$ python3 -I tools/experiments/characterbench_judge_audit.py <bundle>/results/judge_audit.json <validation>/val_zh.jsonl <validation>/val_en.jsonl --repeat <validation>/judge_rep0_zh.jsonl <validation>/judge_rep0_en.jsonl --seeds <validation>/judge_rep1_zh.jsonl <validation>/judge_rep1_en.jsonl <validation>/judge_rep2_zh.jsonl <validation>/judge_rep2_en.jsonl <validation>/judge_rep3_zh.jsonl <validation>/judge_rep3_en.jsonl
# started 2026-10-27T14:03:49+00:00  ended 2026-10-27T14:03:50+00:00  exit 0

zh SC avg Pearson 0.6754 paper 0.68 within 0.05: 13 /13
  repeat (same seed): {'n': 3250, 'identical_10_outputs': 3061, 'share': 0.9418, 'mean_abs_score_diff': 0.0099, 'max_abs_score_diff': 1.2, 'avg_pearson': 0.6759}
  seeds: [0.6754, 0.6759, 0.6747, 0.6783] avg score [3.5964, 3.5921, 3.5931, 3.5936]
  vs released SC: {'n': 3250, 'mean_abs_diff_raw_points': 0.2092, 'pearson_released_sc_avg': 0.6837} stops: 0
en SC avg Pearson 0.6306 paper 0.64 within 0.05: 10 /13
  repeat (same seed): {'n': 3250, 'identical_10_outputs': 3029, 'share': 0.932, 'mean_abs_score_diff': 0.0129, 'max_abs_score_diff': 0.9000000000000004, 'avg_pearson': 0.6281}
  seeds: [0.6306, 0.6273, 0.6259, 0.6236] avg score [3.6631, 3.6605, 3.6624, 3.6621]
  vs released SC: {'n': 3250, 'mean_abs_diff_raw_points': 0.2736, 'pearson_released_sc_avg': 0.655} stops: 0