Datasets:
benchflow
/
frontierphysics-pr75-evidence

Dataset card Data Studio Files
xet
 Community
text
stringclasses
371 values
Computing standard resonance baseline...
Writing summary.json...
Computing convergence diagnostics...
Computing phi bound sensitivity...
Writing analysis_diagnostics.json...
Generating posterior_diagnostics.png...
Generating report.md...
Analysis complete. Results written to /tmp/pytest-of-root/pytest-3/hidden-cases0/hidden_output_0/
Processing posterior trajectories...
Writing posterior_trajectories.csv...
Writing checkpoint_posteriors.npz...
Computing precision gain...
Writing precision_gain.csv...
Computing strategy statistics...
Computing standard resonance baseline...
Writing summary.json...
Computing convergence diagnostics...
Computing phi bound sensitivity...
Writing analysis_diagnostics.json...
Generating posterior_diagnostics.png...
Generating report.md...
Analysis complete. Results written to /tmp/pytest-of-root/pytest-3/hidden-cases0/hidden_output_1/
________________ test_adversarial_regressions_reject_shortcuts _________________
hidden_outputs = [({'checkpoint_cases': [{'case_id': 'zero-prefix-cd89de4e4359cddf', 'run_id': 'hidden-cd89de4e4359cddf-adaptive-a', 's...pytest-3/hidden-cases0/hidden_input_1.json'), PosixPath('/tmp/pytest-of-root/pytest-3/hidden-cases0/hidden_output_1'))]
tmp_path = PosixPath('/tmp/pytest-of-root/pytest-3/test_adversarial_regressions_r0')
def test_adversarial_regressions_reject_shortcuts(
hidden_outputs: list[tuple[dict[str, Any], Path, Path]],
tmp_path: Path,
) -> None:
payload, _, valid_output = hidden_outputs[0]
dummy_trajectory = tmp_path / "dummy-trajectory"
shutil.copytree(valid_output, dummy_trajectory)
rows = load_trajectory(dummy_trajectory)
for row in rows[4:]:
row["theta_mean"] = "1.0"
row["theta_sd"] = "1.0"
row["phi_mean"] = "10.0"
row["phi_sd"] = "1.0"
with (dummy_trajectory / "posterior_trajectories.csv").open("w", newline="", encoding="utf-8") as handle:
writer = csv.DictWriter(handle, fieldnames=TRAJECTORY_COLUMNS)
writer.writeheader()
writer.writerows(rows)
with pytest.raises(AssertionError):
validate_trajectory(payload, dummy_trajectory)
fake_design = tmp_path / "fake-design"
shutil.copytree(valid_output, fake_design)
design_rows = load_design(fake_design)
for row in design_rows:
row["detuning_mhz"] = "999"
row["zeta"] = "999"
row["expected_variance_reduction"] = "999"
with (fake_design / "precision_gain.csv").open("w", newline="", encoding="utf-8") as handle:
writer = csv.DictWriter(handle, fieldnames=DESIGN_COLUMNS)
writer.writeheader()
writer.writerows(design_rows)
valid_rows = [numeric_trajectory_row(row) for row in load_trajectory(fake_design)]
normalization_error = validate_checkpoints(payload, fake_design, valid_rows)
with pytest.raises(AssertionError):
validate_summary_and_design(payload, fake_design, valid_rows, normalization_error)
fake_summary = tmp_path / "fake-summary"
shutil.copytree(valid_output, fake_summary)
summary = load_json(fake_summary / "summary.json")
summary["model"]["linewidth_mhz"] = 999
summary["design_cases"][0]["case_id"] = "fabricated"
summary["headline_comparison"]["adaptive_final_nsr"] = 999
summary["max_checkpoint_normalization_error"] = 999
(fake_summary / "summary.json").write_text(json.dumps(summary), encoding="utf-8")
valid_rows = [numeric_trajectory_row(row) for row in load_trajectory(fake_summary)]
normalization_error = validate_checkpoints(payload, fake_summary, valid_rows)
with pytest.raises(AssertionError):
validate_summary_and_design(payload, fake_summary, valid_rows, normalization_error)
fake_diagnostics = tmp_path / "fake-diagnostics"
shutil.copytree(valid_output, fake_diagnostics)
diagnostics = load_json(fake_diagnostics / "analysis_diagnostics.json")
diagnostics["convergence"]["max_abs_theta_mean_change"] = 999
diagnostics["phi_bound_sensitivity"]["expanded_bounds"] = [999, 999]
(fake_diagnostics / "analysis_diagnostics.json").write_text(json.dumps(diagnostics), encoding="utf-8")
valid_rows = [numeric_trajectory_row(row) for row in load_trajectory(fake_diagnostics)]
normalization_error = validate_checkpoints(payload, fake_diagnostics, valid_rows)
> valid_summary = validate_summary_and_design(payload, fake_diagnostics, valid_rows, normalization_error)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
/verifier/test_outputs.py:740:
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
/verifier/test_outputs.py:278: in validate_summary_and_design
assert_close(actual[f"shot_{shot}"][key], from_rows[strategy][f"shot_{shot}"][key], atol=1e-10, rtol=1e-10)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
actual = 0.22320398204402384, expected = 0.049820017600308915
def assert_close(actual: float, expected: float, *, atol: float = 3e-3, rtol: float = 3e-3, message: str = "") -> None:
> assert math.isclose(actual, expected, abs_tol=atol, rel_tol=rtol), message or f"{actual} != {expected}"
E AssertionError: 0.22320398204402384 != 0.049820017600308915
E assert False