diff --git a/quartz/src/commonMain/kotlin/com/vitorpamplona/quartz/nip01Core/relay/client/INostrClient.kt b/quartz/src/commonMain/kotlin/com/vitorpamplona/quartz/nip01Core/relay/client/INostrClient.kt
index 0596190836..b77d033c7d 100644
--- a/quartz/src/commonMain/kotlin/com/vitorpamplona/quartz/nip01Core/relay/client/INostrClient.kt
+++ b/quartz/src/commonMain/kotlin/com/vitorpamplona/quartz/nip01Core/relay/client/INostrClient.kt
@@ -57,6 +57,22 @@ interface INostrClient : AutoCloseable {
      */
     fun resetBackoff() { }
 
+    /**
+     * Puts [url] back in the connection pool if it is no longer there, without dialing it.
+     *
+     * [reconnect] can only act on relays the pool still holds, so a caller that means to
+     * revive one it stopped hearing from has to restore that precondition first —
+     * otherwise the reconnect iterates past an empty pool and does nothing at all, which
+     * is indistinguishable from a relay that was asked and stayed silent.
+     *
+     * A relay leaves the pool when nothing wants it any more (no subscription, no count,
+     * no pending publish). That is normally the right call and normally permanent, so
+     * this is deliberately narrow: it restores membership and leaves dialing, backoff and
+     * filter syncing to [reconnect]. Defaults to a no-op — a client with no pool has
+     * nothing to restore and should not be forced to implement one.
+     */
+    fun ensureInPool(url: NormalizedRelayUrl) { }
+
     fun isActive(): Boolean
 
     /**
diff --git a/quartz/src/commonMain/kotlin/com/vitorpamplona/quartz/nip01Core/relay/client/NostrClient.kt b/quartz/src/commonMain/kotlin/com/vitorpamplona/quartz/nip01Core/relay/client/NostrClient.kt
index 5b75850ea3..21c16d9b64 100644
--- a/quartz/src/commonMain/kotlin/com/vitorpamplona/quartz/nip01Core/relay/client/NostrClient.kt
+++ b/quartz/src/commonMain/kotlin/com/vitorpamplona/quartz/nip01Core/relay/client/NostrClient.kt
@@ -274,6 +274,12 @@ class NostrClient(
         relayPool.resetBackoff()
     }
 
+    override fun ensureInPool(url: NormalizedRelayUrl) {
+        // Membership only. Connecting is reconnect()'s job, and going through the pool's
+        // own create keeps the relay client identical to the one publish would have made.
+        relayPool.createRelayIfAbsent(url)
+    }
+
     override fun subscribe(
         subId: String,
         filters: Map<NormalizedRelayUrl, List<Filter>>,
diff --git a/quartz/src/commonMain/kotlin/com/vitorpamplona/quartz/nip01Core/relay/client/accessories/NostrClientPublishExt.kt b/quartz/src/commonMain/kotlin/com/vitorpamplona/quartz/nip01Core/relay/client/accessories/NostrClientPublishExt.kt
index 222e743746..4a4428f890 100644
--- a/quartz/src/commonMain/kotlin/com/vitorpamplona/quartz/nip01Core/relay/client/accessories/NostrClientPublishExt.kt
+++ b/quartz/src/commonMain/kotlin/com/vitorpamplona/quartz/nip01Core/relay/client/accessories/NostrClientPublishExt.kt
@@ -273,9 +273,20 @@ suspend fun INostrClient.publishAndCollectResults(
                                         }
                                         // The event is still in the pool's outbox for this relay,
                                         // so the dial is the whole job: the pool flushes what it
-                                        // owes the relay once the socket is back. Ignore the
-                                        // accumulated backoff — this is a user-visible publish
-                                        // waiting on it, not a background refresh.
+                                        // owes the relay once the socket is back.
+                                        //
+                                        // Except that a reconnect can only dial relays the pool
+                                        // still holds, and a relay can leave it — the desired set
+                                        // is sampled, so a snapshot taken before this publish
+                                        // claimed the relay retires it, socket and all. Restoring
+                                        // membership first is what keeps the retry from being a
+                                        // silent no-op: issued, logged, dialing nothing, and
+                                        // reported at the deadline as the hang-up we already knew
+                                        // about. A no-op when the relay is still there, which is
+                                        // the ordinary case.
+                                        ensureInPool(result.relay)
+                                        // Ignore the accumulated backoff — this is a user-visible
+                                        // publish waiting on it, not a background refresh.
                                         resetBackoff()
                                         reconnect(onlyIfChanged = false, ignoreRetryDelays = true)
                                     }