| |
| """Check public headline values using only packaged aggregate artifacts.""" |
|
|
| from __future__ import annotations |
|
|
| import csv |
| import json |
| import math |
| from pathlib import Path |
|
|
|
|
| ROOT = Path(__file__).resolve().parent.parent |
|
|
|
|
| def close(value: float, expected: float, tolerance: float = 5e-4) -> None: |
| if not math.isclose(value, expected, abs_tol=tolerance, rel_tol=0.0): |
| raise AssertionError(f"expected {expected}, found {value}") |
|
|
|
|
| def rows(path: Path) -> list[dict[str, str]]: |
| with path.open(newline="", encoding="utf-8") as handle: |
| return list(csv.DictReader(handle)) |
|
|
|
|
| controlled = rows(ROOT / "results" / "controlled" / "main_results.csv") |
| for budget, local_expected, topo_expected in [ |
| (0.10, 0.261, 0.147), |
| (0.20, 0.172, 0.059), |
| ]: |
| subset = [r for r in controlled if float(r["budget"]) == budget] |
| for method, expected in [("local-risk", local_expected), ("toporisk", topo_expected)]: |
| values = [float(r["terminal_error"]) for r in subset if r["method"] == method] |
| close(sum(values) / len(values), expected) |
|
|
| fixed = json.loads((ROOT / "results" / "fixed_trace_summary" / "summary.json").read_text()) |
| assert fixed["workflows"] == 18 |
| close(fixed["deepseek_scores_qwen_verifier"]["paired_difference"], -0.526) |
| close(fixed["qwen_scores_deepseek_verifier"]["paired_difference"], -0.544) |
|
|
| online = json.loads((ROOT / "results" / "online_summary" / "combined_summary.json").read_text()) |
| assert online["paired_runs"] == 45 |
| assert online["baseline_success"] == online["verified_success"] == 39 |
| assert online["paired_wins"] == online["paired_losses"] == 2 |
| close(online["relative_total_cost_overhead_percent"], 11.6, tolerance=0.05) |
|
|
| glm = json.loads((ROOT / "results" / "online_summary" / "glm_airline_summary.json").read_text()) |
| assert glm["paired_runs"] == 45 |
| assert glm["baseline_success"] == 40 |
| assert glm["verified_success"] == 39 |
| assert glm["baseline_database_match"] == glm["verified_database_match"] == 42 |
| assert (glm["paired_wins"], glm["paired_losses"], glm["paired_ties"]) == (2, 3, 40) |
| close(glm["success_difference_percentage_points"], -2.2222222222) |
| assert glm["task_cluster_bootstrap_95_ci_percentage_points"] == [ |
| -11.1111111111, |
| 6.6666666667, |
| ] |
|
|
| real_topology = json.loads( |
| (ROOT / "results" / "real_topology_summary" / "summary.json").read_text() |
| ) |
| assert real_topology["graphs"] == 44 |
| assert real_topology["budget_2"]["dynamic_toporisk_exact_optimal_graphs"] == 44 |
| assert real_topology["budget_2"]["static_marginal_exact_optimal_graphs"] == 38 |
| assert real_topology["budget_2"]["wins_losses_ties"] == [6, 0, 38] |
| close(real_topology["budget_2"]["dynamic_toporisk_mean_loss"], 0.136080) |
| close(real_topology["budget_2"]["static_marginal_mean_loss"], 0.140501) |
| close(real_topology["budget_3"]["dynamic_toporisk_mean_loss"], 0.056320) |
| close(real_topology["budget_3"]["static_marginal_mean_loss"], 0.060765) |
|
|
| print("PASS: packaged aggregate artifacts match the public headline values.") |
|
|
