diff --git a/WORK-RECORDS.md b/WORK-RECORDS.md
index 86302df..b78789d 100644
--- a/WORK-RECORDS.md
+++ b/WORK-RECORDS.md
@@ -9,12 +9,12 @@
 | Kind | ID | Status | Lane | Source |
 | --- | --- | --- | --- | --- |
 | workplan | RCP-WP-0001 | finished | — | workplans/RCP-WP-0001-statehub-bootstrap.md |
-| workplan | RCP-WP-0002 | finished | — | workplans/RCP-WP-0002-first-deployment.md |
+| workplan | RCP-WP-0002 | active | — | workplans/RCP-WP-0002-first-deployment.md |
 | task | RCP-WP-0001-T01 | done | — | workplans/RCP-WP-0001-statehub-bootstrap.md |
 | task | RCP-WP-0001-T02 | done | — | workplans/RCP-WP-0001-statehub-bootstrap.md |
 | task | RCP-WP-0001-T03 | done | — | workplans/RCP-WP-0001-statehub-bootstrap.md |
 | task | RCP-WP-0002-T01 | done | — | workplans/RCP-WP-0002-first-deployment.md |
 | task | RCP-WP-0002-T02 | done | — | workplans/RCP-WP-0002-first-deployment.md |
 | task | RCP-WP-0002-T03 | done | — | workplans/RCP-WP-0002-first-deployment.md |
-| task | RCP-WP-0002-T04 | done | — | workplans/RCP-WP-0002-first-deployment.md |
+| task | RCP-WP-0002-T04 | progress | — | workplans/RCP-WP-0002-first-deployment.md |
 | task | RCP-WP-0002-T05 | wait | — | workplans/RCP-WP-0002-first-deployment.md |
diff --git a/docs/evidence/RCP-WP-0002-T04-first-deployment-2026-09-08.md b/docs/evidence/RCP-WP-0002-T04-first-deployment-2026-09-08.md
index fa0d0df..a9a0f7b 100644
--- a/docs/evidence/RCP-WP-0002-T04-first-deployment-2026-09-08.md
+++ b/docs/evidence/RCP-WP-0002-T04-first-deployment-2026-09-08.md
@@ -74,3 +74,34 @@ lease window cannot distinguish a service that works from one that works *once*.
 The criterion for `verified` is therefore observation across a full rotation,
 not a passing check at a single instant — an availability property is not
 provable by one sample.
+
+
+## The watcher was also wrong
+
+The first lease watch returned **FAILED** on one sample out of 21. That sample
+was the instrument, not the service: it treated an empty `kubectl` result as
+"not ready", so a transient API hiccup was recorded as an outage.
+
+The service had returned **zero** 503s across the whole window. The readiness
+probe fires every 5s, so a genuine two-minute outage would have left roughly
+twenty-four of them, and the pod ran 46 minutes past a 30-minute lease with no
+restarts.
+
+I did not override the verdict by argument — a verdict that can be talked around
+is worth nothing. The watcher is fixed (`tools/lease-watch.sh`) to distinguish
+a failed *query* from a failed *service*, and to corroborate its own sampling
+against the kubelet's probe history, which is a far better instrument than one
+sample a minute. Then re-measured.
+
+**Three checks in this rollout were themselves defective**, all the same shape —
+unable to distinguish their own failure from the failure they were watching for:
+
+| Check | How it lied |
+|---|---|
+| `live-image-digest-match` | line-offset `grep` returned empty; reported "not pinned yet" while a digest was pinned |
+| `check_readiness` (earlier) | one `except` reported "database unreachable" for a reachable but unmigrated database |
+| lease watcher v1 | empty `kubectl` output counted as an outage |
+
+A verification step that cannot fail correctly is worse than none, because it is
+trusted. That is the durable lesson from this deployment, more than any single
+defect it found.
diff --git a/tools/lease-watch.sh b/tools/lease-watch.sh
new file mode 100755
index 0000000..7f8f5b1
--- /dev/null
+++ b/tools/lease-watch.sh
@@ -0,0 +1,46 @@
+#!/usr/bin/env bash
+# Verification for RCP-WP-0002-T04: stay ready across a credential lease
+# rotation (runtime lease TTL 30m).
+#
+# v2. The first version treated an empty kubectl result as "not ready", so a
+# transient API hiccup was recorded as a service failure and produced a FAILED
+# verdict for a service that never returned a single 503. A check that cannot
+# tell its own failure from the failure it watches for is worse than no check.
+OUT="$1"; MINUTES="${2:-20}"
+DEADLINE=$(( $(date +%s) + MINUTES * 60 ))
+: > "$OUT"
+echo "started $(date -Is) — ${MINUTES}m watch (runtime lease TTL 30m)" >> "$OUT"
+NOTREADY=0; QUERYFAIL=0; SAMPLES=0
+while [ "$(date +%s)" -lt "$DEADLINE" ]; do
+  SAMPLES=$((SAMPLES+1))
+  if ! OUTJSON=$(kubectl -n canned-prompts get deploy canned-prompts \
+        -o jsonpath='{.status.readyReplicas}|{.status.replicas}' 2>/dev/null) || [ -z "${OUTJSON%%|*}${OUTJSON##*|}" ]; then
+    QUERYFAIL=$((QUERYFAIL+1))
+    echo "$(date -Is) QUERY-FAILED (instrument, not service)" >> "$OUT"
+  else
+    READY="${OUTJSON%%|*}"; WANT="${OUTJSON##*|}"
+    if [ "$READY" = "$WANT" ] && [ -n "$READY" ]; then
+      echo "$(date -Is) ready=$READY/$WANT" >> "$OUT"
+    else
+      NOTREADY=$((NOTREADY+1))
+      echo "$(date -Is) NOT READY ready=${READY:-0}/${WANT:-?}" >> "$OUT"
+    fi
+  fi
+  sleep 60
+done
+# Corroborate with the kubelet, which probes every 5s and is a far better
+# instrument than this loop's 1/min sampling.
+P503=$(kubectl -n canned-prompts logs deploy/canned-prompts --since="${MINUTES}m" 2>/dev/null | grep -c '" 503' || echo "?")
+UP=$(kubectl -n canned-prompts get pod -l app.kubernetes.io/name=canned-prompts -o jsonpath='{.items[0].status.startTime}' 2>/dev/null)
+R=$(kubectl -n canned-prompts get pod -l app.kubernetes.io/name=canned-prompts -o jsonpath='{.items[0].status.containerStatuses[0].restartCount}' 2>/dev/null)
+{
+  echo "finished $(date -Is)"
+  echo "samples=$SAMPLES not_ready=$NOTREADY query_failed=$QUERYFAIL"
+  echo "kubelet readiness 503s in window: $P503 (probe every 5s)"
+  echo "pod started $UP, restarts=$R"
+  if [ "$NOTREADY" -eq 0 ] && [ "$P503" = "0" ]; then
+    echo "RESULT: survived — continuously ready past the lease TTL"
+  else
+    echo "RESULT: FAILED"
+  fi
+} >> "$OUT"