapsed={}ms)",
+                    lead, p.token(), cfg.turnSettleSeconds(), turnResult.elapsedMillis());
             return;
         }
 
@@ -395,15 +400,22 @@ public final class LeadRollover {
         // /clear is housekeeping, not a delegated turn, and routing it through Injector wedges the
         // pane forever (see this class's javadoc).
         agents.send(lead, "/clear");
-        boolean clearSettled = waitForClearPickupAndSettle(lead, cfg.clearSettleSeconds());
-        if (!clearSettled) {
-            log.warn("lead-rollover: pane {} did not reach a turn boundary (IDLE or DONE) within {}s "
-                            + "after /clear — NOT sending bootstrapText (token={})",
-                    lead, cfg.clearSettleSeconds(), p.token());
+        ClearSettleResult clearResult = waitForClearPickupAndSettle(lead, cfg.clearSettleSeconds());
+        if (!clearResult.settled()) {
+            // fleetd #494: cfg.clearSettleSeconds() is the CONFIGURED budget, not how long the wait
+            // actually ran — an operator reading only that number wrongly believes it is a measured
+            // duration. Print the measured elapsed time and nudge count alongside it, each labelled,
+            // so the two can be compared at a glance.
+            log.warn("lead-rollover: pane {} did not reach a turn boundary (IDLE or DONE) after "
+                            + "/clear — NOT sending bootstrapText (token={}, configured={}s "
+                            + "elapsed={}ms nudges={})",
+                    lead, p.token(), cfg.clearSettleSeconds(), clearResult.elapsedMillis(),
+                    clearResult.nudges());
             return;
         }
         agents.send(lead, cfg.bootstrapTextFor(p.handoverPath()));
-        log.info("lead-rollover: rolled token={} lead={}", p.token(), lead);
+        long rollElapsedMillis = nowMillis.getAsLong() - rollStartMillis;
+        log.info("lead-rollover: rolled token={} lead={} elapsedMs={}", p.token(), lead, rollElapsedMillis);
     }
 
     /** Drop a pending request without rolling. @return whether a pending request existed for {@code token} */
@@ -475,9 +487,15 @@ public final class LeadRollover {
      * gate exists to prevent. Do not "simplify" this back to {@code injectable()}. ({@link
      * #waitForClearPickupAndSettle} keeps the same exclusion of {@code BLOCKED}, for the same
      * reason, on the second wait.)
+     *
+     * @return a {@link TurnSettleResult} whose {@code settled()} is {@code true} once a real
+     *         boundary was observed, {@code false} if {@code settleSeconds} elapses first.
+     *         {@code elapsedMillis()} is a MEASURED value from the injected {@link #nowMillis}
+     *         clock, never the configured {@code settleSeconds} budget (fleetd #494 follow-up).
      */
-    private boolean waitUntilAtTurnBoundary(String target, int settleSeconds) {
-        long deadline = nowMillis.getAsLong() + TimeUnit.SECONDS.toMillis(settleSeconds);
+    private TurnSettleResult waitUntilAtTurnBoundary(String target, int settleSeconds) {
+        long startMillis = nowMillis.getAsLong();
+        long deadline = startMillis + TimeUnit.SECONDS.toMillis(settleSeconds);
         while (nowMillis.getAsLong() < deadline) {
             AgentStatus status;
             try {
@@ -488,13 +506,20 @@ public final class LeadRollover {
                 status = null;
             }
             if (status == AgentStatus.IDLE || status == AgentStatus.DONE) {
-                return true;
+                return new TurnSettleResult(true, nowMillis.getAsLong() - startMillis);
             }
             settleSleeper.run();
         }
-        return false;
+        return new TurnSettleResult(false, nowMillis.getAsLong() - startMillis);
     }
 
+    /**
+     * The measured outcome of {@link #waitUntilAtTurnBoundary} — fleetd #494 follow-up. The sibling
+     * of {@link ClearSettleResult} for the FIRST wait, which never nudges, so it carries no nudge
+     * count.
+     */
+    private record TurnSettleResult(boolean settled, long elapsedMillis) {}
+
     /**
      * The SECOND wait in {@link #runRollover} — after {@code /clear} has been sent, waits for it to
      * settle, bounded by {@code settleSeconds}. <strong>fleetd #489 — the paste-race fix.</strong>
@@ -524,8 +549,9 @@ public final class LeadRollover {
      *       Claude Code prompt is a no-op, so repeating it is safe;</li>
      *   <li>the {@code PICKUP_GRACE_POLLS}th consecutive such poll, with {@code WORKING} still never
      *       observed, releases rather than wedges the roll instead of nudging again — the same
-     *       choice {@code Injector} makes — and returns {@code true} anyway, logged at {@code info}
-     *       so an operator can see which path ran;</li>
+     *       choice {@code Injector} makes — and returns {@code settled() == true} anyway, logged at
+     *       {@code warn} with the measured elapsed time (fleetd #494) so an operator can see which
+     *       path ran and how long it actually took;</li>
      *   <li>once {@code WORKING} has been observed, nudging stops and this instead waits for a real
      *       {@code working → IDLE/DONE} completion boundary before returning {@code true}.</li>
      * </ul>
@@ -541,15 +567,20 @@ public final class LeadRollover {
      * swallowed and logged at {@code debug}, exactly like {@code Injector.java:437-442} — a failed
      * nudge must not abort the roll.
      *
-     * @return {@code true} once {@code /clear} has settled, or once the nudge budget was exhausted
-     *         with no pickup ever observed (released rather than wedged); {@code false} if {@code
-     *         settleSeconds} elapses first — the caller must NOT send {@code bootstrapText} in that
-     *         case, exactly as before this fix
+     * @return a {@link ClearSettleResult} whose {@code settled()} is {@code true} once {@code
+     *         /clear} has settled, or once the nudge budget was exhausted with no pickup ever
+     *         observed (released rather than wedged); {@code false} if {@code settleSeconds} elapses
+     *         first — the caller must NOT send {@code bootstrapText} in that case, exactly as before
+     *         this fix. {@code elapsedMillis()} and {@code nudges()} are MEASURED values (from the
+     *         injected {@link #nowMillis} clock and an actual nudge count), never the configured
+     *         {@code settleSeconds} budget (fleetd #494).
      */
-    private boolean waitForClearPickupAndSettle(String target, int settleSeconds) {
-        long deadline = nowMillis.getAsLong() + TimeUnit.SECONDS.toMillis(settleSeconds);
+    private ClearSettleResult waitForClearPickupAndSettle(String target, int settleSeconds) {
+        long startMillis = nowMillis.getAsLong();
+        long deadline = startMillis + TimeUnit.SECONDS.toMillis(settleSeconds);
         boolean pickedUp = false; // a WORKING sample has been observed since /clear was sent
         int idlePollsAwaitingPickup = 0;
+        int nudges = 0;
         while (nowMillis.getAsLong() < deadline) {
             AgentStatus status;
             try {
@@ -563,26 +594,56 @@ public final class LeadRollover {
                 pickedUp = true;
             } else if (status == AgentStatus.IDLE || status == AgentStatus.DONE) {
                 if (pickedUp) {
-                    return true; // a real WORKING -> IDLE/DONE completion boundary
+                    // a real WORKING -> IDLE/DONE completion boundary
+                    return new ClearSettleResult(true, nowMillis.getAsLong() - startMillis, nudges);
                 }
                 if (++idlePollsAwaitingPickup >= PICKUP_GRACE_POLLS) {
-                    log.info("lead-rollover: /clear on {} was never observed as WORKING after {} "
+                    long elapsedMillis = nowMillis.getAsLong() - startMillis;
+                    // fleetd #494: this release trades a possibly-unsubmitted /clear for progress
+                    // instead of wedging the roll — that trade is deliberate and stays. But it is
+                    // also exactly the case that reported false success in the real incident (the
+                    // whole roll "succeeded" after 438ms of a 20s budget), so raise it to WARN and
+                    // print the MEASURED elapsed time next to the target pane, not just the count.
+                    //
+                    // fleetd #494 follow-up (2nd pass): BOTH numbers in this line must come from
+                    // the loop's own counters, never from the PICKUP_GRACE_POLLS constant.
+                    // `idlePollsAwaitingPickup` and `nudges` each have exactly one write site in
+                    // this loop, on the same branch, so on this branch they cannot differ from
+                    // PICKUP_GRACE_POLLS / PICKUP_GRACE_POLLS - 1 today — no test can prove the
+                    // difference on this line, and printing the counters does not change that.
+                    // What it does buy: one source of truth instead of two, so a later change to
+                    // the loop (an early return, a second increment site, a different exit
+                    // condition) cannot leave this message reporting a number the loop no longer
+                    // produces. The place where `nudges` genuinely varies with the run — and is
+                    // covered by a test that can tell it apart from a constant — is the
+                    // /clear-timeout warn in runRollover, which prints clearResult.nudges().
+                    log.warn("lead-rollover: /clear on {} was never observed as WORKING after {} "
                                     + "consecutive IDLE/DONE polls ({} of those were nudged) — "
-                                    + "releasing rather than wedging the roll",
-                            target, PICKUP_GRACE_POLLS, PICKUP_GRACE_POLLS - 1);
-                    return true;
+                                    + "releasing rather than wedging the roll (elapsed={}ms)",
+                            target, idlePollsAwaitingPickup, nudges, elapsedMillis);
+                    return new ClearSettleResult(true, elapsedMillis, nudges);
                 }
                 try {
                     agents.submit(target); // nudge a raced Enter (CB-113) so /clear actually submits
                 } catch (RuntimeException e) {
                     log.debug("lead-rollover: resubmit to {} failed (will retry next poll): {}",
                             target, e.getMessage());
+                } finally {
+                    nudges++; // an attempted nudge, whether or not the submit call itself threw
                 }
             }
             // AgentStatus.BLOCKED or UNKNOWN (or an unreadable status, above): neither a pickup
             // signal nor a boundary — keep polling without nudging or releasing.
             settleSleeper.run();
         }
-        return false;
+        return new ClearSettleResult(false, nowMillis.getAsLong() - startMillis, nudges);
     }
+
+    /**
+     * The measured outcome of {@link #waitForClearPickupAndSettle} — fleetd #494. Carries the
+     * MEASURED elapsed time (from the injected {@link #nowMillis} clock) and nudge count alongside
+     * the settle/timeout decision, so callers can log them instead of the configured budget, which
+     * is not how long the wait actually ran.
+     */
+    private record ClearSettleResult(boolean settled, long elapsedMillis, int nudges) {}
 }

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

@@ -19,6 +19,8 @@ import java.util.Set;
 import java.util.concurrent.CompletableFuture;
 import java.util.concurrent.ExecutionException;
 import java.util.concurrent.TimeUnit;
+import java.util.concurrent.atomic.AtomicInteger;
+import java.util.function.LongSupplier;
 
 import static org.junit.jupiter.api.Assertions.*;
 
@@ -521,6 +523,97 @@ class InjectorTest {
         }
     }
 
+    @Test
+    void readinessGraceExpiryLogsTheMeasuredPollCountNextToTheConfiguredBudget() {
+        // fleetd #501, defect 1: READINESS_GRACE_POLLS (240) used to be printed twice — once as
+        // "after {} polls" and once inside the parenthesised budget — even though the loop's own
+        // counter (Target.notReadySincePoll) was in scope at the same call site. On THIS branch the
+        // counter has just reached the threshold, so it equals the constant by construction and this
+        // test cannot tell the two apart — it only pins that the message still carries both a poll
+        // count and a labelled configured budget, using literal numbers (240, 60), never
+        // READINESS_GRACE_POLLS or POLL_INTERVAL_MILLIS, so the assertion can't silently track a
+        // constant change instead of catching a real regression.
+        LoggerContext ctx = (LoggerContext) LoggerFactory.getILoggerFactory();
+        ch.qos.logback.classic.Logger injectorLog =
+                (ch.qos.logback.classic.Logger) LoggerFactory.getLogger(Injector.class);
+        ListAppender<ILoggingEvent> appender = new ListAppender<>();
+        appender.setContext(ctx);
+        appender.start();
+        injectorLog.addAppender(appender);
+        injectorLog.setLevel(Level.WARN);
+        try {
+            Injector inj = new Injector(new AgentControl(herdr), TurnListener.NOOP, _ -> false, _ -> {
+            });
+            inj.enqueue(T, "task", TestTurnTokens.inert(T));
+
+            for (int i = 0; i < READINESS_SAMPLES; i++) inj.onStatus(T, AgentStatus.IDLE);
+
+            String warn = appender.list.stream()
+                    .filter(e -> e.getLevel().equals(Level.WARN))
+                    .map(ILoggingEvent::getFormattedMessage)
+                    .findFirst()
+                    .orElse("no grace-expiry WARN logged");
+            assertTrue(warn.contains("after 240 polls"),
+                    "must print the measured poll count as a plain number: " + warn);
+            assertTrue(warn.contains("configured=240 polls/60s"),
+                    "must print the configured budget, clearly labelled: " + warn);
+        } finally {
+            injectorLog.detachAppender(appender);
+        }
+    }
+
+    @Test
+    void readinessGraceExpiryLogsTheMeasuredElapsedTimeNotArithmeticOnConstants() {
+        // fleetd #501, defect 2: the old line computed "({}s)" as READINESS_GRACE_POLLS *
+        // POLL_INTERVAL_MILLIS / 1000 — arithmetic on two constants, never a measurement, and wrong
+        // in the direction that says everything ran on schedule. This stub clock returns two FIXED
+        // values (1_000ms at the first non-ready sample, 318_412ms at the poll that trips the grace)
+        // whose difference — 317_412ms — does NOT equal 240 * POLL_INTERVAL_MILLIS (=60_000ms).
+        // Asserting on that literal, non-derived number is what makes this test able to fail if the
+        // production code goes back to printing the constant-arithmetic value instead of the
+        // injected clock's measurement.
+        long[] readings = {1_000L, 318_412L};
+        AtomicInteger call = new AtomicInteger(0);
+        LongSupplier stubClock = () -> {
+            int i = call.getAndIncrement();
+            if (i >= readings.length) {
+                throw new AssertionError("nowMillis read more times than this fixture expects (" + i
+                        + "); the readiness-not-ready branch should read the clock exactly twice — "
+                        + "once to stamp the first non-ready sample, once at grace expiry");
+            }
+            return readings[i];
+        };
+
+        LoggerContext ctx = (LoggerContext) LoggerFactory.getILoggerFactory();
+        ch.qos.logback.classic.Logger injectorLog =
+                (ch.qos.logback.classic.Logger) LoggerFactory.getLogger(Injector.class);
+        ListAppender<ILoggingEvent> appender = new ListAppender<>();
+        appender.setContext(ctx);
+        appender.start();
+        injectorLog.addAppender(appender);
+        injectorLog.setLevel(Level.WARN);
+        try {
+            Injector inj = new Injector(new AgentControl(herdr), TurnListener.NOOP, _ -> false, _ -> {
+            }, stubClock);
+            inj.enqueue(T, "task", TestTurnTokens.inert(T));
+
+            for (int i = 0; i < READINESS_SAMPLES; i++) inj.onStatus(T, AgentStatus.IDLE);
+
+            String warn = appender.list.stream()
+                    .filter(e -> e.getLevel().equals(Level.WARN))
+                    .map(ILoggingEvent::getFormattedMessage)
+                    .findFirst()
+                    .orElse("no grace-expiry WARN logged");
+            assertTrue(warn.contains("elapsed=317412ms"), "must print the MEASURED elapsed time from "
+                    + "the injected clock (318412 - 1000 = 317412), not an arithmetic value: " + warn);
+            assertFalse(warn.contains("elapsed=60000ms"), "must not print "
+                    + "READINESS_GRACE_POLLS * POLL_INTERVAL_MILLIS (240 * 250 = 60000ms) as if it "
+                    + "were the measured elapsed time: " + warn);
+        } finally {
+            injectorLog.detachAppender(appender);
+        }
+    }
+
     @Test
     void aWorkerThatBecomesReadyWithinTheGraceIsDeliveredNormally() {
         // The readiness grace must not fail a worker that is merely slow to boot: once it becomes

fleetd/src/test/java/dev/ltms/fleet/lead/LeadRolloverTest.java

@@ -1,5 +1,9 @@
 package dev.ltms.fleet.lead;
 
+import ch.qos.logback.classic.Level;
+import ch.qos.logback.classic.Logger;
+import ch.qos.logback.classic.spi.ILoggingEvent;
+import ch.qos.logback.core.read.ListAppender;
 import com.fasterxml.jackson.databind.JsonNode;
 import dev.ltms.fleet.config.FleetConfig;
 import dev.ltms.fleet.herdr.AgentControl;
@@ -9,6 +13,7 @@ import dev.ltms.fleet.herdr.HerdrException;
 import org.junit.jupiter.api.DisplayName;
 import org.junit.jupiter.api.Test;
 import org.junit.jupiter.api.io.TempDir;
+import org.slf4j.LoggerFactory;
 
 import java.io.IOException;
 import java.nio.file.Files;
@@ -854,4 +859,207 @@ class LeadRolloverTest {
         assertFalse(bootstrapSent.contains("\"handover.md\""),
                 "must not name the raw relative configured value in the text actually sent");
     }
+
+    // ---- fleetd #494: the log lines must print MEASURED values, never the configured budget --
+
+    private static ListAppender<ILoggingEvent> attachLog() {
+        Logger logger = (Logger) LoggerFactory.getLogger(LeadRollover.class);
+        logger.setLevel(Level.DEBUG);
+        ListAppender<ILoggingEvent> appender = new ListAppender<>();
+        appender.start();
+        logger.addAppender(appender);
+        return appender;
+    }
+
+    private static void detachLog(ListAppender<ILoggingEvent> appender) {
+        ((Logger) LoggerFactory.getLogger(LeadRollover.class)).detachAppender(appender);
+    }
+
+    private static ILoggingEvent lastEventContaining(ListAppender<ILoggingEvent> events, String substring) {
+        return events.list.stream()
+                .filter(e -> e.getFormattedMessage().contains(substring))
+                .reduce((_, b) -> b)
+                .orElseThrow(() -> new AssertionError("no log event contained \"" + substring
+                        + "\"; got: " + events.list.stream().map(ILoggingEvent::getFormattedMessage).toList()));
+    }
+
+    @Test
+    @DisplayName("[fleetd #494] the /clear-timeout warn line prints the MEASURED elapsed time and "
+            + "nudge count next to the configured budget, never the configured value alone")
+    void clearTimeoutLogPrintsMeasuredElapsedAndNudgesNotJustConfigured() throws IOException {
+        // Idle until /clear is sent, then permanently WORKING (a genuinely stuck /clear that never
+        // reaches a completion boundary) — isolates the SECOND wait exactly like
+        // clearThatNeverSettlesAfterwardsNeverSendsBootstrapText, but with a clock that ADVANCES on
+        // every read so the measured elapsed time is a deterministic, non-zero value distinct from
+        // the configured budget — pinning fleetd #494's fix, not just its absence of a hang.
+        FakeHerdr fake = new FakeHerdr();
+        HerdrClient flipsAfterClear = new HerdrClient() {
+            @Override
+            public JsonNode call(String method, Object params) throws HerdrException {
+                JsonNode result = fake.call(method, params);
+                if ("agent.prompt".equals(method) && String.valueOf(params).contains("/clear")) {
+                    fake.agentStatus("working");
+                }
+                return result;
+            }
+
+            @Override
+            public void close() {
+                fake.close();
+            }
+        };
+        Path handover = writeHandover("handover contents");
+        FleetConfig.LeadRollover config =
+                new FleetConfig.LeadRollover(handover.toString(), true, 3600, 20, 1 /*clearSettleSeconds*/, "boot text");
+        AtomicLong clock = new AtomicLong(1_000);
+        LeadRollover rollover = newRollover(flipsAfterClear, config, () -> clock.addAndGet(500));
+
+        ListAppender<ILoggingEvent> events = attachLog();
+        try {
+            LeadRollover.PendingRollover pending = rollover.open(LEAD, "context is full");
+            LeadRollover.RollDecision decision = rollover.confirm(LEAD, pending.token(), true);
+
+            assertTrue(decision.accepted(), "every synchronous gate passes; the refusal is logged "
+                    + "only, deep inside the deferred continuation");
+
+            ILoggingEvent event = lastEventContaining(events, "NOT sending bootstrapText");
+            assertEquals(Level.WARN, event.getLevel());
+            String message = event.getFormattedMessage();
+            assertTrue(message.contains("configured=1s"), "must label the configured budget: " + message);
+            assertTrue(message.contains("elapsed=1500ms"), "must print the MEASURED elapsed time — "
+                    + "with this fixture's advancing clock, the wait actually ran 1500ms against a "
+                    + "1s(=1000ms) configured budget: " + message);
+            assertTrue(message.contains("nudges=0"), "must print the measured nudge count (0 here — "
+                    + "the pane was WORKING throughout, never IDLE/DONE, so no nudge was ever sent): "
+                    + message);
+            assertFalse(message.contains("within 1s"), "must not present the configured budget as if "
+                    + "it were the measured wait duration: " + message);
+        } finally {
+            detachLog(events);
+        }
+    }
+
+    @Test
+    @DisplayName("[fleetd #494] the /clear pickup-grace release line is WARN (was INFO) and prints "
+            + "the measured elapsed time next to the target pane")
+    void clearGraceReleaseLogIsWarnWithMeasuredElapsed() throws IOException {
+        // Default idle throughout — no WORKING sample is ever observed, so the pickup-grace wait
+        // exhausts PICKUP_GRACE_POLLS and releases rather than wedging (see
+        // clearPickupIsNudgedBeforeBootstrapTextWhenPaneStaysIdle for the un-logged half of this
+        // scenario). A self-advancing clock makes the measured elapsed time deterministic and
+        // provably distinct from a bare poll/nudge count.
+        FakeHerdr herdr = new FakeHerdr();
+        Path handover = writeHandover("handover contents");
+        FleetConfig.LeadRollover config = cfg(handover.toString()); // turnSettleSeconds=clearSettleSeconds=20
+        AtomicLong clock = new AtomicLong(1_000);
+        LeadRollover rollover = newRollover(herdr, config, () -> clock.addAndGet(500));
+
+        ListAppender<ILoggingEvent> events = attachLog();
+        try {
+            LeadRollover.PendingRollover pending = rollover.open(LEAD, "context is full");
+            LeadRollover.RollDecision decision = rollover.confirm(LEAD, pending.token(), true);
+
+            assertTrue(decision.accepted(), "expected approval; got: " + decision.reason()
+                    + " / " + decision.detail());
+
+            ILoggingEvent event = lastEventContaining(events, "releasing rather than wedging the roll");
+            assertEquals(Level.WARN, event.getLevel(), "the grace-limit release must be WARN, not "
+                    + "INFO — it is exactly the case that reported false success in the real incident "
+                    + "this fix comes from (a roll that 'succeeded' after 438ms of a 20s budget)");
+            String message = event.getFormattedMessage();
+            assertTrue(message.contains(LEAD), "must name the target pane: " + message);
+            assertTrue(message.contains("elapsed=4500ms"), "must print the MEASURED elapsed time — "
+                    + "with this fixture's advancing clock, the wait ran 4500ms before releasing: "
+                    + message);
+            // fleetd #494 follow-up (2nd pass): both numbers here are DELIBERATE plain literals,
+            // not derived from LeadRollover.PICKUP_GRACE_POLLS. A version of this assertion that
+            // reads "(" + (LeadRollover.PICKUP_GRACE_POLLS - 1) + " of those were nudged)" builds
+            // its expectation the same way the production code used to build the log line, so it
+            // cannot tell a fixed constant apart from the measured counter — proved by reverting
+            // the production fix and re-running: that mutant stayed green under the old assertion.
+            // If PICKUP_GRACE_POLLS ever changes, THIS TEST MUST FAIL and a human must look at the
+            // new message and update the literals below, not just re-derive them.
+            assertTrue(message.contains("after 8 consecutive IDLE/DONE polls (7 of those were nudged)"),
+                    "must print the measured poll count and nudge count as plain numbers, not the "
+                            + "PICKUP_GRACE_POLLS constant standing in for either: " + message);
+        } finally {
+            detachLog(events);
+        }
+    }
+
+    @Test
+    @DisplayName("[fleetd #494] the success line prints the measured elapsed time for the whole roll")
+    void successLogPrintsMeasuredElapsedForTheWholeRoll() throws IOException {
+        // Same fixture as clearGraceReleaseLogIsWarnWithMeasuredElapsed: default idle throughout, so
+        // the grace release fires and the roll still goes on to send bootstrapText and log success.
+        // This is deliberately the SAME shape as the real incident (a roll that "succeeds" quickly)
+        // — the missing signal was the elapsed time on this exact line.
+        FakeHerdr herdr = new FakeHerdr();
+        Path handover = writeHandover("handover contents");
+        FleetConfig.LeadRollover config = cfg(handover.toString());
+        AtomicLong clock = new AtomicLong(1_000);
+        LeadRollover rollover = newRollover(herdr, config, () -> clock.addAndGet(500));
+
+        ListAppender<ILoggingEvent> events = attachLog();
+        try {
+            LeadRollover.PendingRollover pending = rollover.open(LEAD, "context is full");
+            LeadRollover.RollDecision decision = rollover.confirm(LEAD, pending.token(), true);
+
+            assertTrue(decision.accepted(), "expected approval; got: " + decision.reason()
+                    + " / " + decision.detail());
+
+            ILoggingEvent event = lastEventContaining(events, "lead-rollover: rolled");
+            assertEquals(Level.INFO, event.getLevel());
+            String message = event.getFormattedMessage();
+            // fleetd #494 follow-up: waitUntilAtTurnBoundary now also reads the injected clock one
+            // extra time (to compute ITS OWN measured elapsed on the success path), so the shared
+            // fixture clock advances by one more 500ms tick before the roll finishes than it did
+            // before that follow-up — 7000ms, not 6500ms.
+            assertTrue(message.contains("elapsedMs=7000"), "must print the MEASURED elapsed time for "
+                    + "the whole roll — with this fixture's advancing clock, the full roll (turn-settle "
+                    + "wait + /clear wait + bootstrapText) took 7000ms: " + message);
+        } finally {
+            detachLog(events);
+        }
+    }
+
+    @Test
+    @DisplayName("[fleetd #494 follow-up] the turn-settle timeout warn line prints the MEASURED "
+            + "elapsed time next to the configured budget, never the configured value alone")
+    void turnTimeoutLogPrintsMeasuredElapsedNotJustConfigured() throws IOException {
+        // The brief that named the four items this ticket fixed left this exact sibling line out —
+        // "did not reach a turn boundary (IDLE or DONE) within {}s after confirm()" — even though it
+        // has the identical defect shape one method below. Same fixture shape as
+        // clearTimeoutLogPrintsMeasuredElapsedAndNudgesNotJustConfigured, but for the FIRST wait: the
+        // calling lead's own pane never goes idle, so waitUntilAtTurnBoundary times out.
+        FakeHerdr herdr = new FakeHerdr();
+        herdr.agentStatus("working"); // the calling lead's own pane — never goes idle in this test
+        Path handover = writeHandover("handover contents");
+        FleetConfig.LeadRollover config =
+                new FleetConfig.LeadRollover(handover.toString(), true, 3600, 1 /*turnSettleSeconds*/, 20, "text");
+        AtomicLong clock = new AtomicLong(1_000);
+        LeadRollover rollover = newRollover(herdr, config, () -> clock.addAndGet(500));
+
+        ListAppender<ILoggingEvent> events = attachLog();
+        try {
+            LeadRollover.PendingRollover pending = rollover.open(LEAD, "context is full");
+            LeadRollover.RollDecision decision = rollover.confirm(LEAD, pending.token(), true);
+
+            assertTrue(decision.accepted(), "every synchronous gate should pass; the refusal happens "
+                    + "only inside the deferred continuation, which this test's synchronous runner "
+                    + "has already run to completion by the time confirm() returns");
+
+            ILoggingEvent event = lastEventContaining(events, "refusing to send /clear at all");
+            assertEquals(Level.WARN, event.getLevel());
+            String message = event.getFormattedMessage();
+            assertTrue(message.contains("configured=1s"), "must label the configured budget: " + message);
+            assertTrue(message.contains("elapsed=1500ms"), "must print the MEASURED elapsed time — "
+                    + "with this fixture's advancing clock, the wait actually ran 1500ms against a "
+                    + "1s(=1000ms) configured budget: " + message);
+            assertFalse(message.contains("within 1s"), "must not present the configured budget as if "
+                    + "it were the measured wait duration: " + message);
+        } finally {
+            detachLog(events);
+        }
+    }
 }