Prevent passing run verdicts after failed scheduled realizations

Assistant: codex
Assistant-Model: gpt-6-astra
Assistant-Session: 01a0e76f-be98-7ae3-965d-e0b31290a4c4

tests/test_evidence_completeness.py

@@ -155,3 +155,26 @@ def test_truncating_the_manifest_too_cannot_hide_a_missing_baseline_step(baselin
     candidate["verdicts"] = [v for v in candidate["verdicts"] if v["step_id"] != missing]
     candidate["observations"] = [o for o in candidate["observations"] if o["step_id"] != missing]
     assert not classify(baseline, candidate).safe_to_accept
+
+
+@pytest.mark.parametrize('mutation,expected', [(None, Verdict.INCONCLUSIVE), ('M17', Verdict.FAIL)])
+def test_failed_trailing_realization_cannot_leave_a_passing_receipt(tmp_path, mutation, expected):
+    from testdriver import Step, SemanticAction
+    from testdriver.storage import EvidenceStore
+    world, driver, observer, asset, oracle = build(*([mutation] if mutation else []))
+    asset.scenario = replace(
+        asset.scenario,
+        use_case=replace(asset.scenario.use_case, invariants=()),
+        steps=(*asset.scenario.steps, Step('trailing-read', 'bob', SemanticAction(
+            'read_resource', {'resource_id': 'absent'}, frozenset({'api'})))),
+    )
+    store = EvidenceStore(tmp_path)
+    result = Runner(world, driver, observer, oracle).run(asset, evidence_store=store)
+    assert result.verdict is expected
+    assert store.load(result.run_id)['run_verdict'] == expected.value
+    assert any(o.kind == 'realization' and o.step_id == 'trailing-read' and o.data['raised']
+               for o in result.evidence.observations)
+    assert not any(j.step_id == 'trailing-read' for j in result.judgments)
+    if not mutation:
+        assert all(j.verdict is Verdict.PASS for j in result.judgments)
+    assert not classify(store.load(result.run_id), store.load(result.run_id)).safe_to_accept