diff --git a/uts/docs/integration-testing.md b/uts/docs/integration-testing.md
index 352d66136..933bb260b 100644
--- a/uts/docs/integration-testing.md
+++ b/uts/docs/integration-testing.md
@@ -302,6 +302,7 @@ The following integration test specs are annotated with `## Protocol Variants`:
 **LiveObjects:**
 - `objects/integration/objects_lifecycle_test.md`
 - `objects/integration/objects_sync_test.md`
+- `objects/integration/objects_gc_test.md`
 
 ## Writing Proxy Tests
 
diff --git a/uts/docs/writing-test-specs.md b/uts/docs/writing-test-specs.md
index aa403e6b2..0e14160d2 100644
--- a/uts/docs/writing-test-specs.md
+++ b/uts/docs/writing-test-specs.md
@@ -35,7 +35,7 @@ Every test in the UTS suite has a unique identifier. The ID appears explicitly i
 
 | Field | Description |
 |-------|-------------|
-| `category` | One of: `rest/unit`, `rest/integration`, `rest/proxy`, `realtime/unit`, `realtime/integration`, `realtime/proxy` |
+| `category` | One of: `rest/unit`, `rest/integration`, `rest/proxy`, `realtime/unit`, `realtime/integration`, `realtime/proxy`, `objects/unit`, `objects/integration`, `objects/proxy` |
 | `spec-point` | The primary spec point being tested (e.g. `RSC15l2`, `RTN14a`) |
 | `descriptive-name` | 2–4 hyphenated words describing the specific behaviour (e.g. `timeout-fallback`, `cloudfront-header`) |
 | `n` | 0-based index disambiguating multiple tests for the same spec point within the same file |
@@ -343,7 +343,7 @@ Tests that [behaviour] when the proxy injects [fault].
 
 ```pseudo
 session = create_proxy_session(
-  target: TargetConfig(realtimeHost: "sandbox.realtime.ably-nonprod.net", restHost: "sandbox.realtime.ably-nonprod.net"),
+  endpoint: "nonprod:sandbox",
   rules: [{
     "match": { ... },
     "action": { ... },
@@ -658,6 +658,18 @@ In **integration and proxy** specs these timeouts are **wall-clock time** (the t
 real server — see the *Timeout Strategy* section in `docs/integration-testing.md`); in unit specs they may be
 realised with fake/virtual timers.
 
+### Closing Clients (`CLOSE_CLIENT`)
+
+Realtime clients hold internal timers (retry, heartbeat, suspend) that keep the test process alive after the assertions complete. A test that creates a Realtime client should end with `CLOSE_CLIENT(client)`, one call per client it created:
+
+```pseudo
+ASSERT client.connection.state == ConnectionState.connected
+
+CLOSE_CLIENT(client)
+```
+
+`CLOSE_CLIENT` is a cleanup step, not an assertion: it closes the client so its timers stop, and doesn't wait for or assert the CLOSED state. Derived tests should close the client even when the test fails (see *Cleanup with afterEach* in `writing-derived-tests.md`).
+
 ## Timer Mocking
 
 Tests verifying timeout behavior should use timer mocking where practical to avoid slow tests.
diff --git a/uts/realtime/integration/proxy/connection_resume.md b/uts/realtime/integration/proxy/connection_resume.md
index 90fddc848..c31f32e44 100644
--- a/uts/realtime/integration/proxy/connection_resume.md
+++ b/uts/realtime/integration/proxy/connection_resume.md
@@ -779,6 +779,10 @@ As of specification version 6.1.0 this replaces RTN15g (deleted). The client no
 3. Refuse the second ws_connect (fires once). This keeps the client in DISCONNECTED while the TTL clock runs out, so when the TTL expires the client transitions to SUSPENDED. The third ws_connect (when the `suspendedRetryTimeout` fires) should be a fresh connection.
 
 ```pseudo
+# The replace action sends its message verbatim (uts-proxy has no placeholder that
+# keeps the server's real value), so the client receives the fixed fake
+# connectionKey below. The resume after TTL expiry is expected to fail
+# server-side either way; the test checks only that the client attempts it (RTN14h).
 session = create_proxy_session(
   endpoint: "nonprod:sandbox",
   rules: [
@@ -789,9 +793,9 @@ session = create_proxy_session(
         "message": {
           "action": 4,
           "connectionId": "proxy-ttl-test-id",
-          "connectionKey": "__PASSTHROUGH__",
+          "connectionKey": "proxy-ttl-test-key",
           "connectionDetails": {
-            "connectionKey": "__PASSTHROUGH__",
+            "connectionKey": "proxy-ttl-test-key",
             "clientId": null,
             "maxMessageSize": 65536,
             "maxInboundRate": 250,
diff --git a/uts/realtime/integration/proxy/presence_reentry.md b/uts/realtime/integration/proxy/presence_reentry.md
index 86c411a18..ae1a849dd 100644
--- a/uts/realtime/integration/proxy/presence_reentry.md
+++ b/uts/realtime/integration/proxy/presence_reentry.md
@@ -71,7 +71,10 @@ key_name = key_parts[0]
 key_secret = key_parts[1]
 
 # Create proxy session with clean passthrough (no fault rules)
-session = create_proxy_session(rules: [])
+session = create_proxy_session(
+  endpoint: "nonprod:sandbox",
+  rules: []
+)
 
 # client: the presence member, connects through the proxy so we can inject ATTACHED
 # Needs a clientId for presence — use authCallback with JWT that includes clientId
@@ -191,6 +194,7 @@ key_secret = key_parts[1]
 # 1. Close the WebSocket 3s after connect (giving time to attach + enter presence)
 # 2. Replace the 2nd ATTACHED on the channel with a non-resumed one
 session = create_proxy_session(
+  endpoint: "nonprod:sandbox",
   rules: [
     {
       match: { type: "delay_after_ws_connect", delayMs: 3000 },