[REFACTOR]: Clarify stop-check naming in the trace runner

Rename TraceRun.latest_online_evaluation to latest_stop_check, matching EvaluationPurpose.STOP_CHECK, and describe the final-trace evaluator as producing evidence rather than the safety status.

Author: spencrr

rampart/core/trace.py

@@ -27,7 +27,7 @@
 
 @dataclass(frozen=True, kw_only=True, eq=False)
 class EvaluationRecord:
-    """One online evaluation and the exact context it judged.
+    """One stop-condition check and the exact context it judged.
 
     Args:
         evaluator: Evaluator object that produced the result. Identity is the
@@ -43,7 +43,7 @@ class EvaluationRecord:
 
 @dataclass(kw_only=True)
 class TraceRun:
-    """A completed linear trace and its latest online evaluation.
+    """A completed linear trace and its latest stop-condition check.
 
     ``turns`` is the driver/report view and may carry online evidence.
     ``raw_turns`` is the evaluator view and never carries framework-produced
@@ -55,15 +55,16 @@ class TraceRun:
         manifest: Agent capabilities used to create evaluator contexts.
         turns: Annotated history passed to prompt drivers and results.
         raw_turns: Annotation-free history passed to evaluators.
-        latest_online_evaluation: Most recent stop-condition evaluation.
+        latest_stop_check: Most recent ``stop_when`` evaluation, or None when
+            no stop condition was configured.
     """
 
     trace_end_reason: TraceEndReason
     observability_level: ObservabilityLevel
     manifest: AppManifest | None = None
     turns: list[Turn] = field(default_factory=list[Turn])
     raw_turns: list[Turn] = field(default_factory=list[Turn])
-    latest_online_evaluation: EvaluationRecord | None = None
+    latest_stop_check: EvaluationRecord | None = None
 
 
 def _evaluation_context(
@@ -166,7 +167,7 @@ async def run_trace_async(
             manifest=manifest,
         )
         evaluation = await stop_when.evaluate_async(context=context)
-        run.latest_online_evaluation = EvaluationRecord(
+        run.latest_stop_check = EvaluationRecord(
             evaluator=stop_when,
             context=context,
             result=evaluation,
@@ -190,10 +191,10 @@ async def evaluate_final_trace_async(
     evaluator: Evaluator,
     run: TraceRun,
 ) -> EvalResult | None:
-    """Evaluate the final raw trace, reusing an identical online judgment.
+    """Evaluate the final raw trace, reusing an identical stop-condition check.
 
     Args:
-        evaluator: Evaluator responsible for the final verdict.
+        evaluator: Evaluator applied to the final trace.
         run: Completed trace from :func:`run_trace_async`.
 
     Returns:
@@ -207,7 +208,7 @@ async def evaluate_final_trace_async(
     if not run.raw_turns:
         return None
 
-    record = run.latest_online_evaluation
+    record = run.latest_stop_check
     if (
         record is not None
         and record.evaluator is evaluator

tests/unit/core/test_trace.py

@@ -61,7 +61,7 @@ async def test_driver_exhaustion_returns_raw_turns_async(self) -> None:
         assert run.trace_end_reason is TraceEndReason.DRIVER_EXHAUSTED
         assert [turn.response.text for turn in run.turns] == ["r1", "r2"]
         assert run.turns == run.raw_turns
-        assert run.latest_online_evaluation is None
+        assert run.latest_stop_check is None
 
     async def test_turn_budget_is_a_normal_termination_async(self) -> None:
         run = await run_trace_async(
@@ -183,8 +183,8 @@ async def test_changed_observability_requires_new_judgment_async(self) -> None:
 
         result = await evaluate_final_trace_async(evaluator=evaluator, run=updated_run)
 
-        assert run.latest_online_evaluation is not None
-        assert run.latest_online_evaluation.result.outcome is EvalOutcome.NOT_DETECTED
+        assert run.latest_stop_check is not None
+        assert run.latest_stop_check.result.outcome is EvalOutcome.NOT_DETECTED
         assert result is not None
         assert result.outcome is EvalOutcome.UNDETERMINED
 
@@ -278,7 +278,7 @@ async def test_empty_trace_skips_evaluator_async(self) -> None:
             (EvalOutcome.NOT_DETECTED,),
         ],
     )
-    async def test_reuses_identical_latest_online_evaluation_async(
+    async def test_reuses_identical_latest_stop_check_async(
         self,
         outcomes: tuple[EvalOutcome, ...],
     ) -> None:
@@ -291,14 +291,14 @@ async def test_reuses_identical_latest_online_evaluation_async(
             stop_when=evaluator,
             manifest=AppManifest(name="agent"),
         )
-        online_result = run.latest_online_evaluation
-        assert online_result is not None
+        stop_check = run.latest_stop_check
+        assert stop_check is not None
 
         result = await evaluate_final_trace_async(evaluator=evaluator, run=run)
 
-        assert result == online_result.result
-        assert result is not online_result.result
-        assert result.evidence is not online_result.result.evidence
+        assert result == stop_check.result
+        assert result is not stop_check.result
+        assert result.evidence is not stop_check.result.evidence
         assert evaluator.evaluate_async.await_count == 1
 
     async def test_non_firing_stop_reuses_final_prefix_without_extra_call_async(