| """ | |
| run_jss_example.py — minimal end-to-end JSS demo. | |
| Loads the factuality task, simulates two sets of judge decisions, | |
| and computes the Judge Sensitivity Score. | |
| """ | |
| import random | |
| import sys | |
| from pathlib import Path | |
| sys.path.insert(0, str(Path(__file__).parent.parent)) | |
| from utils.load_judgesense import load_task | |
| from utils.compute_jss import compute_jss, flip_rate | |
| random.seed(42) | |
| pairs = load_task("factuality", data_dir=Path(__file__).parent.parent / "data") | |
| print(f"Loaded {len(pairs)} factuality pairs") | |
| print(f"Sample pair_id: {pairs[0]['pair_id']}") | |
| print(f"Sample prompt_a: {pairs[0]['prompt_a'][:60]}...") | |
| decisions_a = [random.choice(["accurate", "inaccurate"]) for _ in pairs] | |
| decisions_b = [random.choice(["accurate", "inaccurate"]) for _ in pairs] | |
| jss = compute_jss(decisions_a, decisions_b) | |
| fr = flip_rate(decisions_a, decisions_b) | |
| print(f"\nJSS (simulated): {jss:.3f}") | |
| print(f"Flip rate: {fr:.3f}") | |
| print("(Simulated random decisions — replace with real judge outputs)") | |