{
  "review_id": "2026-07-21-d48-hold1-training",
  "decision": "reject_training",
  "result_archive": "backtest-result-2026-07-21_06-39-39.zip",
  "result_sha256": "5ec42e651ebfe2a3b44330dedd00824966c66e17bbe97ba9fdafe0417bfa873c",
  "chart": "hold1_training_comparison.png",
  "visual_review": {
    "inspected_at": "2026-07-21T06:45:00Z",
    "status": "passed",
    "notes": [
      "The cumulative-profit panel is readable and shows Hold1 below baseline through most of the training window, not only in one isolated regime.",
      "The pair-delta panel confirms concentrated damage in SOL, XRP, ADA, and ETH despite improvements in EDGE, ZEC, and POWER.",
      "The exact +5m matched-trade scatter shows multiple baseline winners becoming full-stop Hold1 losses; delayed entry changes both price and downstream trailing/stop paths.",
      "The training-gate panel matches the result JSON: profit and PF fall, win rate falls, drawdown rises, and trade count remains above the 70% minimum.",
      "No unreadable panels, mislabeled series, or contradictions with summary.json were found."
    ]
  }
}