{
 "zh": {
  "n": 3250,
  "base": {
   "columns": {
    "memory_consistency": 0.7975,
    "fact_accuracy": 0.9052,
    "boundary_consistency": 0.7495,
    "attribute_consistency_bot": 0.7942,
    "attribute_consistency_human": 0.6464,
    "behavior_consistency_bot": 0.632,
    "behavior_consistency_human": 0.6325,
    "emotion_self_regulation": 0.6497,
    "empathetic_responsiveness": 0.6608,
    "morality_stability": 0.6285,
    "morality_robustness": 0.5633,
    "human_likeness": 0.5084,
    "engagement": 0.6118
   },
   "avg": 0.6754,
   "paper_avg": 0.68,
   "columns_within_0.05_of_paper": 13
  },
  "vs_released_sc": {
   "n": 3250,
   "mean_abs_diff_raw_points": 0.2092,
   "mean_diff_5pt_per_column": {
    "memory_consistency": -0.072,
    "fact_accuracy": -0.0595,
    "boundary_consistency": 0.006,
    "attribute_consistency_bot": -0.066,
    "attribute_consistency_human": 0.041,
    "behavior_consistency_bot": 0.015,
    "behavior_consistency_human": 0.042,
    "emotion_self_regulation": 0.02,
    "empathetic_responsiveness": 0.0295,
    "morality_stability": -0.045,
    "morality_robustness": -0.041,
    "human_likeness": -0.1708,
    "engagement": -0.0524
   },
   "pearson_released_sc_avg": 0.6837
  },
  "judge_minus_human_5pt": {
   "memory_consistency": 0.0565,
   "fact_accuracy": -0.1205,
   "boundary_consistency": 0.118,
   "attribute_consistency_bot": 0.105,
   "attribute_consistency_human": 0.21,
   "behavior_consistency_bot": 0.053,
   "behavior_consistency_human": 0.1553,
   "emotion_self_regulation": -0.1255,
   "empathetic_responsiveness": -0.17,
   "morality_stability": 0.019,
   "morality_robustness": 0.027,
   "human_likeness": -0.0192,
   "engagement": -0.1148
  },
  "sampled_out_of_range": {},
  "sampled_where_all_eval_stops": 0,
  "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
  },
  "seed_variation": {
   "seed_sets": 4,
   "avg_pearson_per_seed_set": [
    0.6754,
    0.6759,
    0.6747,
    0.6783
   ],
   "avg_pearson_range": 0.0036,
   "column_score_range_5pt": {
    "memory_consistency": 0.0155,
    "fact_accuracy": 0.004,
    "boundary_consistency": 0.006,
    "attribute_consistency_bot": 0.0085,
    "attribute_consistency_human": 0.0095,
    "behavior_consistency_bot": 0.011,
    "behavior_consistency_human": 0.0093,
    "emotion_self_regulation": 0.006,
    "empathetic_responsiveness": 0.0095,
    "morality_stability": 0.008,
    "morality_robustness": 0.003,
    "human_likeness": 0.0188,
    "engagement": 0.0044
   },
   "avg_score_per_seed_set": [
    3.5964,
    3.5921,
    3.5931,
    3.5936
   ]
  }
 },
 "en": {
  "n": 3250,
  "base": {
   "columns": {
    "memory_consistency": 0.7869,
    "fact_accuracy": 0.88,
    "boundary_consistency": 0.6676,
    "attribute_consistency_bot": 0.7716,
    "attribute_consistency_human": 0.5493,
    "behavior_consistency_bot": 0.5321,
    "behavior_consistency_human": 0.6286,
    "emotion_self_regulation": 0.6639,
    "empathetic_responsiveness": 0.6219,
    "morality_stability": 0.5613,
    "morality_robustness": 0.4572,
    "human_likeness": 0.4827,
    "engagement": 0.5947
   },
   "avg": 0.6306,
   "paper_avg": 0.64,
   "columns_within_0.05_of_paper": 10
  },
  "vs_released_sc": {
   "n": 3250,
   "mean_abs_diff_raw_points": 0.2736,
   "mean_diff_5pt_per_column": {
    "memory_consistency": 0.067,
    "fact_accuracy": -0.077,
    "boundary_consistency": -0.2267,
    "attribute_consistency_bot": 0.096,
    "attribute_consistency_human": 0.1145,
    "behavior_consistency_bot": 0.212,
    "behavior_consistency_human": 0.156,
    "emotion_self_regulation": -0.0735,
    "empathetic_responsiveness": -0.119,
    "morality_stability": 0.044,
    "morality_robustness": 0.031,
    "human_likeness": 0.07,
    "engagement": -0.076
   },
   "pearson_released_sc_avg": 0.655
  },
  "judge_minus_human_5pt": {
   "memory_consistency": 0.0435,
   "fact_accuracy": -0.103,
   "boundary_consistency": 0.1573,
   "attribute_consistency_bot": 0.3295,
   "attribute_consistency_human": 0.213,
   "behavior_consistency_bot": 0.254,
   "behavior_consistency_human": 0.2647,
   "emotion_self_regulation": -0.012,
   "empathetic_responsiveness": -0.095,
   "morality_stability": 0.069,
   "morality_robustness": 0.036,
   "human_likeness": 0.0776,
   "engagement": -0.1744
  },
  "sampled_out_of_range": {
   "attribute_consistency_bot": 339
  },
  "sampled_where_all_eval_stops": 0,
  "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
  },
  "seed_variation": {
   "seed_sets": 4,
   "avg_pearson_per_seed_set": [
    0.6306,
    0.6273,
    0.6259,
    0.6236
   ],
   "avg_pearson_range": 0.007,
   "column_score_range_5pt": {
    "memory_consistency": 0.0075,
    "fact_accuracy": 0.005,
    "boundary_consistency": 0.0147,
    "attribute_consistency_bot": 0.0075,
    "attribute_consistency_human": 0.01,
    "behavior_consistency_bot": 0.013,
    "behavior_consistency_human": 0.02,
    "emotion_self_regulation": 0.012,
    "empathetic_responsiveness": 0.017,
    "morality_stability": 0.001,
    "morality_robustness": 0.005,
    "human_likeness": 0.0116,
    "engagement": 0.024
   },
   "avg_score_per_seed_set": [
    3.6631,
    3.6605,
    3.6624,
    3.6621
   ]
  }
 }
}