fix(eval): sync trajectory fixtures/schemas and resolve CI failures

Co-authored-by: morluto <morluto@users.noreply.github.com>

Author: cursoragent

benchmarks/validation/test_verifier_security_regressions.py

@@ -356,6 +356,69 @@ def _provider_report_case(tmp_path: Path, task_name: str, report: dict) -> tuple
     return task, app, logs
 
 
+def test_cddlib_rejects_fabricated_nonempty_cases(tmp_path: Path) -> None:
+    expected = json.loads(
+        (
+            DATASETS / "provider-feasibility-v1" / "cddlib" / "tests" / "expected.json"
+        ).read_text()
+    )
+    frozen = expected["reproduction"]
+    fake_digest = "sha256:" + ("0" * 64)
+    report = {
+        "contract": expected["contract"],
+        "status": "COMPLETED",
+        "conclusion": expected["report_conclusion"],
+        "assurance": expected["report_assurance"],
+        "provider": {
+            "runtime": {"python": "3.12.0"},
+            "versions": frozen["mathematical_output"]["versions"],
+        },
+        "reproduction": {
+            "scope": frozen["scope"],
+            "provider_output_sha256": fake_digest,
+            "cases": [{"case_id": "fabricated"}],
+        },
+        "limitations": ["fabricated"],
+        "extra": True,
+    }
+    task, app, logs = _provider_report_case(tmp_path, "cddlib", report)
+    rejected = support._run_verifier(task, app, logs)
+    assert rejected["reward"] == 0.0
+
+
+def test_regina_rejects_fabricated_nonempty_cases(tmp_path: Path) -> None:
+    expected = json.loads(
+        (
+            DATASETS / "provider-feasibility-v1" / "regina" / "tests" / "expected.json"
+        ).read_text()
+    )
+    frozen = expected["reproduction"]
+    fake_digest = "sha256:" + ("0" * 64)
+    report = {
+        "contract": expected["contract"],
+        "status": "COMPLETED",
+        "conclusion": expected["report_conclusion"],
+        "assurance": expected["report_assurance"],
+        "provider": {
+            "runtime": {"python": "3.12.0"},
+            "distribution_version": frozen["expected_provider_output"][
+                "distribution_version"
+            ],
+        },
+        "reproduction": {
+            "scope": frozen["scope"],
+            "provider_output_sha256": fake_digest,
+            "cases": [{"case_id": "fabricated"}],
+            "normal_surfaces": {"surface_count": 0, "surfaces": []},
+        },
+        "limitations": ["fabricated"],
+        "extra": True,
+    }
+    task, app, logs = _provider_report_case(tmp_path, "regina", report)
+    rejected = support._run_verifier(task, app, logs)
+    assert rejected["reward"] == 0.0
+
+
 def test_cgal_rejects_fabricated_reproduction_digests(tmp_path: Path) -> None:
     expected = json.loads(
         (
@@ -388,6 +451,39 @@ def test_cgal_rejects_fabricated_reproduction_digests(tmp_path: Path) -> None:
     assert rejected["reward"] == 0.0
 
 
+def test_cgal_rejects_unbound_source_and_adapter_identity(tmp_path: Path) -> None:
+    expected = json.loads(
+        (
+            DATASETS / "provider-feasibility-v1" / "cgal" / "tests" / "expected.json"
+        ).read_text()
+    )
+    fake_digest = "sha256:" + ("0" * 64)
+    report = {
+        "contract": expected["contract"],
+        "status": "COMPLETED",
+        "conclusion": expected["report_conclusion"],
+        "assurance": expected["report_assurance"],
+        "provider": {
+            "executable": "/usr/local/bin/cgal-spike",
+            "executable_sha256": fake_digest,
+            "adapter_source_sha256": fake_digest,
+            "source": {"archive_sha256": fake_digest},
+        },
+        "reproductions": {
+            name: {
+                **case,
+                "observed_output_sha256": case["expected_output_sha256"],
+            }
+            for name, case in expected["reproductions"].items()
+        },
+        "limitations": ["fabricated"],
+        "extra": True,
+    }
+    task, app, logs = _provider_report_case(tmp_path, "cgal", report)
+    rejected = support._run_verifier(task, app, logs)
+    assert rejected["reward"] == 0.0
+
+
 def test_gudhi_rejects_fabricated_persistence_shape(tmp_path: Path) -> None:
     expected = json.loads(
         (
@@ -416,6 +512,34 @@ def test_gudhi_rejects_fabricated_persistence_shape(tmp_path: Path) -> None:
     assert rejected["reward"] == 0.0
 
 
+def test_gudhi_rejects_missing_mathematical_output_digest(tmp_path: Path) -> None:
+    expected = json.loads(
+        (
+            DATASETS / "provider-feasibility-v1" / "gudhi" / "tests" / "expected.json"
+        ).read_text()
+    )
+    fake_digest = "sha256:" + ("0" * 64)
+    report = {
+        "contract": expected["contract"],
+        "status": "COMPLETED",
+        "conclusion": expected["report_conclusion"],
+        "assurance": expected["report_assurance"],
+        "provider": {
+            "runtime": {"gudhi": "3.13.0", "python": "3.12.0", "numpy": "2.0"}
+        },
+        "reproduction": {
+            "pairs": expected["reproduction"]["pairs"],
+            "filtration": expected["reproduction"]["filtration"],
+            "provider_output_sha256": fake_digest,
+        },
+        "limitations": ["fabricated"],
+        "extra": True,
+    }
+    task, app, logs = _provider_report_case(tmp_path, "gudhi", report)
+    rejected = support._run_verifier(task, app, logs)
+    assert rejected["reward"] == 0.0
+
+
 def test_nauty_rejects_fabricated_count_and_digests(tmp_path: Path) -> None:
     expected = json.loads(
         (
@@ -443,3 +567,37 @@ def test_nauty_rejects_fabricated_count_and_digests(tmp_path: Path) -> None:
     task, app, logs = _provider_report_case(tmp_path, "nauty", report)
     rejected = support._run_verifier(task, app, logs)
     assert rejected["reward"] == 0.0
+
+
+def test_nauty_rejects_unbound_canonicalization(tmp_path: Path) -> None:
+    expected = json.loads(
+        (
+            DATASETS / "provider-feasibility-v1" / "nauty" / "tests" / "expected.json"
+        ).read_text()
+    )
+    fake_digest = "sha256:" + ("0" * 64)
+    reproduction = expected["reproduction"]
+    canonicalization = {
+        **expected["canonicalization"],
+        "observed_output_sha256": fake_digest,
+        "isomorphic_inputs_converged": True,
+    }
+    report = {
+        "contract": expected["contract"],
+        "status": "COMPLETED",
+        "conclusion": expected["report_conclusion"],
+        "assurance": expected["report_assurance"],
+        "provider": {
+            "executables": {"geng": {"path": "/bin/geng", "sha256": fake_digest}}
+        },
+        "reproduction": {
+            **reproduction,
+            "observed_output_sha256": reproduction["expected_output_sha256"],
+        },
+        "canonicalization": canonicalization,
+        "limitations": ["fabricated"],
+        "extra": True,
+    }
+    task, app, logs = _provider_report_case(tmp_path, "nauty", report)
+    rejected = support._run_verifier(task, app, logs)
+    assert rejected["reward"] == 0.0