{
  "models": [
    {
      "model": "gemma_4_26b",
      "failure_rates": {
        "F1_agency": {
          "rate_pct": 2.27,
          "ci_half_pct": 5.7,
          "n": 44,
          "failures": 1
        },
        "F2_pov_tense": {
          "rate_pct": 0.0,
          "ci_half_pct": 5.21,
          "n": 33,
          "failures": 0
        }
      },
      "behavioral": {
        "word_count": {
          "value": 350.0042,
          "population_avg": 264.5894,
          "delta": 85.4148
        },
        "unique_word_ratio": {
          "value": 0.5971,
          "population_avg": 0.655,
          "delta": -0.0579
        },
        "bigram_repetition": {
          "value": 0.0688,
          "population_avg": 0.049,
          "delta": 0.0198
        }
      },
      "subjective": {
        "engagement": 4.49,
        "tone_consistency": 4.59,
        "collaboration": 4.32
      },
      "flaw_hunter": {
        "mean": 32.6,
        "median": 33.0,
        "fatal_per_session": 0.62,
        "major_per_session": 7.38,
        "top_flaws": [
          "MAJOR:purple_prose",
          "MAJOR:recycled_description",
          "MAJOR:narrating_emotions"
        ],
        "n_sessions": 16
      },
      "community": {
        "rank": 1,
        "elo": 1535.1,
        "n_votes": 302
      },
      "bayesian": {
        "rank": 1,
        "model": "gemma_4_26b",
        "elo_mean": 1533.7,
        "elo_std": 74.8,
        "ci_low_95": 1404.6,
        "ci_high_95": 1662.4,
        "ci_width_95": 257.8
      },
      "multiturn_judge_mean": 4.29,
      "strength": "Community top tier (#1, ELO 1535)",
      "weakness": "No standout weakness on tested dimensions"
    },
    {
      "model": "mistral_small_creative",
      "failure_rates": {
        "F1_agency": {
          "rate_pct": 15.91,
          "ci_half_pct": 10.72,
          "n": 44,
          "failures": 7
        },
        "F2_pov_tense": {
          "rate_pct": 0.0,
          "ci_half_pct": 5.21,
          "n": 33,
          "failures": 0
        }
      },
      "behavioral": {
        "word_count": {
          "value": 439.2375,
          "population_avg": 264.5894,
          "delta": 174.6481
        },
        "unique_word_ratio": {
          "value": 0.5567,
          "population_avg": 0.655,
          "delta": -0.0983
        },
        "bigram_repetition": {
          "value": 0.0945,
          "population_avg": 0.049,
          "delta": 0.0455
        }
      },
      "subjective": {
        "engagement": 4.46,
        "tone_consistency": 4.59,
        "collaboration": 4.35
      },
      "flaw_hunter": {
        "mean": 27.1,
        "median": 37.0,
        "fatal_per_session": 0.95,
        "major_per_session": 7.7,
        "top_flaws": [
          "MAJOR:purple_prose",
          "MAJOR:recycled_description",
          "FATAL:agency_violation"
        ],
        "n_sessions": 20
      },
      "community": {
        "rank": 2,
        "elo": 1525.7,
        "n_votes": 646
      },
      "bayesian": {
        "rank": 2,
        "model": "mistral_small_creative",
        "elo_mean": 1533.5,
        "elo_std": 74.7,
        "ci_low_95": 1396.3,
        "ci_high_95": 1660.3,
        "ci_width_95": 264.0
      },
      "multiturn_judge_mean": 4.22,
      "strength": "Community top tier (#2, ELO 1526)",
      "weakness": "High agency violation rate (15.9%)"
    },
    {
      "model": "gemini_2_5_flash",
      "failure_rates": {
        "F1_agency": {
          "rate_pct": 0.0,
          "ci_half_pct": 4.01,
          "n": 44,
          "failures": 0
        },
        "F2_pov_tense": {
          "rate_pct": 0.0,
          "ci_half_pct": 5.21,
          "n": 33,
          "failures": 0
        }
      },
      "behavioral": {
        "word_count": {
          "value": 141.1034,
          "population_avg": 264.5894,
          "delta": -123.486
        },
        "unique_word_ratio": {
          "value": 0.7278,
          "population_avg": 0.655,
          "delta": 0.0728
        },
        "bigram_repetition": {
          "value": 0.0299,
          "population_avg": 0.049,
          "delta": -0.0191
        }
      },
      "subjective": {
        "engagement": 4.07,
        "tone_consistency": 4.5,
        "collaboration": 4.33
      },
      "flaw_hunter": {
        "mean": 43.6,
        "median": 41.5,
        "fatal_per_session": 0.19,
        "major_per_session": 6.44,
        "top_flaws": [
          "MAJOR:purple_prose",
          "MAJOR:recycled_description",
          "MAJOR:narrating_emotions"
        ],
        "n_sessions": 16
      },
      "community": {
        "rank": 3,
        "elo": 1515.4,
        "n_votes": 241
      },
      "bayesian": {
        "rank": 3,
        "model": "gemini_2_5_flash",
        "elo_mean": 1529.1,
        "elo_std": 77.2,
        "ci_low_95": 1389.1,
        "ci_high_95": 1664.6,
        "ci_width_95": 275.4
      },
      "multiturn_judge_mean": 4.14,
      "strength": "Community top tier (#3, ELO 1515)",
      "weakness": "Catastrophic floor on agency respect (lowest session: 2.8)"
    },
    {
      "model": "grok_4_1",
      "failure_rates": {
        "F1_agency": {
          "rate_pct": 0.0,
          "ci_half_pct": 4.1,
          "n": 43,
          "failures": 0
        },
        "F2_pov_tense": {
          "rate_pct": 0.0,
          "ci_half_pct": 5.21,
          "n": 33,
          "failures": 0
        }
      },
      "behavioral": {
        "word_count": {
          "value": 136.728,
          "population_avg": 264.5894,
          "delta": -127.8614
        },
        "unique_word_ratio": {
          "value": 0.796,
          "population_avg": 0.655,
          "delta": 0.141
        },
        "bigram_repetition": {
          "value": 0.0151,
          "population_avg": 0.049,
          "delta": -0.0339
        }
      },
      "subjective": {
        "engagement": 4.36,
        "tone_consistency": 4.53,
        "collaboration": 4.26
      },
      "flaw_hunter": {
        "mean": 12.8,
        "median": 34.5,
        "fatal_per_session": 1.33,
        "major_per_session": 8.17,
        "top_flaws": [
          "MAJOR:purple_prose",
          "MAJOR:recycled_description",
          "FATAL:agency_violation"
        ],
        "n_sessions": 12
      },
      "community": {
        "rank": 5,
        "elo": 1505.9,
        "n_votes": 322
      },
      "bayesian": {
        "rank": 4,
        "model": "grok_4_1",
        "elo_mean": 1517.0,
        "elo_std": 75.9,
        "ci_low_95": 1384.8,
        "ci_high_95": 1650.9,
        "ci_width_95": 266.1
      },
      "multiturn_judge_mean": 4.19,
      "strength": "Strong on tone consistency (4.53/5)",
      "weakness": "Bottom-1 on narrative momentum (rank 12/12)"
    },
    {
      "model": "minimax_m2_7",
      "failure_rates": {
        "F1_agency": {
          "rate_pct": 9.3,
          "ci_half_pct": 8.96,
          "n": 43,
          "failures": 4
        },
        "F2_pov_tense": {
          "rate_pct": 0.0,
          "ci_half_pct": 5.21,
          "n": 33,
          "failures": 0
        }
      },
      "behavioral": {
        "word_count": {
          "value": 260.5021,
          "population_avg": 264.5894,
          "delta": -4.0873
        },
        "unique_word_ratio": {
          "value": 0.649,
          "population_avg": 0.655,
          "delta": -0.006
        },
        "bigram_repetition": {
          "value": 0.0458,
          "population_avg": 0.049,
          "delta": -0.0032
        }
      },
      "subjective": {
        "engagement": 4.49,
        "tone_consistency": 4.59,
        "collaboration": 4.33
      },
      "flaw_hunter": {
        "mean": 41.5,
        "median": 44.5,
        "fatal_per_session": 0.79,
        "major_per_session": 6.0,
        "top_flaws": [
          "MAJOR:purple_prose",
          "MAJOR:recycled_description",
          "MAJOR:narrating_emotions"
        ],
        "n_sessions": 14
      },
      "community": {
        "rank": 4,
        "elo": 1510.3,
        "n_votes": 393
      },
      "bayesian": {
        "rank": 5,
        "model": "minimax_m2_7",
        "elo_mean": 1514.4,
        "elo_std": 75.2,
        "ci_low_95": 1386.4,
        "ci_high_95": 1645.6,
        "ci_width_95": 259.2
      },
      "multiturn_judge_mean": 4.34,
      "strength": "Top-2 on narrative momentum (4.30/5)",
      "weakness": "Frequent fatal flaws (0.79 per session, mean score 42/100)"
    },
    {
      "model": "claude_sonnet_4_5",
      "failure_rates": {
        "F1_agency": {
          "rate_pct": 0.0,
          "ci_half_pct": 4.01,
          "n": 44,
          "failures": 0
        },
        "F2_pov_tense": {
          "rate_pct": 0.0,
          "ci_half_pct": 5.21,
          "n": 33,
          "failures": 0
        }
      },
      "behavioral": {
        "word_count": {
          "value": 313.7542,
          "population_avg": 264.5894,
          "delta": 49.1648
        },
        "unique_word_ratio": {
          "value": 0.6253,
          "population_avg": 0.655,
          "delta": -0.0297
        },
        "bigram_repetition": {
          "value": 0.0526,
          "population_avg": 0.049,
          "delta": 0.0036
        }
      },
      "subjective": {
        "engagement": 4.58,
        "tone_consistency": 4.67,
        "collaboration": 4.37
      },
      "flaw_hunter": {
        "mean": 45.3,
        "median": 44.5,
        "fatal_per_session": 0.22,
        "major_per_session": 6.22,
        "top_flaws": [
          "MAJOR:purple_prose",
          "MAJOR:recycled_description",
          "MAJOR:narrating_emotions"
        ],
        "n_sessions": 18
      },
      "community": {
        "rank": 6,
        "elo": 1505.9,
        "n_votes": 194
      },
      "bayesian": {
        "rank": 6,
        "model": "claude_sonnet_4_5",
        "elo_mean": 1513.0,
        "elo_std": 76.5,
        "ci_low_95": 1373.0,
        "ci_high_95": 1653.2,
        "ci_width_95": 280.2
      },
      "multiturn_judge_mean": 4.42,
      "strength": "Top-1 on long-context attention (4.60/5)",
      "weakness": "No standout weakness on tested dimensions"
    },
    {
      "model": "qwen3_5_flash",
      "failure_rates": {
        "F1_agency": {
          "rate_pct": 0.0,
          "ci_half_pct": 4.01,
          "n": 44,
          "failures": 0
        },
        "F2_pov_tense": {
          "rate_pct": 0.0,
          "ci_half_pct": 5.21,
          "n": 33,
          "failures": 0
        }
      },
      "behavioral": {
        "word_count": {
          "value": 229.4625,
          "population_avg": 264.5894,
          "delta": -35.1269
        },
        "unique_word_ratio": {
          "value": 0.6341,
          "population_avg": 0.655,
          "delta": -0.0209
        },
        "bigram_repetition": {
          "value": 0.0693,
          "population_avg": 0.049,
          "delta": 0.0203
        }
      },
      "subjective": {
        "engagement": 4.18,
        "tone_consistency": 4.16,
        "collaboration": 4.13
      },
      "flaw_hunter": {
        "mean": 39.6,
        "median": 39.5,
        "fatal_per_session": 0.5,
        "major_per_session": 6.5,
        "top_flaws": [
          "MAJOR:recycled_description",
          "MAJOR:purple_prose",
          "MAJOR:narrating_emotions"
        ],
        "n_sessions": 18
      },
      "community": {
        "rank": 8,
        "elo": 1486.6,
        "n_votes": 401
      },
      "bayesian": {
        "rank": 7,
        "model": "qwen3_5_flash",
        "elo_mean": 1492.6,
        "elo_std": 75.7,
        "ci_low_95": 1361.0,
        "ci_high_95": 1634.2,
        "ci_width_95": 273.2
      },
      "multiturn_judge_mean": 3.98,
      "strength": "No standout strength on tested dimensions",
      "weakness": "Catastrophic floor on agency respect (lowest session: 2.5)"
    },
    {
      "model": "deepseek_v3_2",
      "failure_rates": {
        "F1_agency": {
          "rate_pct": 4.55,
          "ci_half_pct": 6.94,
          "n": 44,
          "failures": 2
        },
        "F2_pov_tense": {
          "rate_pct": 0.0,
          "ci_half_pct": 5.21,
          "n": 33,
          "failures": 0
        }
      },
      "behavioral": {
        "word_count": {
          "value": 178.1792,
          "population_avg": 264.5894,
          "delta": -86.4102
        },
        "unique_word_ratio": {
          "value": 0.7126,
          "population_avg": 0.655,
          "delta": 0.0576
        },
        "bigram_repetition": {
          "value": 0.0288,
          "population_avg": 0.049,
          "delta": -0.0202
        }
      },
      "subjective": {
        "engagement": 4.61,
        "tone_consistency": 4.63,
        "collaboration": 4.32
      },
      "flaw_hunter": {
        "mean": 46.9,
        "median": 47,
        "fatal_per_session": 0.4,
        "major_per_session": 5.53,
        "top_flaws": [
          "MAJOR:purple_prose",
          "MAJOR:recycled_description",
          "MAJOR:narrating_emotions"
        ],
        "n_sessions": 15
      },
      "community": {
        "rank": 7,
        "elo": 1489.0,
        "n_votes": 241
      },
      "bayesian": {
        "rank": 8,
        "model": "deepseek_v3_2",
        "elo_mean": 1492.1,
        "elo_std": 76.6,
        "ci_low_95": 1357.6,
        "ci_high_95": 1625.4,
        "ci_width_95": 267.8
      },
      "multiturn_judge_mean": 4.38,
      "strength": "Top-2 on lore consistency (4.50/5)",
      "weakness": "NSFW collapse (30.4% NSFW win rate)"
    },
    {
      "model": "glm_4_7",
      "failure_rates": {
        "F1_agency": {
          "rate_pct": 9.09,
          "ci_half_pct": 8.78,
          "n": 44,
          "failures": 4
        },
        "F2_pov_tense": {
          "rate_pct": 0.0,
          "ci_half_pct": 5.36,
          "n": 32,
          "failures": 0
        }
      },
      "behavioral": {
        "word_count": {
          "value": 221.9331,
          "population_avg": 264.5894,
          "delta": -42.6563
        },
        "unique_word_ratio": {
          "value": 0.6667,
          "population_avg": 0.655,
          "delta": 0.0117
        },
        "bigram_repetition": {
          "value": 0.0382,
          "population_avg": 0.049,
          "delta": -0.0108
        }
      },
      "subjective": {
        "engagement": 4.41,
        "tone_consistency": 4.62,
        "collaboration": 4.42
      },
      "flaw_hunter": {
        "mean": 36.8,
        "median": 37,
        "fatal_per_session": 0.71,
        "major_per_session": 6.76,
        "top_flaws": [
          "MAJOR:purple_prose",
          "MAJOR:recycled_description",
          "MAJOR:narrating_emotions"
        ],
        "n_sessions": 17
      },
      "community": {
        "rank": 9,
        "elo": 1483.4,
        "n_votes": 285
      },
      "bayesian": {
        "rank": 9,
        "model": "glm_4_7",
        "elo_mean": 1490.0,
        "elo_std": 75.6,
        "ci_low_95": 1352.6,
        "ci_high_95": 1623.1,
        "ci_width_95": 270.6
      },
      "multiturn_judge_mean": 4.37,
      "strength": "Strong on tone consistency (4.62/5)",
      "weakness": "Lower community half (#9)"
    },
    {
      "model": "llama_4_maverick",
      "failure_rates": {
        "F1_agency": {
          "rate_pct": 6.82,
          "ci_half_pct": 7.94,
          "n": 44,
          "failures": 3
        },
        "F2_pov_tense": {
          "rate_pct": 0.0,
          "ci_half_pct": 5.21,
          "n": 33,
          "failures": 0
        }
      },
      "behavioral": {
        "word_count": {
          "value": 171.5292,
          "population_avg": 264.5894,
          "delta": -93.0602
        },
        "unique_word_ratio": {
          "value": 0.6456,
          "population_avg": 0.655,
          "delta": -0.0094
        },
        "bigram_repetition": {
          "value": 0.0637,
          "population_avg": 0.049,
          "delta": 0.0147
        }
      },
      "subjective": {
        "engagement": 4.15,
        "tone_consistency": 4.3,
        "collaboration": 4.2
      },
      "flaw_hunter": {
        "mean": 30.6,
        "median": 36.5,
        "fatal_per_session": 0.95,
        "major_per_session": 6.65,
        "top_flaws": [
          "MAJOR:recycled_description",
          "MAJOR:purple_prose",
          "FATAL:agency_violation"
        ],
        "n_sessions": 20
      },
      "community": {
        "rank": 10,
        "elo": 1473.0,
        "n_votes": 474
      },
      "bayesian": {
        "rank": 10,
        "model": "llama_4_maverick",
        "elo_mean": 1483.3,
        "elo_std": 75.8,
        "ci_low_95": 1349.8,
        "ci_high_95": 1616.5,
        "ci_width_95": 266.7
      },
      "multiturn_judge_mean": 3.96,
      "strength": "No standout strength on tested dimensions",
      "weakness": "Catastrophic floor on agency respect (lowest session: 3.0)"
    },
    {
      "model": "gpt_4_1",
      "failure_rates": {
        "F1_agency": {
          "rate_pct": 0.0,
          "ci_half_pct": 4.01,
          "n": 44,
          "failures": 0
        },
        "F2_pov_tense": {
          "rate_pct": 0.0,
          "ci_half_pct": 5.21,
          "n": 33,
          "failures": 0
        }
      },
      "behavioral": {
        "word_count": {
          "value": 211.675,
          "population_avg": 264.5894,
          "delta": -52.9144
        },
        "unique_word_ratio": {
          "value": 0.6878,
          "population_avg": 0.655,
          "delta": 0.0328
        },
        "bigram_repetition": {
          "value": 0.0313,
          "population_avg": 0.049,
          "delta": -0.0177
        }
      },
      "subjective": {
        "engagement": 4.38,
        "tone_consistency": 4.6,
        "collaboration": 4.38
      },
      "flaw_hunter": {
        "mean": 27.6,
        "median": 42.0,
        "fatal_per_session": 0.75,
        "major_per_session": 6.83,
        "top_flaws": [
          "MAJOR:purple_prose",
          "MAJOR:recycled_description",
          "FATAL:agency_violation"
        ],
        "n_sessions": 12
      },
      "community": {
        "rank": 11,
        "elo": 1469.7,
        "n_votes": 215
      },
      "bayesian": {
        "rank": 11,
        "model": "gpt_4_1",
        "elo_mean": 1471.9,
        "elo_std": 76.8,
        "ci_low_95": 1337.6,
        "ci_high_95": 1610.1,
        "ci_width_95": 272.5
      },
      "multiturn_judge_mean": 4.34,
      "strength": "Top-1 on narrative momentum (4.30/5)",
      "weakness": "Bottom community ELO (#11/11)"
    },
    {
      "model": "claude_opus_4_7",
      "failure_rates": {
        "F1_agency": {
          "rate_pct": 6.06,
          "ci_half_pct": 8.96,
          "n": 33,
          "failures": 2
        },
        "F2_pov_tense": {
          "rate_pct": 0.0,
          "ci_half_pct": 5.21,
          "n": 33,
          "failures": 0
        }
      },
      "behavioral": {
        "word_count": {
          "value": 407.2431,
          "population_avg": 264.5894,
          "delta": 142.6537
        },
        "unique_word_ratio": {
          "value": 0.5711,
          "population_avg": 0.655,
          "delta": -0.0839
        },
        "bigram_repetition": {
          "value": 0.071,
          "population_avg": 0.049,
          "delta": 0.022
        }
      },
      "subjective": {
        "engagement": 4.61,
        "tone_consistency": 4.75,
        "collaboration": 4.53
      },
      "flaw_hunter": {
        "mean": 42.8,
        "median": 48.0,
        "fatal_per_session": 0.75,
        "major_per_session": 5.92,
        "top_flaws": [
          "MAJOR:purple_prose",
          "MAJOR:recycled_description",
          "FATAL:agency_violation"
        ],
        "n_sessions": 12
      },
      "community": {
        "rank": null,
        "elo": null,
        "n_votes": null
      },
      "bayesian": null,
      "multiturn_judge_mean": 4.54,
      "strength": "Top-1 on agency respect (4.60/5)",
      "weakness": "Frequent fatal flaws (0.75 per session, mean score 43/100)"
    },
    {
      "model": "claude_opus_4_6",
      "failure_rates": {
        "F1_agency": {
          "rate_pct": 4.55,
          "ci_half_pct": 6.94,
          "n": 44,
          "failures": 2
        },
        "F2_pov_tense": {
          "rate_pct": 0.0,
          "ci_half_pct": 5.21,
          "n": 33,
          "failures": 0
        }
      },
      "behavioral": {
        "word_count": {
          "value": 533.525,
          "population_avg": 264.5894,
          "delta": 268.9356
        },
        "unique_word_ratio": {
          "value": 0.551,
          "population_avg": 0.655,
          "delta": -0.104
        },
        "bigram_repetition": {
          "value": 0.0757,
          "population_avg": 0.049,
          "delta": 0.0267
        }
      },
      "subjective": {
        "engagement": 4.67,
        "tone_consistency": 4.75,
        "collaboration": 4.46
      },
      "flaw_hunter": {
        "mean": 40.9,
        "median": 42,
        "fatal_per_session": 0.29,
        "major_per_session": 6.82,
        "top_flaws": [
          "MAJOR:purple_prose",
          "MAJOR:recycled_description",
          "MAJOR:narrating_emotions"
        ],
        "n_sessions": 17
      },
      "community": {
        "rank": null,
        "elo": null,
        "n_votes": null
      },
      "bayesian": null,
      "multiturn_judge_mean": 4.51,
      "strength": "Top-2 on agency respect (4.55/5)",
      "weakness": "High phrase repetition (0.076 vs population 0.049)"
    },
    {
      "model": "deepseek_v4_pro",
      "failure_rates": {
        "F1_agency": {
          "rate_pct": 0.0,
          "ci_half_pct": 5.21,
          "n": 33,
          "failures": 0
        },
        "F2_pov_tense": {
          "rate_pct": 0.0,
          "ci_half_pct": 5.21,
          "n": 33,
          "failures": 0
        }
      },
      "behavioral": {
        "word_count": {
          "value": 258.8333,
          "population_avg": 264.5894,
          "delta": -5.7561
        },
        "unique_word_ratio": {
          "value": 0.6642,
          "population_avg": 0.655,
          "delta": 0.0092
        },
        "bigram_repetition": {
          "value": 0.0402,
          "population_avg": 0.049,
          "delta": -0.0088
        }
      },
      "subjective": {
        "engagement": 4.53,
        "tone_consistency": 4.7,
        "collaboration": 4.38
      },
      "flaw_hunter": {
        "mean": 19.4,
        "median": 46.5,
        "fatal_per_session": 0.5,
        "major_per_session": 9.0,
        "top_flaws": [
          "MAJOR:purple_prose",
          "MAJOR:recycled_description",
          "MAJOR:narrating_emotions"
        ],
        "n_sessions": 8
      },
      "community": {
        "rank": null,
        "elo": null,
        "n_votes": null
      },
      "bayesian": null,
      "multiturn_judge_mean": 4.42,
      "strength": "Strong on tone consistency (4.70/5)",
      "weakness": "No standout weakness on tested dimensions"
    },
    {
      "model": "kimi_k2_5",
      "failure_rates": {
        "F1_agency": {
          "rate_pct": 6.06,
          "ci_half_pct": 8.96,
          "n": 33,
          "failures": 2
        },
        "F2_pov_tense": {
          "rate_pct": 0.0,
          "ci_half_pct": 5.21,
          "n": 33,
          "failures": 0
        }
      },
      "behavioral": {
        "word_count": {
          "value": 252.9444,
          "population_avg": 264.5894,
          "delta": -11.645
        },
        "unique_word_ratio": {
          "value": 0.681,
          "population_avg": 0.655,
          "delta": 0.026
        },
        "bigram_repetition": {
          "value": 0.037,
          "population_avg": 0.049,
          "delta": -0.012
        }
      },
      "subjective": {
        "engagement": 4.56,
        "tone_consistency": 4.67,
        "collaboration": 4.39
      },
      "flaw_hunter": {
        "mean": 42,
        "median": 42,
        "fatal_per_session": 0.44,
        "major_per_session": 6.33,
        "top_flaws": [
          "MAJOR:recycled_description",
          "MAJOR:purple_prose",
          "MAJOR:narrating_emotions"
        ],
        "n_sessions": 9
      },
      "community": {
        "rank": null,
        "elo": null,
        "n_votes": null
      },
      "bayesian": null,
      "multiturn_judge_mean": 4.4,
      "strength": "Strong on tone consistency (4.67/5)",
      "weakness": "No standout weakness on tested dimensions"
    },
    {
      "model": "glm_5_1",
      "failure_rates": {
        "F1_agency": {
          "rate_pct": 3.03,
          "ci_half_pct": 7.39,
          "n": 33,
          "failures": 1
        },
        "F2_pov_tense": {
          "rate_pct": 0.0,
          "ci_half_pct": 5.21,
          "n": 33,
          "failures": 0
        }
      },
      "behavioral": {
        "word_count": {
          "value": 240.0903,
          "population_avg": 264.5894,
          "delta": -24.4991
        },
        "unique_word_ratio": {
          "value": 0.6534,
          "population_avg": 0.655,
          "delta": -0.0016
        },
        "bigram_repetition": {
          "value": 0.0409,
          "population_avg": 0.049,
          "delta": -0.0081
        }
      },
      "subjective": {
        "engagement": 4.49,
        "tone_consistency": 4.65,
        "collaboration": 4.34
      },
      "flaw_hunter": {
        "mean": 45.8,
        "median": 46,
        "fatal_per_session": 0.11,
        "major_per_session": 6.44,
        "top_flaws": [
          "MAJOR:purple_prose",
          "MAJOR:recycled_description",
          "MAJOR:convenient_world"
        ],
        "n_sessions": 9
      },
      "community": {
        "rank": null,
        "elo": null,
        "n_votes": null
      },
      "bayesian": null,
      "multiturn_judge_mean": 4.39,
      "strength": "Strong on tone consistency (4.65/5)",
      "weakness": "No standout weakness on tested dimensions"
    },
    {
      "model": "deepseek_v4_flash",
      "failure_rates": {
        "F1_agency": {
          "rate_pct": 0.0,
          "ci_half_pct": 5.36,
          "n": 32,
          "failures": 0
        },
        "F2_pov_tense": {
          "rate_pct": 0.0,
          "ci_half_pct": 5.36,
          "n": 32,
          "failures": 0
        }
      },
      "behavioral": {
        "word_count": {
          "value": 172.9286,
          "population_avg": 264.5894,
          "delta": -91.6608
        },
        "unique_word_ratio": {
          "value": 0.7088,
          "population_avg": 0.655,
          "delta": 0.0538
        },
        "bigram_repetition": {
          "value": 0.0302,
          "population_avg": 0.049,
          "delta": -0.0188
        }
      },
      "subjective": {
        "engagement": 4.47,
        "tone_consistency": 4.6,
        "collaboration": 4.27
      },
      "flaw_hunter": {
        "mean": 50.6,
        "median": 58,
        "fatal_per_session": 0.36,
        "major_per_session": 5.27,
        "top_flaws": [
          "MAJOR:purple_prose",
          "MAJOR:recycled_description",
          "MAJOR:convenient_world"
        ],
        "n_sessions": 11
      },
      "community": {
        "rank": null,
        "elo": null,
        "n_votes": null
      },
      "bayesian": null,
      "multiturn_judge_mean": 4.38,
      "strength": "Strong on tone consistency (4.60/5)",
      "weakness": "No standout weakness on tested dimensions"
    },
    {
      "model": "gemini_3_1_pro",
      "failure_rates": {
        "F1_agency": {
          "rate_pct": 0.0,
          "ci_half_pct": 5.51,
          "n": 31,
          "failures": 0
        },
        "F2_pov_tense": {
          "rate_pct": 0.0,
          "ci_half_pct": 5.36,
          "n": 32,
          "failures": 0
        }
      },
      "behavioral": {
        "word_count": {
          "value": 263.1241,
          "population_avg": 264.5894,
          "delta": -1.4653
        },
        "unique_word_ratio": {
          "value": 0.6675,
          "population_avg": 0.655,
          "delta": 0.0125
        },
        "bigram_repetition": {
          "value": 0.0401,
          "population_avg": 0.049,
          "delta": -0.0089
        }
      },
      "subjective": {
        "engagement": 4.5,
        "tone_consistency": 4.62,
        "collaboration": 4.42
      },
      "flaw_hunter": {
        "mean": 29.2,
        "median": 35.5,
        "fatal_per_session": 1.0,
        "major_per_session": 6.75,
        "top_flaws": [
          "MAJOR:recycled_description",
          "MAJOR:purple_prose",
          "FATAL:agency_violation"
        ],
        "n_sessions": 12
      },
      "community": {
        "rank": null,
        "elo": null,
        "n_votes": null
      },
      "bayesian": null,
      "multiturn_judge_mean": 4.33,
      "strength": "Strong on tone consistency (4.62/5)",
      "weakness": "Frequent fatal flaws (1.00 per session, mean score 29/100)"
    },
    {
      "model": "gemini_3_1_flash_lite",
      "failure_rates": {
        "F1_agency": {
          "rate_pct": 3.03,
          "ci_half_pct": 7.39,
          "n": 33,
          "failures": 1
        },
        "F2_pov_tense": {
          "rate_pct": 0.0,
          "ci_half_pct": 5.21,
          "n": 33,
          "failures": 0
        }
      },
      "behavioral": {
        "word_count": {
          "value": 263.9236,
          "population_avg": 264.5894,
          "delta": -0.6658
        },
        "unique_word_ratio": {
          "value": 0.6435,
          "population_avg": 0.655,
          "delta": -0.0115
        },
        "bigram_repetition": {
          "value": 0.0485,
          "population_avg": 0.049,
          "delta": -0.0005
        }
      },
      "subjective": {
        "engagement": 4.43,
        "tone_consistency": 4.6,
        "collaboration": 4.31
      },
      "flaw_hunter": {
        "mean": 34.2,
        "median": 34.0,
        "fatal_per_session": 0.17,
        "major_per_session": 8.0,
        "top_flaws": [
          "MAJOR:purple_prose",
          "MAJOR:recycled_description",
          "MAJOR:narrating_emotions"
        ],
        "n_sessions": 6
      },
      "community": {
        "rank": null,
        "elo": null,
        "n_votes": null
      },
      "bayesian": null,
      "multiturn_judge_mean": 4.3,
      "strength": "Strong on tone consistency (4.60/5)",
      "weakness": "No standout weakness on tested dimensions"
    },
    {
      "model": "kimi_k2_6",
      "failure_rates": {
        "F1_agency": {
          "rate_pct": 0.0,
          "ci_half_pct": 5.51,
          "n": 31,
          "failures": 0
        },
        "F2_pov_tense": {
          "rate_pct": 0.0,
          "ci_half_pct": 5.68,
          "n": 30,
          "failures": 0
        }
      },
      "behavioral": {
        "word_count": {
          "value": 221.1797,
          "population_avg": 264.5894,
          "delta": -43.4097
        },
        "unique_word_ratio": {
          "value": 0.677,
          "population_avg": 0.655,
          "delta": 0.022
        },
        "bigram_repetition": {
          "value": 0.0376,
          "population_avg": 0.049,
          "delta": -0.0114
        }
      },
      "subjective": {
        "engagement": 4.44,
        "tone_consistency": 4.39,
        "collaboration": 4.12
      },
      "flaw_hunter": {
        "mean": 49.5,
        "median": 52.0,
        "fatal_per_session": 0.12,
        "major_per_session": 5.25,
        "top_flaws": [
          "MAJOR:recycled_description",
          "MAJOR:purple_prose",
          "MAJOR:convenient_world"
        ],
        "n_sessions": 8
      },
      "community": {
        "rank": null,
        "elo": null,
        "n_votes": null
      },
      "bayesian": null,
      "multiturn_judge_mean": 4.18,
      "strength": "Top-2 on flaw hunter (49.5/100)",
      "weakness": "Catastrophic floor on agency respect (lowest session: 2.5)"
    }
  ]
}