Index: docs/SDD/Seq_RealtimeDataTransfer.puml
===================================================================
diff -u -r6d2ef8c97f4bb34204e95811839b3995000c47c1 -r8b98ff34313ee5d3f5d49672d622162acfc755c9
--- docs/SDD/Seq_RealtimeDataTransfer.puml	(.../Seq_RealtimeDataTransfer.puml)	(revision 6d2ef8c97f4bb34204e95811839b3995000c47c1)
+++ docs/SDD/Seq_RealtimeDataTransfer.puml	(.../Seq_RealtimeDataTransfer.puml)	(revision 8b98ff34313ee5d3f5d49672d622162acfc755c9)
@@ -2,52 +2,90 @@
 
 title Real-time Data Transfer — End-to-End Scenario
 
-participant "CANDumpPlayer" as CDP
 participant "CAN Bus\n(SocketCAN)" as CANBUS
-participant "CloudConnect\n(CloudConnectController)" as LRC
-participant "CloudConnectServer" as RS
-participant "Luis application" as APP
-participant "Cloud pipeline\n(planned)" as PIPE
+participant "CloudConnect\n(CloudConnectController)" as CC
+participant "AWS IoT Core\n(MQTT broker)" as AWS
 
-note over CDP
-  Replays a candump log onto SocketCAN
-  at a configurable speed multiplier.
-  See Seq_CANDumpPlayer for detail.
+== Startup (main.cpp, on the controller thread) ==
+
+CC -> CC : listenForApp()\nbind /tmp/cloudconnect.sock
+CC -> AWS : connectToCloud() → MqttClient::open()\nconnectToHostEncrypted(), mTLS, MQTT 3.1.1, port 8883
+AWS --> CC : CONNACK
+CC -> CC : onCloudStateChanged(Connected)\nreset CAN/MQTT stats counters
+CC -> AWS : SUBSCRIBE {prefix}/{deviceId}/clinical (QoS 1)
+CC -> CC : startCan() → CanInterface::init()
+
+note over CC
+  Order is deliberate: the app socket and the broker
+  connection come up before CAN, so frames arriving
+  before CONNACK are dropped rather than queued.
 end note
 
-CDP -> CANBUS : writeFrame(can_id, payload)
-CANBUS -> LRC : didFrameReceive(QCanBusFrame)
+== Steady state: CAN frame → cloud ==
 
-LRC -> LRC : MessageDispatcher reassembles\nmulti-frame Can::Message\n(see Seq_CANToCloud for detail)
+CANBUS -> CC : didFrameReceive(QCanBusFrame)
 
-alt action == Drop (or msgId not in _msgHandling)
-    LRC -> LRC : discard
+CC -> CC : MessageDispatcher::onFrameReceive(canId, payload)\nreassembles multi-frame Can::Message\n→ didActionReceive(Can::Message)
+
+CC -> CC : record CAN stats\n(count, inter-arrival, running average)
+
+alt msgId not in _canRouting
+    CC -> CC : discard (no action defined)
+else action == Drop
+    CC -> CC : discard
 else action == SendDelta AND payload unchanged\n(cached data.chopped(1) == new data.chopped(1))
-    LRC -> LRC : discard (no change)
+    CC -> CC : discard (no change)
 else action == SendAlways OR SendDelta with new payload
-    LRC -> LRC : canMessageToProtobufByteArray()\n→ serialised protobuf bytes
+    CC -> CC : leahi::canMessageToProtobufByteArray()\n→ serialised protobuf bytes
 
-    LRC -> RS : send(topic, seq++, protobuf payload)
-    RS -> APP : CloudConnectFrame frame over\n/tmp/cloudconnect.sock
+    alt serialisation succeeded
+        CC -> AWS : PUBLISH (QoS 1, retain=false)\ntopic {prefix}/{deviceId}/clinical
 
-    LRC -->> PIPE : route → spool → publish
+        alt publish accepted by the client
+            CC -> CC : _inflight[pubId] = { Sent, msgId, sequence, sentUs }
+            AWS --> CC : PUBACK
+            CC -> CC : onCloudMessageStatusChanged(Acknowledged)\nmark ACK, record publish→ack latency
+        else publish rejected
+            CC -> CC : log warning, message lost\n(no spool today)
+        end
+    else serialisation failed
+        CC -> CC : log warning, message dropped
+    end
+
+    CC -> CC : cachedMsg = msg (delta cache updated)
 end
 
-note over PIPE
-  NOT YET IMPLEMENTED.
+== Echo path (loopback via own subscription) ==
 
-  The Connectivity Agent process and its UDS have been
-  removed. The replacement in-process pipeline is:
+AWS -> CC : PUBLISH on the subscribed topic\n→ didMessageReceived(payload, topic)
 
-    Router      → resolve MQTT topic from Type,
-                  build Envelope
-    MessageSpool → durable insert BEFORE publish
-    MqttPublisher → QoS 1 PUBLISH to AWS IoT Core
-                  → PUBACK → MessageSpool::remove()
-    SpoolDrain  → replay spooled messages on reconnect
+CC -> CC : parse Envelope → Header (msgId, sequence, timestamp)
+CC -> CC : match _inflight by (msgId, sequence)\nrecord publish→receive latency
+CC -> CC : msgIdToProtoName() → DescriptorPool lookup\n→ DynamicMessageFactory → MessageToJsonString()\n→ log decoded message
 
-  Until this lands, nothing is sent to the cloud.
-  See docs/SDD/AgentMigrationPlan.md Phases 2-3.
+note over CANBUS, AWS
+  The device both publishes to and subscribes to the same
+  topic. The echo is a round-trip latency measurement for
+  bring-up, not a cloud→device command channel.
 end note
 
+note over CC
+  No persistence today. A message that cannot be published
+  is logged and lost; the spool-on-failure design
+  (AgentMigrationPlan.md section 4.6) is still outstanding.
+
+  Bring-up shortcuts currently in CloudConnectController:
+    * unknown routing action defaults to SendAlways, not Drop
+    * mqttTopic() ignores the per-message Topic and always
+      returns the "clinical" suffix
+    * the delta cache's "received" flag is never set, so the
+      SendDelta suppression branch never fires
+end note
+
+note over CC
+  CloudConnectServer still binds the app Unix domain socket
+  at startup, but no message is written to it — the CAN path
+  now publishes straight to AWS IoT Core.
+end note
+
 @enduml