test(integration): address round-7 review — keep RAG assertion live, fix races

- test_no_results_for_unrelated_query: replace pytest.skip with `unrelated =
  ... or 0.0` and fall through. The physics query almost always returns nothing
  on this corpus, so the skip meant the comparison (and the manual-is-indexed
  check) never ran. Treating no-results as score 0.0 keeps the test live and
  vacuously satisfies `0.0 <= relevant`.
- test_sampling: the three limit/threshold/max-tokens tests now gate on a
  representative created note being searchable (search_term + note_id) instead
  of a bare idle signal that can fire before the new notes are enqueued.
- _get_with_retry: only sleep between attempts, not before giving up.
- _search_helpers: log the id/doc_type schema-drift mismatch at WARNING (CI runs
  --log-cli-level=WARN) so it surfaces instead of hiding behind a timeout.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

tests/integration/_search_helpers.py

@@ -49,9 +49,10 @@ async def document_is_searchable(
             if r.get("id") == note_id:
                 if r.get("doc_type") == "note":
                     return True
-                # id matched but not a note — surface possible schema drift
+                # id matched but not a note — surface possible schema drift at
-                # rather than silently timing out.
+                # WARNING (CI runs --log-cli-level=WARN) instead of letting the
-                logger.debug(
+                # caller time out with a generic message.
+                logger.warning(
                     "search hit id=%s has doc_type=%s (expected note)",
                     note_id,
                     r.get("doc_type"),

tests/integration/test_astrolabe_session_jwt_search.py

@@ -56,7 +56,8 @@ async def _get_with_retry(
             logger.warning(
                 "GET %s failed (attempt %s/%s): %s", url, attempt, max_attempts, e
             )
-            await anyio.sleep(2)
+            if attempt < max_attempts:
+                await anyio.sleep(2)  # no point sleeping before we give up
     assert last_exc is not None  # loop ran at least once, so this is set
     raise last_exc
 

tests/integration/test_rag.py

@@ -432,13 +432,16 @@ async def test_no_results_for_unrelated_query(nc_mcp_client, indexed_manual_pdf)
     query happened to retrieve any chunk at all). Comparing against a relevant
     query on the same corpus is self-calibrating and stable.
     """
-    unrelated = await _top_score(
+    # No results for the nonsense query is the ideal outcome — treat as score
-        nc_mcp_client, "quantum entanglement hadron collider particle physics"
+    # 0.0 and fall through, so the comparison (and the manual-is-indexed check
+    # below) still runs instead of the test silently skipping every time the
+    # physics query finds nothing.
+    unrelated = (
+        await _top_score(
+            nc_mcp_client, "quantum entanglement hadron collider particle physics"
+        )
+        or 0.0
     )
-    if unrelated is None:
-        # No results for the nonsense query is the ideal outcome; skip (rather
-        # than a silent pass) so the test report shows the path was taken.
-        pytest.skip("No results for nonsense physics query — the ideal outcome")
 
     relevant = await _top_score(
         nc_mcp_client, "how do I enable two-factor authentication"

tests/integration/test_sampling.py

@@ -274,8 +274,12 @@ async def test_semantic_search_answer_with_limit(nc_mcp_client, temporary_note_f
         category="Development",
     )
 
-    # Wait for vector indexing to complete
+    # Wait until the batch is indexed — gate on the last note being searchable
-    await wait_for_vector_sync(nc_mcp_client)
+    # rather than a bare idle signal, which can fire before the new notes are
+    # even enqueued.
+    await wait_for_vector_sync(
+        nc_mcp_client, search_term="async context managers", note_id=_note3["id"]
+    )
 
     call_result = await nc_mcp_client.call_tool(
         "nc_semantic_search_answer",
@@ -315,8 +319,10 @@ async def test_semantic_search_answer_score_threshold(
         category="Test",
     )
 
-    # Wait for vector indexing to complete
+    # Gate on the new note being searchable (not a bare idle signal).
-    await wait_for_vector_sync(nc_mcp_client)
+    await wait_for_vector_sync(
+        nc_mcp_client, search_term="widget manufacturing", note_id=_note["id"]
+    )
 
     # Query with exact match
     call_result = await nc_mcp_client.call_tool(
@@ -362,8 +368,10 @@ async def test_semantic_search_answer_max_tokens(nc_mcp_client, temporary_note_f
         category="Test",
     )
 
-    # Wait for vector indexing to complete
+    # Gate on the new note being searchable (not a bare idle signal).
-    await wait_for_vector_sync(nc_mcp_client)
+    await wait_for_vector_sync(
+        nc_mcp_client, search_term="Long Document content", note_id=_note["id"]
+    )
 
     call_result = await nc_mcp_client.call_tool(
         "nc_semantic_search_answer",