{
  "key": "fresh4-1",
  "idea": "Temper only the first-trade price-belief mixture: q=pi^(1/T)/(pi^(1/T)+(1-pi)^(1/T)); retain raw pi>0.9 for phase and rival updates. Frozen T=0.9.",
  "why": "Hypothesis: misspecified arb/retail likelihoods distort the recursive price posterior even when the cost-sensitive phase call is good. A temperature changes uncertainty carried into later callbacks and silent-step conditioning, rather than training a classifier, changing the lattice, or blending current quotes. Both softening and sharpening were tested; neither improved the base.",
  "settings_tried": [
    {
      "setting": {
        "temperature": 0.5,
        "threshold": 0.9
      },
      "n": 256,
      "mean_diff": -1.0084984741863616,
      "se": 0.05985098470278028
    },
    {
      "setting": {
        "temperature": 0.8,
        "threshold": 0.9
      },
      "n": 256,
      "mean_diff": -0.19296413595452577,
      "se": 0.02571443713701789
    },
    {
      "setting": {
        "temperature": 0.9,
        "threshold": 0.9
      },
      "n": 256,
      "mean_diff": -0.04265217454113346,
      "se": 0.01966864877821828
    },
    {
      "setting": {
        "temperature": 1.1,
        "threshold": 0.9
      },
      "n": 256,
      "mean_diff": -0.04275019437055616,
      "se": 0.01727075941353284
    },
    {
      "setting": {
        "temperature": 1.2,
        "threshold": 0.9
      },
      "n": 256,
      "mean_diff": -0.07686396748066926,
      "se": 0.021489530740328007
    },
    {
      "setting": {
        "temperature": 1.5,
        "threshold": 0.9
      },
      "n": 256,
      "mean_diff": -0.24731005871898004,
      "se": 0.028935347606281223
    },
    {
      "setting": {
        "temperature": 2.0,
        "threshold": 0.9
      },
      "n": 256,
      "mean_diff": -0.771081259812775,
      "se": 0.03740526651725613
    }
  ],
  "confirm_paired": {
    "n": 256,
    "mean_diff": -0.04292407054618819,
    "se": 0.018327806812821437
  },
  "holdout_vs_lattice_qm": {
    "n": 2048,
    "mean_diff": 0.30243290662156325,
    "se": 0.011608511702937188,
    "sum": 619.3825927609615,
    "sumsq": 752.2598000394573,
    "worst5": [
      [
        11487,
        -1.8821859694651266
      ],
      [
        11380,
        -1.7064562648552055
      ],
      [
        11651,
        -1.6136323984217142
      ],
      [
        10816,
        -1.479015283183685
      ],
      [
        10942,
        -1.3441758106857833
      ]
    ]
  },
  "holdout_vs_thr09": {
    "n": 2048,
    "mean_diff": -0.020459013188927955,
    "se": 0.007479915311384826,
    "sum": -41.90005901092445,
    "sumsq": 235.41032264851668,
    "worst5": [
      [
        11137,
        -1.7880563960362679
      ],
      [
        10131,
        -1.2356114844537274
      ],
      [
        10047,
        -0.9799073685521762
      ],
      [
        10776,
        -0.9615029801302057
      ],
      [
        10477,
        -0.9388744324845675
      ]
    ]
  },
  "misread_rate": {
    "base": 0.28883400504733875,
    "mine": 0.28384293745078704
  },
  "price_error_bps": {
    "base": 13.277152756808208,
    "mine": 13.335754069821293
  },
  "gas_estimate": {
    "evm_gas": null,
    "additional_persistent_state_slots": 0,
    "additional_pow_calls_per_ambiguous_first_trade": 2
  },
  "code_zb64": "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",
  "code_sha256": "227d4ea44378a4c6e360b8d40c6104adcb0ab9849b8edcebf704efc156085a6d",
  "notes": "Negative result; retain the thr=0.9 base. Seven temperatures tuned only on 60000..60255; T=0.9 selected by highest paired mean, confirmed once on 60256..60511, then frozen. Holdout scored once per arm in four 512-seed ranges covering 10000..12047 exactly. All scoring used run.py Blind with numba callback primitives derived from the supplied fast_thr.py; no hidden inputs entered policy state. Delivered standard-library policy matched accelerated rows exactly on 16 tuning runs; accelerated T=1 matched thrpol.py on two tuning runs. Misread rates are first-callback weighted; price errors are callback-weighted RMS bps, both over holdout. Gas was not measured: two powers per nondegenerate first-trade posterior, no extra persistent state; the existing lattice/table cost remains. SHA256 hashes uncompressed tar bytes. No engine seeds used."
}