{
 "study": "Puppetmaster against Codex subagents, blind",
 "model": "gpt-6.1-sol, high effort, every worker",
 "puppetmaster": "puppetmaster-ai 1.49.0",
 "worlds": "The 18 crew worlds of the solo-line study (3 per crew per size). No new builds.",
 "sizes": [
  4,
  16,
  64
 ],
 "reps_per_crew_per_size": 3,
 "scenes_per_comparison": 15,
 "rule": "Pooled over all 9 comparisons: Puppetmaster preferred when its scene wins exceed Codex subagent scene wins, the exact two-sided binomial p < 0.05 on decided scenes, and at least 6 of 9 comparisons favor it. The mirror rule for Codex subagents. Per-size results are descriptive. A scene counts only when both judges agree in both orders. Ties are undecided.",
 "pooled": {
  "result": "Puppetmaster preferred",
  "puppetmaster": 39,
  "codex_subagents": 20,
  "ties": 76,
  "comparisons_favor_puppetmaster": 6,
  "comparisons_favor_codex_subagents": 2,
  "two_sided_p": 0.018337087779101986
 },
 "per_size": {
  "4": {
   "result": "no preference shown",
   "puppetmaster": 8,
   "codex_subagents": 17,
   "ties": 20,
   "comparisons_favor_puppetmaster": 1,
   "comparisons_favor_codex_subagents": 2,
   "two_sided_p": 0.10775214433670044
  },
  "16": {
   "result": "Puppetmaster preferred",
   "puppetmaster": 16,
   "codex_subagents": 2,
   "ties": 27,
   "comparisons_favor_puppetmaster": 3,
   "comparisons_favor_codex_subagents": 0,
   "two_sided_p": 0.001312255859375
  },
  "64": {
   "result": "Puppetmaster preferred",
   "puppetmaster": 15,
   "codex_subagents": 1,
   "ties": 29,
   "comparisons_favor_puppetmaster": 2,
   "comparisons_favor_codex_subagents": 0,
   "two_sided_p": 0.000518798828125
  }
 },
 "comparisons": [
  {
   "size": 4,
   "rep": 1,
   "puppetmaster": 3,
   "codex_subagents": 7,
   "ties": 5,
   "two_sided_p": 0.34375
  },
  {
   "size": 4,
   "rep": 2,
   "puppetmaster": 5,
   "codex_subagents": 1,
   "ties": 9,
   "two_sided_p": 0.21875
  },
  {
   "size": 4,
   "rep": 3,
   "puppetmaster": 0,
   "codex_subagents": 9,
   "ties": 6,
   "two_sided_p": 0.00390625
  },
  {
   "size": 16,
   "rep": 1,
   "puppetmaster": 5,
   "codex_subagents": 2,
   "ties": 8,
   "two_sided_p": 0.453125
  },
  {
   "size": 16,
   "rep": 2,
   "puppetmaster": 3,
   "codex_subagents": 0,
   "ties": 12,
   "two_sided_p": 0.25
  },
  {
   "size": 16,
   "rep": 3,
   "puppetmaster": 8,
   "codex_subagents": 0,
   "ties": 7,
   "two_sided_p": 0.0078125
  },
  {
   "size": 64,
   "rep": 1,
   "puppetmaster": 7,
   "codex_subagents": 0,
   "ties": 8,
   "two_sided_p": 0.015625
  },
  {
   "size": 64,
   "rep": 2,
   "puppetmaster": 1,
   "codex_subagents": 1,
   "ties": 13,
   "two_sided_p": 1.0
  },
  {
   "size": 64,
   "rep": 3,
   "puppetmaster": 7,
   "codex_subagents": 0,
   "ties": 8,
   "two_sided_p": 0.015625
  }
 ],
 "judge_cash_usd": 7.69,
 "judge_calls": 541,
 "disclosure": [
  "Every vote is fresh. No earlier vote counts.",
  "Six of the 64-region worlds come from the 64-region study.",
  "At 4 regions, Codex subagents led 17 scenes to 8. That did not pass the per-size test.",
  "A tie or 'no preference shown' does not prove the crews are equal.",
  "One world family, one model, three runs per crew per size. Scenes share worlds and judges."
 ],
 "source_report_sha256": "464e58693adecc2b6e5e6751760ea63b767e8c961d73500e544507a4c922bfa8"
}
