[
  {
    "benchmark": "gsm8k",
    "retention": 1.0,
    "token_savings_pct": -4.0,
    "baseline_accuracy": 0.85,
    "baseline_ci_95": [
      0.6396,
      0.9476
    ],
    "entroly_accuracy": 0.85,
    "entroly_ci_95": [
      0.6396,
      0.9476
    ],
    "baseline_avg_tokens": 283.0,
    "entroly_avg_tokens": 294.4,
    "samples": 20,
    "provenance": {
      "model": "gpt-4o-mini",
      "seed": 42,
      "budget": 50000,
      "timestamp_utc": "2026-05-26T04:39:35.149231+00:00",
      "git_sha": "ea7b7b2d43b4",
      "elapsed_s": 190.7,
      "audit_jsonl": "gsm8k_audit.jsonl",
      "audit_record_count": 20,
      "scoring": "bench.accuracy._check_answer (benchmark='gsm8k')",
      "compressor": "entroly.qccr.select (via bench.accuracy._compress_messages_modal mode='entroly')",
      "dataset_loader": "_load_gsm8k"
    }
  }
]