{
 "analysis": "exact two-sided McNemar on paired per-task outcomes; Wilson 95% intervals; paired bootstrap for the difference",
 "baseline_commit": "f4449aa7913623bc034fa4434ca96c2a2919a889",
 "best_fixed_path_selected_on_train": "m15_direct",
 "hypotheses": {
  "H1": "router accuracy on test > accuracy of the best fixed single path chosen on train (paired, exact McNemar p<0.05)",
  "H2": "at least one of c02..c09 (multi-agent, 0.5B) beats c01_single on test (paired, exact McNemar p<0.05)",
  "H3": "the direction of H1 repeats on the replication split run in a separate process"
 },
 "policy_sha256": "b98b26485b018bb6b3c2a9f3225f310e8f3d93e5dbb4fbd5d341d4e7264172fb",
 "question": "Does a routing policy fitted on train beat the best FIXED execution path on unseen test tasks, and does any multi-agent architecture beat one agent at this model scale?",
 "schema": "cain2027.e80.preregistration.v1",
 "stopping": "no peeking: test rows are written once; no re-run with altered prompts counts as this experiment",
 "test_or_replication_stage_files_existing_when_written": [],
 "test_set_sha256": "357de046001a4bb242821d4097d396ad899b7d58dbce400b3887ad419a90f6be",
 "written_utc": 1794775662.477
}