diff options
Diffstat (limited to 'tests')
| -rw-r--r-- | tests/test_models.py | 83 | ||||
| -rw-r--r-- | tests/test_pipeline.py | 8 |
2 files changed, 88 insertions, 3 deletions
diff --git a/tests/test_models.py b/tests/test_models.py index 71dc854..d19b536 100644 --- a/tests/test_models.py +++ b/tests/test_models.py @@ -3,7 +3,12 @@ from __future__ import annotations import pytest from pydantic import ValidationError -from gap_pipeline.models import JudgeVerdict, KernelCandidate, KernelPlan +from gap_pipeline.models import ( + JudgeVerdict, + KernelCandidate, + KernelPlan, + ProofPlanDAG, +) def test_valid_plan_and_candidate(plan_dict: dict, candidate_dict: dict) -> None: @@ -38,3 +43,79 @@ def test_accept_verdict_cannot_report_blocking_issue() -> None: step_by_step_check="checked", blocking_issues="an error", ) + + +def test_path_dag_is_derived_from_ordered_plan(plan_dict: dict) -> None: + dag = ProofPlanDAG.from_plan(KernelPlan.model_validate(plan_dict)) + assert [node.node_id for node in dag.nodes] == ["n1", "n2"] + assert dag.nodes[0].dependencies == [] + assert dag.nodes[1].dependencies == ["n1"] + assert dag.terminal_node_id == "n2" + assert dag.method_labels_payload()[0].startswith("n1: ") + + +def test_dag_rejects_cycle() -> None: + with pytest.raises(ValidationError, match="cycle"): + ProofPlanDAG.model_validate( + { + "nodes": [ + { + "node_id": "n1", + "method_label": "first", + "dependencies": ["n2"], + }, + { + "node_id": "n2", + "method_label": "second", + "dependencies": ["n1"], + }, + ], + "terminal_node_id": "n2", + } + ) + + +def test_path_dag_rejects_skipped_dependency() -> None: + with pytest.raises(ValidationError, match="lead to the terminal"): + ProofPlanDAG.model_validate( + { + "nodes": [ + { + "node_id": "n1", + "method_label": "first", + "dependencies": [], + }, + { + "node_id": "n2", + "method_label": "second", + "dependencies": [], + }, + ], + "terminal_node_id": "n2", + } + ) + + +def test_judge_verdict_requires_every_dag_node(plan_dict: dict) -> None: + dag = ProofPlanDAG.from_plan(KernelPlan.model_validate(plan_dict)) + verdict = JudgeVerdict( + verdict="accept", + step_by_step_check="n1 is valid; n2 is valid", + ) + assert verdict.validate_coverage(dag) is verdict + + 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_dag_labels_must_match_prompt_a_plan(plan_dict: dict) -> None: + plan = KernelPlan.model_validate(plan_dict) + dag = ProofPlanDAG.from_plan(plan) + changed_plan = plan.model_copy( + update={"core_steps": ["a different method", plan.core_steps[1]]} + ) + with pytest.raises(ValueError, match="must equal Prompt-A core_steps"): + dag.validate_plan(changed_plan) diff --git a/tests/test_pipeline.py b/tests/test_pipeline.py index 864e967..04c3ea9 100644 --- a/tests/test_pipeline.py +++ b/tests/test_pipeline.py @@ -13,7 +13,7 @@ from conftest import changed_candidate def accept_review() -> dict[str, str]: return { "verdict": "accept", - "step_by_step_check": "every method label is instantiated", + "step_by_step_check": "n1 passes; n2 passes", "blocking_issues": "", "patch_suggestion": "", } @@ -22,7 +22,9 @@ def accept_review() -> dict[str, str]: def reject_review(iteration: int) -> dict[str, str]: return { "verdict": "reject", - "step_by_step_check": f"terminal step fails at iteration {iteration}", + "step_by_step_check": ( + f"n1 passes; n2 terminal step fails at iteration {iteration}" + ), "blocking_issues": "terminal algebra is incorrect", "patch_suggestion": "correct the terminal algebra", } @@ -76,6 +78,8 @@ def test_two_consecutive_unanimous_passes_accept( final_path = tmp_path / "items" / item.item_id / "final.json" final = json.loads(final_path.read_text()) assert final["accepted_candidate_sha256"] + assert final["proof_dag"]["structure"] == "path" + assert final["proof_dag"]["terminal_node_id"] == "n2" assert len(list((final_path.parent / "calls").glob("*.json"))) == 12 |