{
 "credit": "Taylor (gtmcknight), github.com/gtmcknight/amm-sim@4812a6382311624e1506aca131ca35c378f930d9 research/solved",
 "table": 3,
 "levels": [
  1,
  2
 ],
 "gain_vs_base": -0.000248,
 "se": 0.004883,
 "n": 1024,
 "seeds": [
  64000,
  65023
 ],
 "valid": {
  "gain": 0.007227,
  "se": 0.002336,
  "n": 4096
 },
 "nonzero": true,
 "advance": true,
 "source_sha256": "e183e5bc3fb794b7067703b10a7da9920267aa5f4caa968e16ca101b02338f40",
 "delta_zb64": "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",
 "delta_sha256": "fce9804278b5262b3574c04d862f29bf4184d6fe942f5eac0f32d089861bab0c",
 "bands": [
  {
   "level": 1,
   "axis": "row",
   "knots": [
    0,
    1,
    2,
    3,
    4,
    5,
    6,
    7,
    8,
    9,
    10,
    11,
    12
   ],
   "es_mean_bps": -0.082
  },
  {
   "level": 1,
   "axis": "row",
   "knots": [
    13,
    14
   ],
   "es_mean_bps": 0.029
  },
  {
   "level": 1,
   "axis": "row",
   "knots": [
    15,
    16
   ],
   "es_mean_bps": 0.591
  },
  {
   "level": 1,
   "axis": "row",
   "knots": [
    17,
    18
   ],
   "es_mean_bps": 0.127
  },
  {
   "level": 1,
   "axis": "row",
   "knots": [
    19,
    20
   ],
   "es_mean_bps": -0.154
  },
  {
   "level": 1,
   "axis": "row",
   "knots": [
    21,
    22,
    23,
    24,
    25,
    26,
    27,
    28,
    29,
    30,
    31,
    32,
    33,
    34,
    35
   ],
   "es_mean_bps": -0.405
  },
  {
   "level": 1,
   "axis": "column",
   "knots": [
    0,
    1,
    2,
    3
   ],
   "es_mean_bps": 0.318
  },
  {
   "level": 1,
   "axis": "column",
   "knots": [
    4,
    5,
    6,
    7
   ],
   "es_mean_bps": 0.926
  },
  {
   "level": 1,
   "axis": "column",
   "knots": [
    8,
    9,
    10
   ],
   "es_mean_bps": 0.209
  },
  {
   "level": 1,
   "axis": "column",
   "knots": [
    11,
    12
   ],
   "es_mean_bps": 0.361
  },
  {
   "level": 1,
   "axis": "column",
   "knots": [
    13,
    14,
    15,
    16,
    17,
    18,
    19,
    20,
    21,
    22,
    23,
    24,
    25
   ],
   "es_mean_bps": 0.264
  },
  {
   "level": 2,
   "axis": "row",
   "knots": [
    0,
    1,
    2,
    3,
    4,
    5,
    6,
    7,
    8,
    9,
    10,
    11,
    12
   ],
   "es_mean_bps": -0.011
  },
  {
   "level": 2,
   "axis": "row",
   "knots": [
    13,
    14
   ],
   "es_mean_bps": 0.243
  },
  {
   "level": 2,
   "axis": "row",
   "knots": [
    15,
    16
   ],
   "es_mean_bps": 0.107
  },
  {
   "level": 2,
   "axis": "row",
   "knots": [
    17,
    18
   ],
   "es_mean_bps": 0.637
  },
  {
   "level": 2,
   "axis": "row",
   "knots": [
    19,
    20
   ],
   "es_mean_bps": -0.497
  },
  {
   "level": 2,
   "axis": "row",
   "knots": [
    21,
    22,
    23,
    24,
    25,
    26,
    27,
    28,
    29,
    30,
    31,
    32,
    33,
    34,
    35
   ],
   "es_mean_bps": 0.563
  },
  {
   "level": 2,
   "axis": "column",
   "knots": [
    0,
    1,
    2,
    3,
    4
   ],
   "es_mean_bps": -0.513
  },
  {
   "level": 2,
   "axis": "column",
   "knots": [
    5,
    6,
    7
   ],
   "es_mean_bps": 0.273
  },
  {
   "level": 2,
   "axis": "column",
   "knots": [
    8,
    9,
    10
   ],
   "es_mean_bps": -0.168
  },
  {
   "level": 2,
   "axis": "column",
   "knots": [
    11,
    12
   ],
   "es_mean_bps": -0.213
  },
  {
   "level": 2,
   "axis": "column",
   "knots": [
    13,
    14,
    15,
    16,
    17,
    18,
    19,
    20,
    21,
    22,
    23,
    24,
    25
   ],
   "es_mean_bps": -0.323
  }
 ],
 "candidates": [
  {
   "name": "es_mean",
   "valid_gain": 0.007227,
   "se": 0.002336
  },
  {
   "name": "es_mean_x0.5",
   "valid_gain": 0.003799,
   "se": 0.001701
  },
  {
   "name": "es_mean_x1.5",
   "valid_gain": 0.00656,
   "se": 0.002666
  }
 ],
 "simulations": 121152,
 "limitations": [
  "Replica only (engine-faithful QLX5 replica of Taylor (gtmcknight)'s solved policy); no engine run, no Solidity, no gas; report seeds 10000..12047 were never run.",
  "The kept correction (ES mean, rounded to the 1.0 bps lattice: 356 nodes changed, {\"1,+1\": 180, \"1,+2\": 8, \"2,-1\": 111, \"2,+1\": 57} as {(level, units): nodes}) is +0.007227 +- 0.002336 on VALID (n 4096) but -0.000248 +- 0.004883 on CHOOSE (n 1024, scored once): not distinguishable from zero on CHOOSE and far below the +0.03 promotion bar. The VALID figure is the best of three candidates and so is biased upward.",
  "Simulations 121152 = controls 192 (original, overlay without delta, overlay with an empty delta file on 64000..64063, byte-equal CSV) + profile 128 (64000..64127) + ES 102400 + VALID 16384 (unchanged QLX5 and 3 candidates on 300000..304095) + CHOOSE 2048 (unchanged QLX5 and the kept one on 64000..65023).",
  "ES as run, fixed before scoring: 16 generations, 12 mirrored pairs, a new 256-seed TRAIN slice each generation (200000..204095 used of 200000..263999), sigma 0.30 -> 0.15 bps geometric, isotropic over 22 band offsets, theta in integer millibps; step = eta * sum(d_i * e_i)/(12 sigma^2) with eta = 10.54 bps^2 per edge unit set from generation-0 noise so a noise-only step is 0.05 bps per offset, steps clipped to 0.15 bps and theta to 2 bps. The ES count includes 4096 unchanged-QLX5 runs (one per slice seed) used only to monitor the pair level.",
  "Monitor only, not a score: mean over pairs of (f+ + f-)/2 minus unchanged QLX5 per generation on its TRAIN slice (each about +-0.005): [-0.0095, 0.0023, -0.0042, 0.009, 0.0003, -0.0065, 0.0178, 0.005, 0.0017, 0.0046, 0.0022, 0.0095, 0.0039, 0.0026, 0.0081, 0.01].",
  "Level 2 carries about 2% of level 1's visit weight on the profile seeds (4580 against 233879), so its offsets are mostly a noise random walk under an isotropic sigma; a node rounds to a unit only where row offset + column offset reaches 0.5 bps.",
  "CHOOSE contains the profile seeds 64000..64127 and the control seeds 64000..64063; the bands use visit weights only, never edges. Rows and columns with zero visit weight sit in the last band of their axis.",
  "score.py paired() was used for VALID and CHOOSE (candidate minus unchanged QLX5 on the same seeds, se = sample sd / sqrt(n)); no early stop was applied because this item has no bar. Seed-to-seed sd of the paired difference on CHOOSE is about 0.16.",
  "delta_sha256 hashes the exact delta file bytes, which begin with one # credit line to Taylor (gtmcknight); zero records are omitted. All three candidates passed qpl.apply(quantize=True) and qpl.encode; column differences in levels 1 and 2 of table 3 stay within int8.",
  "The harness README (delta format, lattice scales, qpl and score usage) is inside implementation_zb64 of the harness artifact, so it cannot be quoted verbatim in sources.json; implementation_sha256 4d4894474104eac9148e52b18d679b994ce73c6a889bafa18169c8dacca22855 and QLX5 source sha256 were recomputed here and matched."
 ]
}