{
  "generated": "2026-07-03",
  "prereg": "preregistration/2026-07-03_study2-2x2.md",
  "n_items": 89,
  "bootstrap": {
    "resamples": 10000,
    "seed": 20260703
  },
  "cells": {
    "HOMO_NT": "7x claude-haiku, no tool",
    "HOMO_RET": "7x claude-haiku, retrieval",
    "HET_NT": "4x claude-haiku + 3x gpt-4o-mini, no tool",
    "HET_RET": "4x claude-haiku + 3x gpt-4o-mini, retrieval"
  },
  "arms": {
    "HOMO_NT_r0": {
      "n": 89,
      "accuracy": 0.7528,
      "mean_confidence": 0.8186,
      "overconfidence": 0.0658,
      "ece": 0.0722,
      "mce": 0.1429,
      "brier": 0.1263,
      "auc": 0.826
    },
    "HOMO_NT_post": {
      "n": 89,
      "accuracy": 0.7528,
      "mean_confidence": 0.9181,
      "overconfidence": 0.1653,
      "ece": 0.1782,
      "mce": 0.4286,
      "brier": 0.1731,
      "auc": 0.6971
    },
    "HOMO_RET_r0": {
      "n": 89,
      "accuracy": 0.7528,
      "mean_confidence": 0.8122,
      "overconfidence": 0.0594,
      "ece": 0.0883,
      "mce": 0.2857,
      "brier": 0.1121,
      "auc": 0.866
    },
    "HOMO_RET_post": {
      "n": 89,
      "accuracy": 0.7416,
      "mean_confidence": 0.8973,
      "overconfidence": 0.1557,
      "ece": 0.1557,
      "mce": 0.6905,
      "brier": 0.1401,
      "auc": 0.862
    },
    "HET_NT_r0": {
      "n": 89,
      "accuracy": 0.7865,
      "mean_confidence": 0.7897,
      "overconfidence": 0.0032,
      "ece": 0.1284,
      "mce": 0.2857,
      "brier": 0.1252,
      "auc": 0.7966
    },
    "HET_NT_post": {
      "n": 89,
      "accuracy": 0.7753,
      "mean_confidence": 0.8748,
      "overconfidence": 0.0995,
      "ece": 0.1798,
      "mce": 0.5714,
      "brier": 0.1463,
      "auc": 0.7413
    },
    "HET_RET_r0": {
      "n": 89,
      "accuracy": 0.8427,
      "mean_confidence": 0.7689,
      "overconfidence": -0.0738,
      "ece": 0.1252,
      "mce": 0.4286,
      "brier": 0.1188,
      "auc": 0.8162
    },
    "HET_RET_post": {
      "n": 89,
      "accuracy": 0.7978,
      "mean_confidence": 0.8748,
      "overconfidence": 0.077,
      "ece": 0.0963,
      "mce": 0.2857,
      "brier": 0.1238,
      "auc": 0.7883
    }
  },
  "P_1_primary_fully_equipped_vs_bare": {
    "delta_ci": {
      "mean": 0.0887,
      "lo": 0.0193,
      "hi": 0.1637
    },
    "verdict": "CONFIRMED"
  },
  "P_2_thesis_het_ret_still_overconfident": {
    "overconf_ci": {
      "mean": 0.0772,
      "lo": 0.008,
      "hi": 0.1509
    },
    "verdict": "STILL OVERCONFIDENT"
  },
  "P_3_interaction_retrieval_by_family": {
    "delta_ci": {
      "mean": -0.0123,
      "lo": -0.0995,
      "hi": 0.0754
    },
    "note": "estimation: >0 means retrieval disciplines more in the homogeneous swarm; <0 more in the heterogeneous swarm (superadditive with independence)"
  },
  "P_4_main_effect_replications": {
    "retrieval_HOMO_NT_minus_HOMO_RET": {
      "mean": 0.0102,
      "lo": -0.069,
      "hi": 0.0867
    },
    "independence_HOMO_NT_minus_HET_NT": {
      "mean": 0.0662,
      "lo": 0.0048,
      "hi": 0.1364
    },
    "note": "estimation vs the Phase B' (retrieval ~ null) and Phase B (independence directional) single-factor findings"
  },
  "descriptives": {
    "round0_unanimity": {
      "HOMO_NT": 0.5618,
      "HOMO_RET": 0.5618,
      "HET_NT": 0.5169,
      "HET_RET": 0.4494
    },
    "diversity_by_round": {
      "HOMO_NT": [
        2.0449,
        1.382
      ],
      "HOMO_RET": [
        2.0899,
        1.573
      ],
      "HET_NT": [
        2.1798,
        1.5955
      ],
      "HET_RET": [
        2.3258,
        1.6292
      ]
    },
    "retrieval_turn_uptake": {
      "HOMO_NT": 0.0,
      "HOMO_RET": 0.9984,
      "HET_NT": 0.0,
      "HET_RET": 1.0
    },
    "debate_inflation": {
      "HOMO_NT": 0.0995,
      "HOMO_RET": 0.0963,
      "HET_NT": 0.0963,
      "HET_RET": 0.1508
    },
    "tokens": {
      "HOMO_NT": 469013,
      "HOMO_RET": 1171971,
      "HET_NT": 412177,
      "HET_RET": 995818
    }
  }
}