diff --git a/MODERNIZATION.md b/MODERNIZATION.md
index 8456deaf..e922cc62 100644
--- a/MODERNIZATION.md
+++ b/MODERNIZATION.md
@@ -188,7 +188,26 @@ Android NDK r28+.
   `mMutex` and rtc's callback mutex in `close()`/`destroy()` (teardown
   deadlock; fixed by calling into rtc outside `mMutex`). Reproduced 3/200
   resp. 2/100 locally before the fix; 0 failures in 925 stress runs
-  (Debug + Release) after.
+  (Debug + Release) after. **A third, rarer mechanism** surfaced on 2-core
+  Linux runners (ConnectionLockingTest.CanLockConnections, PR #42 CI,
+  2026-07-04, twice in a row then pass; never reproduced in 100 stress runs
+  under 10x load on macOS): on the outgoing side the connector starts socket
+  I/O inside `createConnection()`, but the endpoint's listeners on the
+  returned `PendingConnection` are only registered afterwards
+  (`addEndpoint` -> `changeToConnectingState`). If the main thread is
+  descheduled in that window for longer than one localhost websocket +
+  streamr handshake round-trip — realistic only on a loaded 2-core box —
+  `onHandshakeCompleted()` runs first; it disarms the 15 s connect watchdog
+  and then emits `Connected` on a fire-and-forget emitter with zero
+  listeners. The emission is lost, no timeout remains armed, and the
+  endpoint sits in the connecting state forever with the lock-request RPC
+  buffered (signature: rpc FutureTimeout after 10 s, waitForCondition
+  timeout, `hasConnection() == false`). Fixed by making `IPendingConnection`
+  a `ReplayEventEmitter` (late listeners receive the latest
+  `Connected`/`Disconnected`) and by making `emit()`'s latest-event store +
+  handler snapshot atomic with respect to `on()` in the eventemitter, so a
+  concurrently registered listener gets exactly one of live dispatch or
+  replay.
 - **Gate**: build/test green macOS + Linux, **and iOS cross-build +
   `iostest.sh` green — the compiler's output must stay compatible with the
   device's fixed libc++ runtime**.
diff --git a/packages/streamr-dht/include/streamr-dht/connection/IPendingConnection.hpp b/packages/streamr-dht/include/streamr-dht/connection/IPendingConnection.hpp
index 6c0eb6c3..cb52e0c3 100644
--- a/packages/streamr-dht/include/streamr-dht/connection/IPendingConnection.hpp
+++ b/packages/streamr-dht/include/streamr-dht/connection/IPendingConnection.hpp
@@ -21,8 +21,18 @@ using PendingConnectionEvents = std::tuple<
     pendingconnectionevents::Connected,
     pendingconnectionevents::Disconnected>;
 
-class IPendingConnection
-    : public streamr::eventemitter::EventEmitter<PendingConnectionEvents> {
+// ReplayEventEmitter, not the fire-and-forget EventEmitter: the connectors
+// start socket I/O inside createConnection(), but the Connected/Disconnected
+// listeners are only registered afterwards, when ConnectionManager wraps the
+// returned pending connection in an Endpoint (addEndpoint ->
+// changeToConnectingState). On a loaded machine the entire websocket +
+// streamr handshake can win that race, and onHandshakeCompleted() disarms
+// the connect watchdog before emitting — with a fire-and-forget emitter the
+// emission is lost and the endpoint stays in the connecting state forever
+// (the CanLockConnections stall on 2-core CI runners). Replay delivers the
+// missed emission to the late listener instead.
+class IPendingConnection : public streamr::eventemitter::ReplayEventEmitter<
+                               PendingConnectionEvents> {
 public:
     ~IPendingConnection() override = default;
     virtual void onHandshakeCompleted(
diff --git a/packages/streamr-dht/test/unit/PendingConnectionTest.cpp b/packages/streamr-dht/test/unit/PendingConnectionTest.cpp
index 1ff54073..d2d9f681 100644
--- a/packages/streamr-dht/test/unit/PendingConnectionTest.cpp
+++ b/packages/streamr-dht/test/unit/PendingConnectionTest.cpp
@@ -93,6 +93,35 @@ TEST_F(PendingConnectionTest, EmitsConnected) {
     pendingConnection->onHandshakeCompleted(dummyConnection);
 }
 
+// Regression tests for the CanLockConnections stall on slow CI runners:
+// the connectors start socket I/O before ConnectionManager registers the
+// endpoint's listeners on the pending connection, so a fast handshake can
+// complete before registration. IPendingConnection must therefore replay
+// the latest emission to late listeners.
+
+TEST_F(PendingConnectionTest, ReplaysConnectedToLateListener) {
+    auto dummyConnection = std::make_shared<DummyConnection>();
+    // handshake completes before anyone listens
+    pendingConnection->onHandshakeCompleted(dummyConnection);
+    bool isEmitted = false;
+    pendingConnection->once<Connected>(
+        [&](const PeerDescriptor& /* peerDescriptor */,
+            std::shared_ptr<Connection> connection) { // NOLINT
+            isEmitted = true;
+            EXPECT_EQ(dummyConnection, connection);
+        });
+    EXPECT_TRUE(isEmitted);
+}
+
+TEST_F(PendingConnectionTest, ReplaysDisconnectedToLateListener) {
+    // connection fails before anyone listens
+    pendingConnection->close(false);
+    bool isEmitted = false;
+    pendingConnection->once<Disconnected>(
+        [&](bool /* gracefulLeave */) { isEmitted = true; });
+    EXPECT_TRUE(isEmitted);
+}
+
 TEST_F(PendingConnectionTest, DoesNotEmitConnectedIfReplaced) {
     bool isEmitted = false;
     pendingConnection->once<Connected>(
diff --git a/packages/streamr-eventemitter/include/streamr-eventemitter/EventEmitter.hpp b/packages/streamr-eventemitter/include/streamr-eventemitter/EventEmitter.hpp
index 968e0f87..05cdda43 100644
--- a/packages/streamr-eventemitter/include/streamr-eventemitter/EventEmitter.hpp
+++ b/packages/streamr-eventemitter/include/streamr-eventemitter/EventEmitter.hpp
@@ -287,13 +287,25 @@ class EventEmitterImpl {
         MatchingEventType<EmitterEventType> EventType,
         typename... EventArgs>
     void emit(EventArgs... args) {
-        if constexpr (ReplayLatestEventToNewListeners) {
-            StoredEvent<typename EmitterEventType::ArgumentTypes> storedEvent(
-                (args)...);
-            mLatestEvent = std::move(storedEvent);
-        }
         std::lock_guard guard{mEmitLoopMutex};
-        createExecutingEmitLoopHandlersMap();
+        {
+            // The latest-event store and the handler snapshot must form one
+            // atomic step with respect to on(): on() checks mLatestEvent and
+            // registers the handler while holding mEventHandlersMutex, so a
+            // listener registered concurrently with this emit either makes
+            // it into the snapshot (live dispatch) or observes the stored
+            // event (replay) — never neither, never both. mEventHandlersMutex
+            // is only ever taken inside mEmitLoopMutex here, matching the
+            // once-handler removal below, so no new lock order is introduced.
+            std::lock_guard handlersGuard{mEventHandlersMutex};
+            if constexpr (ReplayLatestEventToNewListeners) {
+                StoredEvent<typename EmitterEventType::ArgumentTypes>
+                storedEvent((args)...);
+                std::lock_guard latestGuard{mLatestEventMutex};
+                mLatestEvent = std::move(storedEvent);
+            }
+            createExecutingEmitLoopHandlersMap();
+        }
 
         while (auto ret = popHandlerFromExecutingEmitLoopHandlersMap()) {
             auto handler = ret.value();