refactor(evaluator): narrow the agent-eval run seam without an assert

The assert target is not None in AgentEvaluator.run was dead at runtime --
the both/neither cases were already rejected above -- and existed only so the type
checker could narrow target. Reject the "both" case up front and let the
branch chain itself narrow each arm, with the final arm covering "neither".

Adds the missing regression test for the neither-supplied case.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Signed-off-by: Sandy Chapman <schapman@nvidia.com>

Author: SandyChapman

packages/nemo_evaluator_sdk/src/nemo_evaluator_sdk/agent_eval/evaluator.py

@@ -166,17 +166,19 @@ async def run(
         runtime_config = resolved_config.model_copy(update={"run_id": run_id})
         started_at = datetime.now(UTC)
 
-        if (trials is None) == (target is None):
-            raise ValueError("provide exactly one of trials or target")
+        seam_error = "provide exactly one of trials or target"
+        if trials is not None and target is not None:
+            raise ValueError(seam_error)
 
         async with begin_evaluation_session():
-            # Branch on which seam was supplied so the type checker can narrow ``target`` to a
-            # concrete ``AgentEvalTarget`` without a cast.
+            # Branch on which seam was supplied so the type checker narrows each of ``trials`` and
+            # ``target`` to a concrete type without a cast. The final arm is the "neither" case.
             if trials is not None:
                 trial_list = list(trials)
-            else:
-                assert target is not None
+            elif target is not None:
                 trial_list = await self._generate_trials(tasks=task_list, target=target, config=runtime_config)
+            else:
+                raise ValueError(seam_error)
             scores = await self._score_trials(
                 tasks=task_list,
                 trials=trial_list,

packages/nemo_evaluator_sdk/tests/agent_eval/test_evaluator.py

@@ -296,6 +296,11 @@ def test_run_rejects_trials_and_target_together() -> None:
         )
 
 
+def test_run_rejects_neither_trials_nor_target() -> None:
+    with pytest.raises(ValueError, match="provide exactly one"):
+        AgentEvaluator().run_sync(tasks=[_task()])
+
+
 @pytest.mark.asyncio
 async def test_run_writes_nothing_until_persist_is_called(tmp_path: Path) -> None:
     # The point of the change: computing an evaluation and storing one are separate decisions, so a