fix(tests): migrate eval tests to Langfuse V3 API

lf.trace() and lf.score(trace_id=...) are V2 API removed in V3.

V3 pattern:
  lf.start_as_current_observation(name=...) as context manager → obs
  obs.score(name=..., value=...)
  contextlib.nullcontext() when lf is None so structure stays the same

Updated tests 2.1–2.7 in test_agent_runner_v2.py accordingly.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

tests/test_agent_runner_v2.py

@@ -340,28 +340,35 @@ async def test_2_8_items_created_count():
 
 
 # ── Eval: 2.1–2.7 (real LLM + Langfuse scoring) ──────────────────────────
+#
+# Langfuse V3 pattern:
+#   lf.start_as_current_observation(name=...) as context manager → obs object
+#   obs.score(name=..., value=...)  (not lf.score(trace_id=...))
+#   contextlib.nullcontext() when lf is None → obs is None, no-op
+# ─────────────────────────────────────────────────────────────────────────
 
 
 @pytest.mark.asyncio
 @pytest.mark.eval
 async def test_2_1_email_to_task():
     """2.1 Action email → LLM calls create_task. Score: runner.email_to_task."""
+    from contextlib import nullcontext
     lf = get_langfuse()
-    trace = lf.trace(
-        name="eval-runner-2.1-email-to-task",
-        metadata={"step": "2"},
-    ) if lf else None
 
     config = _make_config()
     run_log = _make_run_log(config.id)
     mgr = _make_manager()
-
     executor, calls = _make_executor(
         file_path="/emails/ProjectAlpha_action.html",
         file_content=_ACTION_EMAIL,
         projects=[_PROJECT_ALPHA, _PROJECT_BETA],
     )
 
+    obs_ctx = lf.start_as_current_observation(
+        name="eval-runner-2.1-email-to-task", metadata={"step": "2"}
+    ) if lf else nullcontext()
+
+    with obs_ctx as obs:
         with patch("app.core.agent_runner._make_agent_executor", return_value=executor), \
              patch("app.core.agent_runner._finalize_run", new_callable=AsyncMock) as mock_fin:
             await run_local_agent(_USER_ID, config, run_log, mgr)
@@ -370,13 +377,14 @@ async def test_2_1_email_to_task():
         task_creates = [c for c in calls if c["action"] == "insert" and c["table"] == "tasks"]
         score = 1.0 if len(task_creates) >= 1 else 0.0
 
-    if lf and trace:
+        if obs is not None:
-        lf.score(
+            obs.score(
-            trace_id=trace.id,
                 name="runner.email_to_task",
                 value=score,
                 comment=f"task_creates={len(task_creates)} items_created={kwargs.get('items_created')}",
             )
+
+    if lf:
         lf.flush()
 
     assert score == 1.0, f"Expected at least 1 task created, got {len(task_creates)}"
@@ -386,19 +394,23 @@ async def test_2_1_email_to_task():
 @pytest.mark.eval
 async def test_2_2_email_to_note():
     """2.2 Informational email → LLM calls create_note. Score: runner.email_to_note."""
+    from contextlib import nullcontext
     lf = get_langfuse()
-    trace = lf.trace(name="eval-runner-2.2-email-to-note", metadata={"step": "2"}) if lf else None
 
     config = _make_config()
     run_log = _make_run_log(config.id)
     mgr = _make_manager()
-
     executor, calls = _make_executor(
         file_path="/emails/ProjectAlpha_info.html",
         file_content=_INFO_EMAIL,
         projects=[_PROJECT_ALPHA, _PROJECT_BETA],
     )
 
+    obs_ctx = lf.start_as_current_observation(
+        name="eval-runner-2.2-email-to-note", metadata={"step": "2"}
+    ) if lf else nullcontext()
+
+    with obs_ctx as obs:
         with patch("app.core.agent_runner._make_agent_executor", return_value=executor), \
              patch("app.core.agent_runner._finalize_run", new_callable=AsyncMock):
             await run_local_agent(_USER_ID, config, run_log, mgr)
@@ -406,9 +418,11 @@ async def test_2_2_email_to_note():
         note_creates = [c for c in calls if c["action"] == "insert" and c["table"] == "notes"]
         score = 1.0 if len(note_creates) >= 1 else 0.0
 
-    if lf and trace:
+        if obs is not None:
-        lf.score(trace_id=trace.id, name="runner.email_to_note", value=score,
+            obs.score(name="runner.email_to_note", value=score,
                       comment=f"note_creates={len(note_creates)}")
+
+    if lf:
         lf.flush()
 
     assert score == 1.0, f"Expected at least 1 note created, got {len(note_creates)}"
@@ -418,19 +432,23 @@ async def test_2_2_email_to_note():
 @pytest.mark.eval
 async def test_2_3_email_to_timeline():
     """2.3 Email with event date → LLM calls create_timeline. Score: runner.email_to_timeline."""
+    from contextlib import nullcontext
     lf = get_langfuse()
-    trace = lf.trace(name="eval-runner-2.3-email-to-timeline", metadata={"step": "2"}) if lf else None
 
     config = _make_config()
     run_log = _make_run_log(config.id)
     mgr = _make_manager()
-
     executor, calls = _make_executor(
         file_path="/emails/ProjectAlpha_kickoff.html",
         file_content=_DATE_EMAIL,
         projects=[_PROJECT_ALPHA, _PROJECT_BETA],
     )
 
+    obs_ctx = lf.start_as_current_observation(
+        name="eval-runner-2.3-email-to-timeline", metadata={"step": "2"}
+    ) if lf else nullcontext()
+
+    with obs_ctx as obs:
         with patch("app.core.agent_runner._make_agent_executor", return_value=executor), \
              patch("app.core.agent_runner._finalize_run", new_callable=AsyncMock):
             await run_local_agent(_USER_ID, config, run_log, mgr)
@@ -438,9 +456,11 @@ async def test_2_3_email_to_timeline():
         tl_creates = [c for c in calls if c["action"] == "insert" and c["table"] == "timelines"]
         score = 1.0 if len(tl_creates) >= 1 else 0.0
 
-    if lf and trace:
+        if obs is not None:
-        lf.score(trace_id=trace.id, name="runner.email_to_timeline", value=score,
+            obs.score(name="runner.email_to_timeline", value=score,
                       comment=f"timeline_creates={len(tl_creates)}")
+
+    if lf:
         lf.flush()
 
     assert score == 1.0, f"Expected at least 1 timeline created, got {len(tl_creates)}"
@@ -450,33 +470,37 @@ async def test_2_3_email_to_timeline():
 @pytest.mark.eval
 async def test_2_4_project_matching_filename():
     """2.4 Filename contains 'ProjectAlpha' → LLM assigns to proj-alpha. Score: runner.project_filename."""
+    from contextlib import nullcontext
     lf = get_langfuse()
-    trace = lf.trace(name="eval-runner-2.4-project-filename", metadata={"step": "2"}) if lf else None
 
     config = _make_config()
     run_log = _make_run_log(config.id)
     mgr = _make_manager()
-
     executor, calls = _make_executor(
         file_path="/emails/ProjectAlpha_report.html",
         file_content=_ACTION_EMAIL,
         projects=[_PROJECT_ALPHA, _PROJECT_BETA],
     )
 
+    obs_ctx = lf.start_as_current_observation(
+        name="eval-runner-2.4-project-filename", metadata={"step": "2"}
+    ) if lf else nullcontext()
+
+    with obs_ctx as obs:
         with patch("app.core.agent_runner._make_agent_executor", return_value=executor), \
              patch("app.core.agent_runner._finalize_run", new_callable=AsyncMock):
             await run_local_agent(_USER_ID, config, run_log, mgr)
 
-    # Check that project_id = proj-alpha was used in any insert
         inserts = [c for c in calls if c["action"] == "insert"]
         correct_project = any(
-        c.get("data", {}).get("projectId") == "proj-alpha"
+            c.get("data", {}).get("projectId") == "proj-alpha" for c in inserts
-        for c in inserts
         )
         score = 1.0 if correct_project else 0.0
 
-    if lf and trace:
+        if obs is not None:
-        lf.score(trace_id=trace.id, name="runner.project_filename", value=score)
+            obs.score(name="runner.project_filename", value=score)
+
+    if lf:
         lf.flush()
 
     assert score == 1.0, "Expected inserts to use proj-alpha based on filename"
@@ -486,32 +510,37 @@ async def test_2_4_project_matching_filename():
 @pytest.mark.eval
 async def test_2_5_project_matching_content():
     """2.5 Email body mentions 'Project Alpha' → correct project assigned. Score: runner.project_content."""
+    from contextlib import nullcontext
     lf = get_langfuse()
-    trace = lf.trace(name="eval-runner-2.5-project-content", metadata={"step": "2"}) if lf else None
 
     config = _make_config()
     run_log = _make_run_log(config.id)
     mgr = _make_manager()
-
     executor, calls = _make_executor(
         file_path="/emails/email_001.html",  # generic filename, no project hint
         file_content=_ACTION_EMAIL,          # body mentions "Project Alpha"
         projects=[_PROJECT_ALPHA, _PROJECT_BETA],
     )
 
+    obs_ctx = lf.start_as_current_observation(
+        name="eval-runner-2.5-project-content", metadata={"step": "2"}
+    ) if lf else nullcontext()
+
+    with obs_ctx as obs:
         with patch("app.core.agent_runner._make_agent_executor", return_value=executor), \
              patch("app.core.agent_runner._finalize_run", new_callable=AsyncMock):
             await run_local_agent(_USER_ID, config, run_log, mgr)
 
         inserts = [c for c in calls if c["action"] == "insert"]
         correct_project = any(
-        c.get("data", {}).get("projectId") == "proj-alpha"
+            c.get("data", {}).get("projectId") == "proj-alpha" for c in inserts
-        for c in inserts
         )
         score = 1.0 if correct_project else 0.0
 
-    if lf and trace:
+        if obs is not None:
-        lf.score(trace_id=trace.id, name="runner.project_content", value=score)
+            obs.score(name="runner.project_content", value=score)
+
+    if lf:
         lf.flush()
 
     assert score == 1.0, "Expected inserts to use proj-alpha based on email body content"
@@ -521,30 +550,35 @@ async def test_2_5_project_matching_content():
 @pytest.mark.eval
 async def test_2_6_no_project_match_global_rule():
     """2.6 Newsletter email + global rule 'no project = no entities' → no creates. Score: runner.no_project."""
+    from contextlib import nullcontext
     lf = get_langfuse()
-    trace = lf.trace(name="eval-runner-2.6-no-project", metadata={"step": "2"}) if lf else None
 
     config = _make_config()
     run_log = _make_run_log(config.id)
     mgr = _make_manager()
-
     executor, calls = _make_executor(
         file_path="/emails/newsletter.html",
         file_content=_NO_PROJECT_EMAIL,
         projects=[_PROJECT_ALPHA, _PROJECT_BETA],
     )
 
+    obs_ctx = lf.start_as_current_observation(
+        name="eval-runner-2.6-no-project", metadata={"step": "2"}
+    ) if lf else nullcontext()
+
+    with obs_ctx as obs:
         with patch("app.core.agent_runner._make_agent_executor", return_value=executor), \
-         patch("app.core.agent_runner._finalize_run", new_callable=AsyncMock) as mock_fin:
+             patch("app.core.agent_runner._finalize_run", new_callable=AsyncMock):
             await run_local_agent(_USER_ID, config, run_log, mgr)
 
-    _, kwargs = mock_fin.call_args
         inserts = [c for c in calls if c["action"] == "insert"]
         score = 1.0 if len(inserts) == 0 else 0.0
 
-    if lf and trace:
+        if obs is not None:
-        lf.score(trace_id=trace.id, name="runner.no_project", value=score,
+            obs.score(name="runner.no_project", value=score,
                       comment=f"inserts={len(inserts)}")
+
+    if lf:
         lf.flush()
 
     assert score == 1.0, f"Expected 0 inserts for unmatched newsletter, got {len(inserts)}"
@@ -554,13 +588,12 @@ async def test_2_6_no_project_match_global_rule():
 @pytest.mark.eval
 async def test_2_7_deduplication():
     """2.7 Existing task with same title → LLM calls update_task, not create_task. Score: runner.dedup."""
+    from contextlib import nullcontext
     lf = get_langfuse()
-    trace = lf.trace(name="eval-runner-2.7-dedup", metadata={"step": "2"}) if lf else None
 
     config = _make_config()
     run_log = _make_run_log(config.id)
     mgr = _make_manager()
-
     executor, calls = _make_executor(
         file_path="/emails/ProjectAlpha_followup.html",
         file_content=_ACTION_EMAIL,       # "Fix the login bug" — already exists
@@ -568,18 +601,24 @@ async def test_2_7_deduplication():
         existing_tasks=[_EXISTING_TASK],  # task already exists
     )
 
+    obs_ctx = lf.start_as_current_observation(
+        name="eval-runner-2.7-dedup", metadata={"step": "2"}
+    ) if lf else nullcontext()
+
+    with obs_ctx as obs:
         with patch("app.core.agent_runner._make_agent_executor", return_value=executor), \
              patch("app.core.agent_runner._finalize_run", new_callable=AsyncMock):
             await run_local_agent(_USER_ID, config, run_log, mgr)
 
         task_creates = [c for c in calls if c["action"] == "insert" and c["table"] == "tasks"]
         task_updates = [c for c in calls if c["action"] == "update" and c.get("table") == "tasks"]
-    # Prefer update over create
         score = 1.0 if len(task_creates) == 0 or len(task_updates) >= 1 else 0.0
 
-    if lf and trace:
+        if obs is not None:
-        lf.score(trace_id=trace.id, name="runner.dedup", value=score,
+            obs.score(name="runner.dedup", value=score,
                       comment=f"creates={len(task_creates)} updates={len(task_updates)}")
+
+    if lf:
         lf.flush()
 
     assert score == 1.0, (