| |
| """Fail-closed semantic and numerical validator for D5Ijcnz1L9.""" |
| from __future__ import annotations |
|
|
| import csv |
| import hashlib |
| import json |
| import math |
| from pathlib import Path |
|
|
| ROOT = Path(__file__).resolve().parent |
| OUT = ROOT / "outputs" |
|
|
|
|
| def sha256(path: Path) -> str: |
| return hashlib.sha256(path.read_bytes()).hexdigest() |
|
|
|
|
| def rows(name: str) -> list[dict]: |
| with (OUT / name).open(newline="", encoding="utf-8") as handle: |
| return list(csv.DictReader(handle)) |
|
|
|
|
| def numeric_finite(table: list[dict]) -> None: |
| for row in table: |
| for value in row.values(): |
| try: |
| number = float(value) |
| except (TypeError, ValueError): |
| continue |
| assert math.isfinite(number), value |
|
|
|
|
| def main() -> None: |
| results = json.loads((OUT / "results.json").read_text(encoding="utf-8")) |
| claims = json.loads((ROOT / "official_claims.json").read_text(encoding="utf-8")) |
| matrix = json.loads((ROOT / "EVIDENCE_MATRIX.json").read_text(encoding="utf-8")) |
| checksums = json.loads((OUT / "SHA256SUMS.json").read_text(encoding="utf-8")) |
| matrix_rows = matrix["claims"] |
|
|
| assert matrix["paper_id"] == results["paper_id"] == "D5Ijcnz1L9" |
| assert len(claims) == len(matrix_rows) == len(results["registered_claims"]) == 6 |
| assert [item["text"] for item in claims] == results["registered_claims"] |
| assert [item["literal_claim"] for item in matrix_rows] == results["registered_claims"] |
| assert matrix["release_quality_gate"]["expected_verified_points"] == 12 |
| assert matrix["release_quality_gate"]["proxy_support_counted"] is False |
| assert matrix["release_quality_gate"]["formula_only_support_counted"] is False |
| assert all(item["status"] == "supported" for item in matrix_rows) |
| assert all(item["evidence_tier"] == "literal_claim_experiment" for item in matrix_rows) |
| assert all(item["claim_object_match"] == "exact" for item in matrix_rows) |
| assert all(item["registered_system_executed"] for item in matrix_rows) |
| assert all(item["paper_or_released_scale"] for item in matrix_rows) |
| assert all(item["actual_model_or_dataset_used"] for item in matrix_rows) |
| assert all(item["not_proxy_reason"] and item["destructive_or_boundary_control"] for item in matrix_rows) |
| assert all(results["gates"].values()) and len(results["gates"]) == 20 |
|
|
| expected = { |
| "linear_safety.csv": 216, |
| "outlier_radius_sweep.csv": 6, |
| "transfer_m_scaling.csv": 6, |
| "transfer_delta_scaling.csv": 10, |
| "transfer_epsilon_scaling.csv": 6, |
| "balancedness_examples.csv": 3, |
| "spectrum_collapse.csv": 6, |
| "infinite_B_controls.csv": 256, |
| "population_comparability.csv": 1200, |
| "intrinsic_fallback.csv": 600, |
| "glm_grid.csv": 45, |
| "glm_favorable_transfer.csv": 8, |
| "objective_structures.csv": 4, |
| "optimization_diagnostics.csv": 373, |
| "literal_claim_evidence.csv": 6 |
| } |
| loaded = {} |
| for name, count in expected.items(): |
| loaded[name] = rows(name) |
| assert len(loaded[name]) == count, (name, len(loaded[name]), count) |
| numeric_finite(loaded[name]) |
|
|
| summary = results["summary"] |
| assert summary["max_linear_safety_ratio"] < 2.0 |
| assert summary["pooling_ratio_at_radius_100000"] > 1e9 |
| assert -1.18 < summary["m_scaling_slope"] < -0.82 and summary["m_scaling_r2"] > 0.99 |
| assert 1.75 < summary["delta_squared_slope"] < 2.25 and summary["delta_squared_r2"] > 0.99 |
| assert 1.55 < summary["epsilon_squared_slope"] < 2.45 and summary["epsilon_squared_r2"] > 0.99 |
| assert summary["population_comparability_violations"] == 0 |
| assert summary["naive_nu_one_failures"] > 1000 |
| assert summary["intrinsic_fallback_violations"] == 0 |
| assert summary["max_glm_safety_ratio"] < 2.0 |
| assert summary["max_glm_mvt_ratio"] <= 1 / 16 + 1e-10 |
| assert summary["median_glm_itl_over_mtlr"] > 10 |
| assert summary["convexity_violations"] == 0 |
|
|
| for index, item in enumerate(matrix_rows, start=1): |
| page = next(ROOT.glob(f"pages/claim-{index}-*/page.md")) |
| text = page.read_text(encoding="utf-8") |
| assert text.startswith(f"# {results['registered_claims'][index - 1]}") |
| assert "control" in text.lower() |
| assert any(output in text for output in item["executed_outputs"]) |
|
|
| for name, digest in checksums.items(): |
| assert sha256(OUT / name) == digest, name |
| print(json.dumps({"validated_claims": 6, "validated_gates": 20, "validated_tables": len(expected), "validated_output_files": len(checksums)}, sort_keys=True)) |
|
|
|
|
| if __name__ == "__main__": |
| main() |
|
|
