{
 "approach": "Measure only, no code changed: frozen candidate stock-entry-gate against current best B (combine-cash), paired on every seed of block 252000..252999. Result on this block: candidate is lower by 0.863 (se 0.450), about 1.9 standard errors below zero; no evidence of gain here.",
 "tried": {
  "hash_check": "Downloaded both artifact JSONs, took the `code` field, SHA-256 of UTF-8 bytes matched both expected hashes before any simulation; the worker re-checked them at start.",
  "code_diff": "diff of B and candidate is the single q1-bank gate line.",
  "simulator": "danrobinson/prediction-market-challenge at b244530ee2a6247b5f930830eeca9dd950da0e7e, unmodified, CPython 3.12.3, default ChallengeConfig() and ParameterVariance(); load_strategy_factory then run_batch(factory, base_config=ChallengeConfig(), variance=ParameterVariance(), n_simulations=1, seed_start=seed, workers=1) per seed; no --workers, no --sandbox.",
  "timing": "First 20 seeds serially in 30.7 s wall (0.76 s per simulation), then the other 980 seeds in 5 OS processes on disjoint ranges of 196 seeds; the 20 timing seeds are retained in the block.",
  "checks": [
   "1000 distinct seeds, exactly 252000..252999, each with both strategies",
   "every returned result seed equals the requested seed",
   "no simulation raised or returned failed=true",
   "total_edge = retail_edge + arb_edge within 4.5e-9 on all 2000 runs",
   "seeds 252000, 252250, 252500, 252999 rerun in new processes with a different PYTHONHASHSEED reproduced all six edges exactly"
  ],
  "tail_note": "The difference is zero on 918 of 1000 seeds and heavy-tailed on the rest: worst seed 252003 at -285.63, best seed 252443 at 39.58; the worst seed alone is 40% of sum_d2. All seeds are kept in the figures.",
  "ledger_sha256": "2183630c723275c405978c606f3c6e2ce122a7d888b1a521fff520615fb727ba",
  "ledger_format": "Not delivered; sha256 of UTF-8 CSV lines in seed order: seed,cand_total,best_total,cand_retail,best_retail,cand_arb,best_arb using Python repr of floats, each ending with a newline."
 },
 "next": "Pool this block with the other segment blocks using sum_d, sum_d2 and n. This one block promotes nothing and the reserved finalist seeds were not touched. Because a few seeds dominate the variance, check the worst seeds in the pooled result before judging the sign.",
 "code_sha256": "a709c7e85c7dc5200cba83d854cc7ecc9cee8a70915ca79798504da648e9ebed",
 "incumbent_code_sha256": "baef955ec10728839f7c3357016ec9eb0550e7ec689bbd5171b1cdd5c9e37273",
 "tuning": null,
 "paired": {
  "seeds": "252000..252999",
  "n": 1000,
  "complete": true,
  "mean_diff": -0.8625623452730999,
  "se": 0.4497737798455021,
  "sum_d": -862.5623452730999,
  "sum_d2": 202838.1703829567,
  "candidate_mean": 97.1492456541299,
  "incumbent_mean": 98.011807999403,
  "retail_diff": -0.3407655129722998,
  "retail_se": 0.14602839676716592,
  "arb_diff": -0.5217968323007,
  "arb_se": 0.33078201487418385,
  "nonzero_diff_seeds": 82,
  "failures": []
 },
 "seconds_per_simulation": 0.711
}