diff --git a/tests/integration/observability/test_failure_redaction.py b/tests/integration/observability/test_failure_redaction.py
index 2b745d8836d..87660dfbd2a 100644
--- a/tests/integration/observability/test_failure_redaction.py
+++ b/tests/integration/observability/test_failure_redaction.py
@@ -392,7 +392,10 @@ _SDK_CLIENTS: Final = ("openai_sync", "openai_async", "httpx")
 def _chat_call(rig: Rig, client: str, model: str, secret: str, stream: bool) -> str:
     messages: Final = [{"role": "user", "content": secret}]
     if client == "httpx":
-        response: Final = _chat(rig, model, messages)
+        response: Final = rig.proxy.request(
+            "POST", "/v1/chat/completions", {"model": model, "messages": messages, "stream": stream}
+        )
+        response.read()
         assert response.status_code == 400, response.text
         return _call_id(response)
     if client == "openai_sync":
@@ -448,8 +451,11 @@ def _messages_call(rig: Rig, client: str, model: str, secret: str, stream: bool)
     messages: Final = [{"role": "user", "content": secret}]
     if client == "httpx":
         response: Final = rig.proxy.request(
-            "POST", "/v1/messages", {"model": model, "max_tokens": 16, "messages": messages}
+            "POST",
+            "/v1/messages",
+            {"model": model, "max_tokens": 16, "messages": messages, "stream": stream},
         )
+        response.read()
         assert response.status_code == 400, response.text
         return _call_id(response)
     if client == "openai_sync":
diff --git a/tests/integration/observability/test_failure_redaction_chaos.py b/tests/integration/observability/test_failure_redaction_chaos.py
index fdd4056acf9..d49f23ac7c6 100644
--- a/tests/integration/observability/test_failure_redaction_chaos.py
+++ b/tests/integration/observability/test_failure_redaction_chaos.py
@@ -253,18 +253,29 @@ def test_g3_proxy_restart_mid_burst(tmp_path: Path) -> None:
                 second: Final = _fire(rig_two, bodies[BURST // 2 :])
             answered: Final = first + second + tuple(probes)
             assert all(status in (400, 500) for status, _ in answered), answered
-            assert any(status == 400 for status, _ in answered), answered
-            batches: Final = endpoint.drain()
-            empty: Final = tuple(batch for batch in batches if not _json_body_ok(batch.body))
-            assert all(not batch.body.strip() for batch in empty), empty
+            post_restart_ids: Final = frozenset(cid for status, cid in second + tuple(probes) if status == 400 and cid)
+            assert post_restart_ids, answered
+            assert any(marker.encode() in request.body for request in provider.drain())
+            collected: list[Request] = []  # mutable-ok: drain consumes batches, later polls keep earlier ones
+
+            def landed() -> tuple[str, ...]:
+                collected.extend(endpoint.drain())
+                return tuple(
+                    str(event.get("litellm_call_id"))
+                    for batch in collected
+                    if _json_body_ok(batch.body)
+                    for event in json.loads(batch.body)
+                )
+
+            landed_ids: Final = eventually(landed, lambda ids: post_restart_ids <= set(ids), seconds=60)
+            assert all(not batch.body.strip() for batch in collected if not _json_body_ok(batch.body)), collected
+            assert len(landed_ids) == len(set(landed_ids)), ("duplicate events after restart", landed_ids)
             events: Final = tuple(
                 object_value(event)
-                for batch in batches
+                for batch in collected
                 if _json_body_ok(batch.body)
                 for event in json.loads(batch.body)
             )
-            call_ids: Final = tuple(str(event.get("litellm_call_id")) for event in events if event)
-            assert len(call_ids) == len(set(call_ids)), ("duplicate events after restart", call_ids)
             for event in events:
                 assert marker not in json.dumps(event), json.dumps(event)[:400]
 
diff --git a/tests/integration/observability/test_failure_redaction_otel.py b/tests/integration/observability/test_failure_redaction_otel.py
index 6e633fcf3c2..840650a4ede 100644
--- a/tests/integration/observability/test_failure_redaction_otel.py
+++ b/tests/integration/observability/test_failure_redaction_otel.py
@@ -202,10 +202,22 @@ def _span_for(rig: Rig, name: str, model: str) -> dict[str, JsonValue]:
     return spans[0]
 
 
-def _auth_exception_span(rig: Rig) -> dict[str, JsonValue]:
+def _auth_exception_span_ids(rig: Rig) -> frozenset[str]:
+    return frozenset(
+        str(span.get("spanId"))
+        for span in rig.sink.all()
+        if str(span.get("name", "")).startswith("auth") and _exception_events(span)
+    )
+
+
+def _auth_exception_span(rig: Rig, exclude: frozenset[str]) -> dict[str, JsonValue]:
     def found() -> tuple[dict[str, JsonValue], ...]:
         return tuple(
-            span for span in rig.sink.all() if str(span.get("name", "")).startswith("auth") and _exception_events(span)
+            span
+            for span in rig.sink.all()
+            if str(span.get("name", "")).startswith("auth")
+            and _exception_events(span)
+            and str(span.get("spanId")) not in exclude
         )
 
     return eventually(found, lambda values: len(values) >= 1, seconds=260)[0]
@@ -280,6 +292,7 @@ def test_c10_v2_auth_span_honors_header_opt_in(rig_v2_off: Rig) -> None:
         allowed: Final = scenario.model(api_base=rig_v2_off.provider.url + "/v1", api_key="synthetic-provider-key")
         denied: Final = scenario.model(api_base=rig_v2_off.provider.url + "/v1", api_key="synthetic-provider-key")
         key: Final = scenario.key(models=[allowed])
+        before: Final = _auth_exception_span_ids(rig_v2_off)
         response: Final = rig_v2_off.proxy.request(
             "POST",
             "/v1/chat/completions",
@@ -288,7 +301,7 @@ def test_c10_v2_auth_span_honors_header_opt_in(rig_v2_off: Rig) -> None:
             headers={"x-litellm-enable-message-redaction": "true"},
         )
         assert response.status_code in (400, 401, 403, 404), response.text
-        auth: Final = _auth_exception_span(rig_v2_off)
+        auth: Final = _auth_exception_span(rig_v2_off, before)
         assert allowed not in _exception_texts(auth), _exception_texts(auth)[:600]
         assert "redacted-by-litellm" in _exception_texts(auth), _exception_texts(auth)[:600]
 
@@ -300,6 +313,7 @@ def test_c11_v2_auth_span_redacts_under_global_on(rig_v2_on: Rig) -> None:
         allowed: Final = scenario.model(api_base=rig_v2_on.provider.url + "/v1", api_key="synthetic-provider-key")
         denied: Final = scenario.model(api_base=rig_v2_on.provider.url + "/v1", api_key="synthetic-provider-key")
         key: Final = scenario.key(models=[allowed])
+        before: Final = _auth_exception_span_ids(rig_v2_on)
         response: Final = rig_v2_on.proxy.request(
             "POST",
             "/v1/chat/completions",
@@ -308,6 +322,6 @@ def test_c11_v2_auth_span_redacts_under_global_on(rig_v2_on: Rig) -> None:
             headers={"litellm-disable-message-redaction": "true"},
         )
         assert response.status_code in (400, 401, 403, 404), response.text
-        auth: Final = _auth_exception_span(rig_v2_on)
+        auth: Final = _auth_exception_span(rig_v2_on, before)
         assert allowed not in _exception_texts(auth), _exception_texts(auth)[:600]
         assert "redacted-by-litellm" in _exception_texts(auth), _exception_texts(auth)[:600]