summaryrefslogtreecommitdiff
path: root/tests
diff options
context:
space:
mode:
authorAnonymous Authors <anonymous@invalid.example>2026-07-25 13:42:41 -0500
committerAnonymous Authors <anonymous@invalid.example>2026-07-25 13:42:41 -0500
commit60e5c8d2ca992e197b640c58a5ba276bb6299bf4 (patch)
treedd714f98a535a0682818704f8222fd79e52ad3cb /tests
parent84fab096b3a2500755fea3f538933dbed0e8c72c (diff)
Remove non-paper judge keyword checksHEADmain
Diffstat (limited to 'tests')
-rw-r--r--tests/test_kernel_models.py13
-rw-r--r--tests/test_models.py24
2 files changed, 23 insertions, 14 deletions
diff --git a/tests/test_kernel_models.py b/tests/test_kernel_models.py
index 0dc653d..8ccfd16 100644
--- a/tests/test_kernel_models.py
+++ b/tests/test_kernel_models.py
@@ -199,3 +199,16 @@ def test_accept_verdict_normalizes_explicit_none_sentinel() -> None:
)
assert verdict.blocking_issues == ""
assert verdict.patch_suggestion == ""
+
+
+def test_reject_verdict_accepts_llm_summary_without_keyword_expansion() -> None:
+ verdict = JudgeVerdict(
+ verdict="reject",
+ step_by_step_check=(
+ "n1–n8 are correct, but the announced slot replacement never "
+ "appears in the candidate problem."
+ ),
+ blocking_issues="the candidate problem is identical to the original",
+ patch_suggestion="apply the declared replacement to the problem",
+ )
+ assert verdict.verdict == "reject"
diff --git a/tests/test_models.py b/tests/test_models.py
index b329ecd..2c08bce 100644
--- a/tests/test_models.py
+++ b/tests/test_models.py
@@ -96,23 +96,20 @@ def test_path_dag_rejects_skipped_dependency() -> None:
)
-def test_judge_verdict_requires_every_dag_node(plan_dict: dict) -> None:
- dag = ProofPlanDAG.from_plan(KernelPlan.model_validate(plan_dict))
+def test_judge_verdict_does_not_require_keyword_coverage() -> None:
verdict = JudgeVerdict(
- verdict="accept",
- step_by_step_check="n1 is valid; n2 is valid",
+ verdict="reject",
+ step_by_step_check=(
+ "The early derivation is correct, but the announced replacement "
+ "never appears in the candidate problem."
+ ),
+ blocking_issues="the candidate is unchanged from the original",
+ patch_suggestion="apply the declared replacement",
)
- assert verdict.validate_coverage(dag) is verdict
+ assert verdict.verdict == "reject"
- with pytest.raises(ValueError, match="does not cover DAG nodes.*n2"):
- JudgeVerdict(
- verdict="accept",
- step_by_step_check="n1 is valid",
- ).validate_coverage(dag)
-
-def test_judge_verdict_normalizes_structured_text_fields(plan_dict: dict) -> None:
- dag = ProofPlanDAG.from_plan(KernelPlan.model_validate(plan_dict))
+def test_judge_verdict_normalizes_structured_text_fields() -> None:
verdict = JudgeVerdict.model_validate(
{
"verdict": "accept",
@@ -128,7 +125,6 @@ def test_judge_verdict_normalizes_structured_text_fields(plan_dict: dict) -> Non
assert '"n1"' in verdict.step_by_step_check
assert verdict.blocking_issues == ""
assert verdict.patch_suggestion == ""
- assert verdict.validate_coverage(dag) is verdict
def test_dag_labels_must_match_prompt_a_plan(plan_dict: dict) -> None: