{
  "schema_version": 1,
  "source_path": "data/lab/experiments/e002.json",
  "source_sha256": "f30d3f6fbb576cf2319472a5ceab62550c771f9c859b0b0f4383397a4d32cdb1",
  "original_summary_sha256": "27d0705c2fcb90cbd968ef86118ee870125d3d6d3853894b0dc37d9bd9487aea",
  "generator": "scripts/render_research_charts.py",
  "matplotlib_version": "3.10.8",
  "scope": "Selected baseline and complete condition matrix when ingested. Numeric original research figures; no game assets.",
  "bootstrap": {
    "resamples": 2000,
    "seed": 2002,
    "unit": "source_run_id",
    "clusters": 71,
    "estimator": "Case-weighted proportion; resample source games and retain all their cases. Paired comparisons retain both predictions per case.",
    "interval": "Percentile 2.5/97.5, exploratory; no multiplicity correction, not proof of equivalence or rare-event safety"
  },
  "series": [
    {
      "model": "qwen3.5:27b",
      "mode": "history3",
      "variant": "crop",
      "n": 200,
      "horizontal_accuracy": 0.795,
      "vertical_accuracy": 0.905,
      "grid_accuracy": 0.755,
      "json_success_rate": 1.0,
      "horizontal_cluster_ci95": [
        0.7264126117179741,
        0.8592292049330883
      ],
      "median_latency_ms": 1375.0,
      "p95_latency_ms": 1422.000000020489,
      "median_encoding_ms": 0.0,
      "median_prefill_ms": 550.865,
      "confusion_matrix": {
        "left": {
          "left": 53,
          "center": 6,
          "right": 7,
          "invalid": 0
        },
        "center": {
          "left": 4,
          "center": 53,
          "right": 9,
          "invalid": 0
        },
        "right": {
          "left": 7,
          "center": 8,
          "right": 53,
          "invalid": 0
        }
      },
      "one_column_error_rate": 0.135,
      "opposite_column_error_rate": 0.07,
      "horizontal_outcome_counts": {
        "exact": 159,
        "adjacent": 27,
        "opposite": 14,
        "invalid": 0
      },
      "one_column_error_cluster_ci95": [
        0.07864617757215245,
        0.19804312892022174
      ],
      "opposite_column_error_cluster_ci95": [
        0.03626359520141903,
        0.10502883922134099
      ],
      "mean_column_error_invalid_penalty_two": 0.275
    },
    {
      "model": "qwen3.5:27b",
      "mode": "history3",
      "variant": "nearest4",
      "n": 200,
      "horizontal_accuracy": 0.91,
      "vertical_accuracy": 0.98,
      "grid_accuracy": 0.91,
      "json_success_rate": 1.0,
      "horizontal_cluster_ci95": [
        0.8620642027052772,
        0.9521339918946302
      ],
      "median_latency_ms": 4390.00000001397,
      "p95_latency_ms": 4452.999999979511,
      "median_encoding_ms": 30.999999959021807,
      "median_prefill_ms": 3463.5815000000002,
      "confusion_matrix": {
        "left": {
          "left": 62,
          "center": 4,
          "right": 0,
          "invalid": 0
        },
        "center": {
          "left": 2,
          "center": 61,
          "right": 3,
          "invalid": 0
        },
        "right": {
          "left": 3,
          "center": 6,
          "right": 59,
          "invalid": 0
        }
      },
      "one_column_error_rate": 0.075,
      "opposite_column_error_rate": 0.015,
      "horizontal_outcome_counts": {
        "exact": 182,
        "adjacent": 15,
        "opposite": 3,
        "invalid": 0
      },
      "one_column_error_cluster_ci95": [
        0.03883092849863219,
        0.11682584576706562
      ],
      "opposite_column_error_cluster_ci95": [
        0.0,
        0.03365519662921347
      ],
      "mean_column_error_invalid_penalty_two": 0.105
    },
    {
      "model": "qwen3.5:27b",
      "mode": "history3",
      "variant": "nearest6",
      "n": 200,
      "horizontal_accuracy": 0.895,
      "vertical_accuracy": 0.935,
      "grid_accuracy": 0.87,
      "json_success_rate": 1.0,
      "horizontal_cluster_ci95": [
        0.8532596230554448,
        0.9365902719372021
      ],
      "median_latency_ms": 9282.00000000652,
      "p95_latency_ms": 9375.0,
      "median_encoding_ms": 61.99999991804361,
      "median_prefill_ms": 8241.4814995,
      "confusion_matrix": {
        "left": {
          "left": 65,
          "center": 1,
          "right": 0,
          "invalid": 0
        },
        "center": {
          "left": 4,
          "center": 55,
          "right": 7,
          "invalid": 0
        },
        "right": {
          "left": 7,
          "center": 2,
          "right": 59,
          "invalid": 0
        }
      },
      "one_column_error_rate": 0.07,
      "opposite_column_error_rate": 0.035,
      "horizontal_outcome_counts": {
        "exact": 179,
        "adjacent": 14,
        "opposite": 7,
        "invalid": 0
      },
      "one_column_error_cluster_ci95": [
        0.03541429611760958,
        0.1055608974358974
      ],
      "opposite_column_error_cluster_ci95": [
        0.010580618212197159,
        0.06573545043847989
      ],
      "mean_column_error_invalid_penalty_two": 0.14
    },
    {
      "model": "qwen3.5:27b",
      "mode": "history3",
      "variant": "raw",
      "n": 200,
      "horizontal_accuracy": 0.755,
      "vertical_accuracy": 0.87,
      "grid_accuracy": 0.725,
      "json_success_rate": 1.0,
      "horizontal_cluster_ci95": [
        0.7018282578413949,
        0.8133416149068323
      ],
      "median_latency_ms": 1359.9999999860302,
      "p95_latency_ms": 1390.999999945052,
      "median_encoding_ms": 0.0,
      "median_prefill_ms": 556.1185,
      "confusion_matrix": {
        "left": {
          "left": 56,
          "center": 6,
          "right": 4,
          "invalid": 0
        },
        "center": {
          "left": 4,
          "center": 58,
          "right": 4,
          "invalid": 0
        },
        "right": {
          "left": 20,
          "center": 11,
          "right": 37,
          "invalid": 0
        }
      },
      "one_column_error_rate": 0.125,
      "opposite_column_error_rate": 0.12,
      "horizontal_outcome_counts": {
        "exact": 151,
        "adjacent": 25,
        "opposite": 24,
        "invalid": 0
      },
      "one_column_error_cluster_ci95": [
        0.08151494565217392,
        0.16810646954560102
      ],
      "opposite_column_error_cluster_ci95": [
        0.07831072894931683,
        0.16326530612244897
      ],
      "mean_column_error_invalid_penalty_two": 0.365
    },
    {
      "model": "qwen3.5:27b",
      "mode": "single",
      "variant": "crop",
      "n": 200,
      "horizontal_accuracy": 0.725,
      "vertical_accuracy": 0.975,
      "grid_accuracy": 0.71,
      "json_success_rate": 1.0,
      "horizontal_cluster_ci95": [
        0.6553520984665052,
        0.7878890253890253
      ],
      "median_latency_ms": 1109.9999999860302,
      "p95_latency_ms": 1140.999999945052,
      "median_encoding_ms": 0.0,
      "median_prefill_ms": 312.7745,
      "confusion_matrix": {
        "left": {
          "left": 41,
          "center": 8,
          "right": 17,
          "invalid": 0
        },
        "center": {
          "left": 6,
          "center": 48,
          "right": 12,
          "invalid": 0
        },
        "right": {
          "left": 8,
          "center": 4,
          "right": 56,
          "invalid": 0
        }
      },
      "one_column_error_rate": 0.15,
      "opposite_column_error_rate": 0.125,
      "horizontal_outcome_counts": {
        "exact": 145,
        "adjacent": 30,
        "opposite": 25,
        "invalid": 0
      },
      "one_column_error_cluster_ci95": [
        0.09826460569777044,
        0.2102710843373493
      ],
      "opposite_column_error_cluster_ci95": [
        0.07909604519774012,
        0.17972350230414746
      ],
      "mean_column_error_invalid_penalty_two": 0.4
    },
    {
      "model": "qwen3.5:27b",
      "mode": "single",
      "variant": "nearest4",
      "n": 200,
      "horizontal_accuracy": 0.845,
      "vertical_accuracy": 0.965,
      "grid_accuracy": 0.84,
      "json_success_rate": 1.0,
      "horizontal_cluster_ci95": [
        0.7965116279069767,
        0.8942765480066762
      ],
      "median_latency_ms": 2170.9999999729916,
      "p95_latency_ms": 2297.7500000153664,
      "median_encoding_ms": 0.0,
      "median_prefill_ms": 1295.8905,
      "confusion_matrix": {
        "left": {
          "left": 64,
          "center": 2,
          "right": 0,
          "invalid": 0
        },
        "center": {
          "left": 2,
          "center": 57,
          "right": 7,
          "invalid": 0
        },
        "right": {
          "left": 19,
          "center": 1,
          "right": 48,
          "invalid": 0
        }
      },
      "one_column_error_rate": 0.06,
      "opposite_column_error_rate": 0.095,
      "horizontal_outcome_counts": {
        "exact": 169,
        "adjacent": 12,
        "opposite": 19,
        "invalid": 0
      },
      "one_column_error_cluster_ci95": [
        0.029757502113271342,
        0.08928761398176291
      ],
      "opposite_column_error_cluster_ci95": [
        0.05586521282931117,
        0.1394291201117318
      ],
      "mean_column_error_invalid_penalty_two": 0.25
    },
    {
      "model": "qwen3.5:27b",
      "mode": "single",
      "variant": "nearest6",
      "n": 200,
      "horizontal_accuracy": 0.605,
      "vertical_accuracy": 0.835,
      "grid_accuracy": 0.57,
      "json_success_rate": 1.0,
      "horizontal_cluster_ci95": [
        0.510989010989011,
        0.6926914350538853
      ],
      "median_latency_ms": 3750.0,
      "p95_latency_ms": 3860.749999986728,
      "median_encoding_ms": 15.000000013969839,
      "median_prefill_ms": 2871.0665,
      "confusion_matrix": {
        "left": {
          "left": 65,
          "center": 1,
          "right": 0,
          "invalid": 0
        },
        "center": {
          "left": 24,
          "center": 33,
          "right": 9,
          "invalid": 0
        },
        "right": {
          "left": 45,
          "center": 0,
          "right": 23,
          "invalid": 0
        }
      },
      "one_column_error_rate": 0.17,
      "opposite_column_error_rate": 0.225,
      "horizontal_outcome_counts": {
        "exact": 121,
        "adjacent": 34,
        "opposite": 45,
        "invalid": 0
      },
      "one_column_error_cluster_ci95": [
        0.12775096072734088,
        0.21910112359550563
      ],
      "opposite_column_error_cluster_ci95": [
        0.16036792452830187,
        0.3005188425493361
      ],
      "mean_column_error_invalid_penalty_two": 0.62
    },
    {
      "model": "qwen3.5:27b",
      "mode": "single",
      "variant": "raw",
      "n": 200,
      "horizontal_accuracy": 0.58,
      "vertical_accuracy": 0.6,
      "grid_accuracy": 0.49,
      "json_success_rate": 1.0,
      "horizontal_cluster_ci95": [
        0.5106327570921986,
        0.6497314068722793
      ],
      "median_latency_ms": 1171.9999999040738,
      "p95_latency_ms": 1218.9999999245629,
      "median_encoding_ms": 0.0,
      "median_prefill_ms": 331.33000000000004,
      "confusion_matrix": {
        "left": {
          "left": 57,
          "center": 4,
          "right": 5,
          "invalid": 0
        },
        "center": {
          "left": 13,
          "center": 53,
          "right": 0,
          "invalid": 0
        },
        "right": {
          "left": 56,
          "center": 6,
          "right": 6,
          "invalid": 0
        }
      },
      "one_column_error_rate": 0.115,
      "opposite_column_error_rate": 0.305,
      "horizontal_outcome_counts": {
        "exact": 116,
        "adjacent": 23,
        "opposite": 61,
        "invalid": 0
      },
      "one_column_error_cluster_ci95": [
        0.0625,
        0.16742134062927494
      ],
      "opposite_column_error_cluster_ci95": [
        0.23231872294372294,
        0.38267975794969133
      ],
      "mean_column_error_invalid_penalty_two": 0.725
    },
    {
      "model": "qwen3.5:35b-a3b",
      "mode": "history3",
      "variant": "crop",
      "n": 200,
      "horizontal_accuracy": 0.865,
      "vertical_accuracy": 0.98,
      "grid_accuracy": 0.86,
      "json_success_rate": 1.0,
      "horizontal_cluster_ci95": [
        0.8258376548545088,
        0.9067417761176045
      ],
      "median_latency_ms": 1375.0,
      "p95_latency_ms": 1469.0000000409782,
      "median_encoding_ms": 0.0,
      "median_prefill_ms": 938.5909999999999,
      "confusion_matrix": {
        "left": {
          "left": 60,
          "center": 5,
          "right": 1,
          "invalid": 0
        },
        "center": {
          "left": 4,
          "center": 53,
          "right": 9,
          "invalid": 0
        },
        "right": {
          "left": 1,
          "center": 7,
          "right": 60,
          "invalid": 0
        }
      },
      "one_column_error_rate": 0.125,
      "opposite_column_error_rate": 0.01,
      "horizontal_outcome_counts": {
        "exact": 173,
        "adjacent": 25,
        "opposite": 2,
        "invalid": 0
      },
      "one_column_error_cluster_ci95": [
        0.08287116866664321,
        0.16666666666666666
      ],
      "opposite_column_error_cluster_ci95": [
        0.0,
        0.02500314070351757
      ],
      "mean_column_error_invalid_penalty_two": 0.145
    },
    {
      "model": "qwen3.5:35b-a3b",
      "mode": "history3",
      "variant": "nearest4",
      "n": 200,
      "horizontal_accuracy": 0.86,
      "vertical_accuracy": 0.995,
      "grid_accuracy": 0.86,
      "json_success_rate": 1.0,
      "horizontal_cluster_ci95": [
        0.803272478887233,
        0.9154262023217248
      ],
      "median_latency_ms": 4031.000000075437,
      "p95_latency_ms": 4062.999999965541,
      "median_encoding_ms": 30.999999959021807,
      "median_prefill_ms": 3493.05,
      "confusion_matrix": {
        "left": {
          "left": 61,
          "center": 5,
          "right": 0,
          "invalid": 0
        },
        "center": {
          "left": 2,
          "center": 53,
          "right": 11,
          "invalid": 0
        },
        "right": {
          "left": 10,
          "center": 0,
          "right": 58,
          "invalid": 0
        }
      },
      "one_column_error_rate": 0.09,
      "opposite_column_error_rate": 0.05,
      "horizontal_outcome_counts": {
        "exact": 172,
        "adjacent": 18,
        "opposite": 10,
        "invalid": 0
      },
      "one_column_error_cluster_ci95": [
        0.043478260869565216,
        0.14000241545893718
      ],
      "opposite_column_error_cluster_ci95": [
        0.017234073641145528,
        0.09340787716955941
      ],
      "mean_column_error_invalid_penalty_two": 0.19
    },
    {
      "model": "qwen3.5:35b-a3b",
      "mode": "history3",
      "variant": "nearest6",
      "n": 200,
      "horizontal_accuracy": 0.87,
      "vertical_accuracy": 0.985,
      "grid_accuracy": 0.87,
      "json_success_rate": 1.0,
      "horizontal_cluster_ci95": [
        0.8275823924321025,
        0.9162310112707297
      ],
      "median_latency_ms": 8875.0,
      "p95_latency_ms": 8921.999999909895,
      "median_encoding_ms": 47.0000000204891,
      "median_prefill_ms": 8216.389499,
      "confusion_matrix": {
        "left": {
          "left": 62,
          "center": 4,
          "right": 0,
          "invalid": 0
        },
        "center": {
          "left": 4,
          "center": 50,
          "right": 12,
          "invalid": 0
        },
        "right": {
          "left": 6,
          "center": 0,
          "right": 62,
          "invalid": 0
        }
      },
      "one_column_error_rate": 0.1,
      "opposite_column_error_rate": 0.03,
      "horizontal_outcome_counts": {
        "exact": 174,
        "adjacent": 20,
        "opposite": 6,
        "invalid": 0
      },
      "one_column_error_cluster_ci95": [
        0.05746988945718528,
        0.14287315823190253
      ],
      "opposite_column_error_cluster_ci95": [
        0.008200747696052862,
        0.06018518518518518
      ],
      "mean_column_error_invalid_penalty_two": 0.16
    },
    {
      "model": "qwen3.5:35b-a3b",
      "mode": "history3",
      "variant": "raw",
      "n": 200,
      "horizontal_accuracy": 0.765,
      "vertical_accuracy": 0.88,
      "grid_accuracy": 0.76,
      "json_success_rate": 1.0,
      "horizontal_cluster_ci95": [
        0.6989749815588887,
        0.825248442938267
      ],
      "median_latency_ms": 1468.4999999590218,
      "p95_latency_ms": 1515.9999999508727,
      "median_encoding_ms": 0.0,
      "median_prefill_ms": 985.4615,
      "confusion_matrix": {
        "left": {
          "left": 60,
          "center": 5,
          "right": 1,
          "invalid": 0
        },
        "center": {
          "left": 4,
          "center": 57,
          "right": 5,
          "invalid": 0
        },
        "right": {
          "left": 30,
          "center": 2,
          "right": 36,
          "invalid": 0
        }
      },
      "one_column_error_rate": 0.08,
      "opposite_column_error_rate": 0.155,
      "horizontal_outcome_counts": {
        "exact": 153,
        "adjacent": 16,
        "opposite": 31,
        "invalid": 0
      },
      "one_column_error_cluster_ci95": [
        0.04244971049216332,
        0.1217444326617179
      ],
      "opposite_column_error_cluster_ci95": [
        0.10526315789473684,
        0.21143842364532014
      ],
      "mean_column_error_invalid_penalty_two": 0.39
    },
    {
      "model": "qwen3.5:35b-a3b",
      "mode": "single",
      "variant": "crop",
      "n": 200,
      "horizontal_accuracy": 0.775,
      "vertical_accuracy": 0.98,
      "grid_accuracy": 0.76,
      "json_success_rate": 1.0,
      "horizontal_cluster_ci95": [
        0.7125653570906966,
        0.8333333333333334
      ],
      "median_latency_ms": 907.0000000065193,
      "p95_latency_ms": 1000.0,
      "median_encoding_ms": 0.0,
      "median_prefill_ms": 536.921,
      "confusion_matrix": {
        "left": {
          "left": 62,
          "center": 4,
          "right": 0,
          "invalid": 0
        },
        "center": {
          "left": 13,
          "center": 36,
          "right": 17,
          "invalid": 0
        },
        "right": {
          "left": 6,
          "center": 5,
          "right": 57,
          "invalid": 0
        }
      },
      "one_column_error_rate": 0.195,
      "opposite_column_error_rate": 0.03,
      "horizontal_outcome_counts": {
        "exact": 155,
        "adjacent": 39,
        "opposite": 6,
        "invalid": 0
      },
      "one_column_error_cluster_ci95": [
        0.13888487475915223,
        0.25159490005604324
      ],
      "opposite_column_error_cluster_ci95": [
        0.009900990099009901,
        0.05527638190954774
      ],
      "mean_column_error_invalid_penalty_two": 0.255
    },
    {
      "model": "qwen3.5:35b-a3b",
      "mode": "single",
      "variant": "nearest4",
      "n": 200,
      "horizontal_accuracy": 0.765,
      "vertical_accuracy": 0.935,
      "grid_accuracy": 0.75,
      "json_success_rate": 1.0,
      "horizontal_cluster_ci95": [
        0.6984002976190475,
        0.8199056672295764
      ],
      "median_latency_ms": 1765.999999945052,
      "p95_latency_ms": 1844.7500000358557,
      "median_encoding_ms": 0.0,
      "median_prefill_ms": 1353.307,
      "confusion_matrix": {
        "left": {
          "left": 65,
          "center": 1,
          "right": 0,
          "invalid": 0
        },
        "center": {
          "left": 5,
          "center": 44,
          "right": 17,
          "invalid": 0
        },
        "right": {
          "left": 24,
          "center": 0,
          "right": 44,
          "invalid": 0
        }
      },
      "one_column_error_rate": 0.115,
      "opposite_column_error_rate": 0.12,
      "horizontal_outcome_counts": {
        "exact": 153,
        "adjacent": 23,
        "opposite": 24,
        "invalid": 0
      },
      "one_column_error_cluster_ci95": [
        0.07514450867052024,
        0.15178688294689602
      ],
      "opposite_column_error_cluster_ci95": [
        0.07368421052631578,
        0.17778735632183904
      ],
      "mean_column_error_invalid_penalty_two": 0.355
    },
    {
      "model": "qwen3.5:35b-a3b",
      "mode": "single",
      "variant": "nearest6",
      "n": 200,
      "horizontal_accuracy": 0.725,
      "vertical_accuracy": 0.865,
      "grid_accuracy": 0.72,
      "json_success_rate": 1.0,
      "horizontal_cluster_ci95": [
        0.6553623611922852,
        0.7894929396662387
      ],
      "median_latency_ms": 3375.0,
      "p95_latency_ms": 3452.999999979511,
      "median_encoding_ms": 15.999999945051968,
      "median_prefill_ms": 2934.7415,
      "confusion_matrix": {
        "left": {
          "left": 66,
          "center": 0,
          "right": 0,
          "invalid": 0
        },
        "center": {
          "left": 9,
          "center": 41,
          "right": 16,
          "invalid": 0
        },
        "right": {
          "left": 30,
          "center": 0,
          "right": 38,
          "invalid": 0
        }
      },
      "one_column_error_rate": 0.125,
      "opposite_column_error_rate": 0.15,
      "horizontal_outcome_counts": {
        "exact": 145,
        "adjacent": 25,
        "opposite": 30,
        "invalid": 0
      },
      "one_column_error_cluster_ci95": [
        0.08092193612424826,
        0.16585365853658537
      ],
      "opposite_column_error_cluster_ci95": [
        0.10214531788286878,
        0.20670872664226544
      ],
      "mean_column_error_invalid_penalty_two": 0.425
    },
    {
      "model": "qwen3.5:35b-a3b",
      "mode": "single",
      "variant": "raw",
      "n": 200,
      "horizontal_accuracy": 0.59,
      "vertical_accuracy": 0.595,
      "grid_accuracy": 0.505,
      "json_success_rate": 1.0,
      "horizontal_cluster_ci95": [
        0.5045878533859304,
        0.677252944188428
      ],
      "median_latency_ms": 952.9999999795109,
      "p95_latency_ms": 1030.9999999590218,
      "median_encoding_ms": 0.0,
      "median_prefill_ms": 557.8530000000001,
      "confusion_matrix": {
        "left": {
          "left": 64,
          "center": 2,
          "right": 0,
          "invalid": 0
        },
        "center": {
          "left": 21,
          "center": 38,
          "right": 7,
          "invalid": 0
        },
        "right": {
          "left": 50,
          "center": 2,
          "right": 16,
          "invalid": 0
        }
      },
      "one_column_error_rate": 0.16,
      "opposite_column_error_rate": 0.25,
      "horizontal_outcome_counts": {
        "exact": 118,
        "adjacent": 32,
        "opposite": 50,
        "invalid": 0
      },
      "one_column_error_cluster_ci95": [
        0.11170145264067394,
        0.20442567133496078
      ],
      "opposite_column_error_cluster_ci95": [
        0.18084198235599377,
        0.32984428131915583
      ],
      "mean_column_error_invalid_penalty_two": 0.66
    },
    {
      "model": "qwen3.5:9b",
      "mode": "history3",
      "variant": "crop",
      "n": 200,
      "horizontal_accuracy": 0.74,
      "vertical_accuracy": 0.995,
      "grid_accuracy": 0.74,
      "json_success_rate": 1.0,
      "horizontal_cluster_ci95": [
        0.666626213592233,
        0.8043581780538301
      ],
      "median_latency_ms": 563.0000000819564,
      "p95_latency_ms": 625.0,
      "median_encoding_ms": 0.0,
      "median_prefill_ms": 217.638,
      "confusion_matrix": {
        "left": {
          "left": 55,
          "center": 4,
          "right": 7,
          "invalid": 0
        },
        "center": {
          "left": 4,
          "center": 39,
          "right": 23,
          "invalid": 0
        },
        "right": {
          "left": 4,
          "center": 10,
          "right": 54,
          "invalid": 0
        }
      },
      "one_column_error_rate": 0.205,
      "opposite_column_error_rate": 0.055,
      "horizontal_outcome_counts": {
        "exact": 148,
        "adjacent": 41,
        "opposite": 11,
        "invalid": 0
      },
      "one_column_error_cluster_ci95": [
        0.1449981884057971,
        0.27108688993261176
      ],
      "opposite_column_error_cluster_ci95": [
        0.02347143609319468,
        0.09743902439024389
      ],
      "mean_column_error_invalid_penalty_two": 0.315
    },
    {
      "model": "qwen3.5:9b",
      "mode": "history3",
      "variant": "nearest4",
      "n": 200,
      "horizontal_accuracy": 0.9,
      "vertical_accuracy": 1.0,
      "grid_accuracy": 0.9,
      "json_success_rate": 1.0,
      "horizontal_cluster_ci95": [
        0.8585354348826507,
        0.9411764705882353
      ],
      "median_latency_ms": 1984.000000054948,
      "p95_latency_ms": 2032.0000000065193,
      "median_encoding_ms": 15.999999945051968,
      "median_prefill_ms": 1587.246,
      "confusion_matrix": {
        "left": {
          "left": 66,
          "center": 0,
          "right": 0,
          "invalid": 0
        },
        "center": {
          "left": 4,
          "center": 49,
          "right": 13,
          "invalid": 0
        },
        "right": {
          "left": 3,
          "center": 0,
          "right": 65,
          "invalid": 0
        }
      },
      "one_column_error_rate": 0.085,
      "opposite_column_error_rate": 0.015,
      "horizontal_outcome_counts": {
        "exact": 180,
        "adjacent": 17,
        "opposite": 3,
        "invalid": 0
      },
      "one_column_error_cluster_ci95": [
        0.04503623735467311,
        0.12390526846714653
      ],
      "opposite_column_error_cluster_ci95": [
        0.0,
        0.04124785223367692
      ],
      "mean_column_error_invalid_penalty_two": 0.115
    },
    {
      "model": "qwen3.5:9b",
      "mode": "history3",
      "variant": "nearest6",
      "n": 200,
      "horizontal_accuracy": 0.845,
      "vertical_accuracy": 0.98,
      "grid_accuracy": 0.845,
      "json_success_rate": 1.0,
      "horizontal_cluster_ci95": [
        0.7873445449406444,
        0.8978546821171312
      ],
      "median_latency_ms": 5000.0,
      "p95_latency_ms": 5125.0,
      "median_encoding_ms": 32.00000000651926,
      "median_prefill_ms": 4557.313,
      "confusion_matrix": {
        "left": {
          "left": 66,
          "center": 0,
          "right": 0,
          "invalid": 0
        },
        "center": {
          "left": 9,
          "center": 44,
          "right": 13,
          "invalid": 0
        },
        "right": {
          "left": 9,
          "center": 0,
          "right": 59,
          "invalid": 0
        }
      },
      "one_column_error_rate": 0.11,
      "opposite_column_error_rate": 0.045,
      "horizontal_outcome_counts": {
        "exact": 169,
        "adjacent": 22,
        "opposite": 9,
        "invalid": 0
      },
      "one_column_error_cluster_ci95": [
        0.06697767145135566,
        0.15609756097560976
      ],
      "opposite_column_error_cluster_ci95": [
        0.019046112115732372,
        0.07653122253686163
      ],
      "mean_column_error_invalid_penalty_two": 0.2
    },
    {
      "model": "qwen3.5:9b",
      "mode": "history3",
      "variant": "raw",
      "n": 200,
      "horizontal_accuracy": 0.655,
      "vertical_accuracy": 0.965,
      "grid_accuracy": 0.655,
      "json_success_rate": 1.0,
      "horizontal_cluster_ci95": [
        0.5707049306625578,
        0.7354555625480735
      ],
      "median_latency_ms": 532.0000000065193,
      "p95_latency_ms": 608.9999999443535,
      "median_encoding_ms": 0.0,
      "median_prefill_ms": 216.698,
      "confusion_matrix": {
        "left": {
          "left": 56,
          "center": 8,
          "right": 2,
          "invalid": 0
        },
        "center": {
          "left": 2,
          "center": 48,
          "right": 16,
          "invalid": 0
        },
        "right": {
          "left": 37,
          "center": 4,
          "right": 27,
          "invalid": 0
        }
      },
      "one_column_error_rate": 0.15,
      "opposite_column_error_rate": 0.195,
      "horizontal_outcome_counts": {
        "exact": 131,
        "adjacent": 30,
        "opposite": 39,
        "invalid": 0
      },
      "one_column_error_cluster_ci95": [
        0.0962274417818351,
        0.20833333333333334
      ],
      "opposite_column_error_cluster_ci95": [
        0.14367663242846662,
        0.2553250591016548
      ],
      "mean_column_error_invalid_penalty_two": 0.54
    },
    {
      "model": "qwen3.5:9b",
      "mode": "single",
      "variant": "crop",
      "n": 200,
      "horizontal_accuracy": 0.58,
      "vertical_accuracy": 0.995,
      "grid_accuracy": 0.58,
      "json_success_rate": 1.0,
      "horizontal_cluster_ci95": [
        0.5024863487440845,
        0.6602997280947663
      ],
      "median_latency_ms": 452.9999999795109,
      "p95_latency_ms": 515.999999945052,
      "median_encoding_ms": 0.0,
      "median_prefill_ms": 122.65100000000001,
      "confusion_matrix": {
        "left": {
          "left": 63,
          "center": 3,
          "right": 0,
          "invalid": 0
        },
        "center": {
          "left": 25,
          "center": 28,
          "right": 13,
          "invalid": 0
        },
        "right": {
          "left": 37,
          "center": 6,
          "right": 25,
          "invalid": 0
        }
      },
      "one_column_error_rate": 0.235,
      "opposite_column_error_rate": 0.185,
      "horizontal_outcome_counts": {
        "exact": 116,
        "adjacent": 47,
        "opposite": 37,
        "invalid": 0
      },
      "one_column_error_cluster_ci95": [
        0.17764323975656954,
        0.2912621359223301
      ],
      "opposite_column_error_cluster_ci95": [
        0.12059998789126354,
        0.2545620265151514
      ],
      "mean_column_error_invalid_penalty_two": 0.605
    },
    {
      "model": "qwen3.5:9b",
      "mode": "single",
      "variant": "nearest4",
      "n": 200,
      "horizontal_accuracy": 0.8,
      "vertical_accuracy": 0.995,
      "grid_accuracy": 0.8,
      "json_success_rate": 1.0,
      "horizontal_cluster_ci95": [
        0.7368276460381723,
        0.8612783459315251
      ],
      "median_latency_ms": 937.9999999655411,
      "p95_latency_ms": 984.9999999860302,
      "median_encoding_ms": 0.0,
      "median_prefill_ms": 580.894,
      "confusion_matrix": {
        "left": {
          "left": 66,
          "center": 0,
          "right": 0,
          "invalid": 0
        },
        "center": {
          "left": 6,
          "center": 50,
          "right": 10,
          "invalid": 0
        },
        "right": {
          "left": 24,
          "center": 0,
          "right": 44,
          "invalid": 0
        }
      },
      "one_column_error_rate": 0.08,
      "opposite_column_error_rate": 0.12,
      "horizontal_outcome_counts": {
        "exact": 160,
        "adjacent": 16,
        "opposite": 24,
        "invalid": 0
      },
      "one_column_error_cluster_ci95": [
        0.047841193893825476,
        0.11310036945812806
      ],
      "opposite_column_error_cluster_ci95": [
        0.07291413834951457,
        0.17649430740037939
      ],
      "mean_column_error_invalid_penalty_two": 0.32
    },
    {
      "model": "qwen3.5:9b",
      "mode": "single",
      "variant": "nearest6",
      "n": 200,
      "horizontal_accuracy": 0.475,
      "vertical_accuracy": 0.875,
      "grid_accuracy": 0.42,
      "json_success_rate": 1.0,
      "horizontal_cluster_ci95": [
        0.38094633792603333,
        0.5652173913043478
      ],
      "median_latency_ms": 1938.0000000237487,
      "p95_latency_ms": 2000.0,
      "median_encoding_ms": 15.000000013969839,
      "median_prefill_ms": 1573.1354999999999,
      "confusion_matrix": {
        "left": {
          "left": 66,
          "center": 0,
          "right": 0,
          "invalid": 0
        },
        "center": {
          "left": 45,
          "center": 19,
          "right": 2,
          "invalid": 0
        },
        "right": {
          "left": 58,
          "center": 0,
          "right": 10,
          "invalid": 0
        }
      },
      "one_column_error_rate": 0.235,
      "opposite_column_error_rate": 0.29,
      "horizontal_outcome_counts": {
        "exact": 95,
        "adjacent": 47,
        "opposite": 58,
        "invalid": 0
      },
      "one_column_error_cluster_ci95": [
        0.18106809675437127,
        0.29546065493646134
      ],
      "opposite_column_error_cluster_ci95": [
        0.21950567366636234,
        0.3687150837988827
      ],
      "mean_column_error_invalid_penalty_two": 0.815
    },
    {
      "model": "qwen3.5:9b",
      "mode": "single",
      "variant": "raw",
      "n": 200,
      "horizontal_accuracy": 0.52,
      "vertical_accuracy": 0.8,
      "grid_accuracy": 0.48,
      "json_success_rate": 1.0,
      "horizontal_cluster_ci95": [
        0.4303012684989429,
        0.6085010741640201
      ],
      "median_latency_ms": 468.4999999590218,
      "p95_latency_ms": 515.9999999508727,
      "median_encoding_ms": 0.0,
      "median_prefill_ms": 128.1519995,
      "confusion_matrix": {
        "left": {
          "left": 63,
          "center": 2,
          "right": 1,
          "invalid": 0
        },
        "center": {
          "left": 23,
          "center": 32,
          "right": 11,
          "invalid": 0
        },
        "right": {
          "left": 57,
          "center": 2,
          "right": 9,
          "invalid": 0
        }
      },
      "one_column_error_rate": 0.19,
      "opposite_column_error_rate": 0.29,
      "horizontal_outcome_counts": {
        "exact": 104,
        "adjacent": 38,
        "opposite": 58,
        "invalid": 0
      },
      "one_column_error_cluster_ci95": [
        0.13989241351222934,
        0.23980102040816323
      ],
      "opposite_column_error_cluster_ci95": [
        0.21527706687437412,
        0.3707491289198605
      ],
      "mean_column_error_invalid_penalty_two": 0.77
    }
  ],
  "files": [
    {
      "path": "e002-localization-v1.svg",
      "sha256": "1b970f99b03853635ba987c8b8b817aedc55306f9cce6016b8d858f218026f5b",
      "size": 66169
    },
    {
      "path": "e002-localization-v1.png",
      "sha256": "c7f8e9b27c7182aa22cec76b8bc6880d4e1d8ccfad34c14b34cf9f4f6de93d6e",
      "size": 84979
    },
    {
      "path": "e002-localization-v1-mobile.svg",
      "sha256": "98df202034fb828a86b95561accbed212dc9be99698dd73ab7788e641cbae16c",
      "size": 70812
    },
    {
      "path": "e002-accuracy-latency-v1.svg",
      "sha256": "32c976c0645bbfa9f66f91df6d2cd9ea8223fcf57a73b7819663d0abde3e6bb8",
      "size": 114842
    },
    {
      "path": "e002-accuracy-latency-v1.png",
      "sha256": "d692f2c095931f8dfda9eb864eb2f1fdf02081ddd1ebe7846220602008bded85",
      "size": 107421
    },
    {
      "path": "e002-accuracy-latency-v1-mobile.svg",
      "sha256": "546ec0418df281b2d7d3adf8cc9da22c346ddf87a422c53ec920d50fcc9f7fe8",
      "size": 117305
    }
  ]
}
