{
 "credit": "Taylor (gtmcknight), github.com/gtmcknight/amm-sim@4812a6382311624e1506aca131ca35c378f930d9 research/solved: the replica, Gaussian filter, fee rule, qm tables and the exact-arithmetic arb/retail tell are his. This item only snaps the fee wads his policy returns and counts his tell's labels. Fast replica: fastlean, https://api.imd.fun/artifacts/a88ff982eba7e078bc6ccb1555b90a211c00c9fd76ec90a62ab13a4074f2c3d0",
 "mech": "tell-coverage",
 "already_present": {
  "value": false,
  "where": "Not as a lattice snap. Read: leanbase.py (Policy.on_grid: label from lean_tell.classify, over/unsupported/neg -> miss, chance = 2^(binade(paid) - binade(reserve))), stalerival.py (fee(): wad = clamp(int(v*1e18), 0, 1e17), then `while 1.0-float(w)/WAD == 1.0-0.003: w+=128`; the staleness pull), tellpol.py (the on-grid test and the RHO*chance posterior), real.py (filter and table lookup: fees are a continuous interpolation). The only fee-lattice operation is stalerival.fee's step off Vanilla's gamma: one excluded gamma value. No file picks fee values for the tell. The measurement shows that excluded value is the centre of the only lattice effect found, so the variant tested is a general snap of the returned fee wad onto a lattice, which is not in the policy."
 },
 "definition": {
  "formula": "After LEANBASE has its two fees and stalerival.fee has made the wad w = clamp(int(f*1e18), 0, 1e17), and before the step off Vanilla's gamma: mode 1 (gamma lattice): g = 1 - w/1e18 in binary64 as the engine has it, G = g*2^53 (an integer), Gs = G rounded to a multiple of 2^par (rnd 0 nearest, +1 down = higher fee, -1 up = lower fee), w' = round((2^53 - Gs)*1e18/2^53), whose engine gamma is exactly Gs/2^53 (0 misses in 12,000,000 random lattice gammas). mode 2 (decimal lattice): w' = w rounded to a multiple of par wads. mode 3 (floor): w' = max(w, par). The snapped wads are returned and kept as the fees in force, for both sides, on every callback and at init. Chosen: mode 1, par 36, rnd +1: gamma on multiples of 2^-17, each fee raised by 0 to 0.0763 bps.",
  "parameters": {
   "mode": "1 gamma lattice, 2 decimal lattice, 3 floor",
   "par": "mode 1: bits t (lattice step 2^(t-53) in gamma); mode 2, 3: wads",
   "rnd": "0 nearest, +1 toward the higher fee, -1 toward the lower fee"
  },
  "state_kept": "none beyond LEANBASE: the two fee wads it already keeps are simply the snapped ones",
  "inputs_used": "only the two fees the policy just computed; nothing outside afterSwap inputs and own state"
 },
 "measurement": {
  "what_undecided_means": "(a) label over or unsupported (LEAN reads both as a miss); (b) an arbitrage trade off the grid; (c) a retail trade on the grid, the one case the tell cannot settle and leaves to the posterior. Hidden truth (which call site of the replica made the callback) is used for counting only, never by the policy.",
  "inloop_lean_64000_64127": {
   "first_own_trades": 895728,
   "retail": 410658,
   "label_over": 0,
   "label_unsupported": 0,
   "arb_off_grid": 0,
   "arb_on_grid": 485070,
   "retail_on_grid": 537,
   "retail_on_grid_per_1000": 1.3077,
   "reference_chance_per_1000": 1.099,
   "by_side": {
    "pool_bought_X": {
     "retail": 204829,
     "on_grid": 320,
     "sum_chance": 253.3
    },
    "pool_sold_X": {
     "retail": 205829,
     "on_grid": 217,
     "sum_chance": 198.0
    }
   },
   "fee_wad_0": {
    "retail": 41817,
    "on_grid": 57,
    "sum_chance": 46.7
   },
   "first_trades_with_fee_within_2000_wad_of_30bps": 0
  },
  "offline_by_fee_class": {
   "method": "the 410,658 retail first trades recorded above, each re-labelled by the same compiled classify under a hypothetical fee wad, amount paid held fixed (sell side: new reserveY recomputed as the engine would under that gamma); pos_per_1000 = retail trades labelled pos per 1000; se = sqrt(pos)/evaluations; fee wads drawn uniformly in 1..500 bps then snapped, or +-10% around a level",
   "classes": {
    "generic_1_500bps": {
     "fees": 48,
     "pos_per_1000": 1.2837,
     "se": 0.0081,
     "mean_gamma": 0.978024,
     "over": 0,
     "unsupported": 0
    },
    "generic_level_1bps": {
     "fees": 48,
     "pos_per_1000": 1.3101,
     "se": 0.0082,
     "mean_gamma": 0.9999,
     "over": 0,
     "unsupported": 0
    },
    "generic_level_10bps": {
     "fees": 48,
     "pos_per_1000": 1.3253,
     "se": 0.0082,
     "mean_gamma": 0.999005,
     "over": 0,
     "unsupported": 0
    },
    "generic_level_30bps": {
     "fees": 48,
     "pos_per_1000": 1.3112,
     "se": 0.0082,
     "mean_gamma": 0.997026,
     "over": 0,
     "unsupported": 0
    },
    "generic_level_100bps": {
     "fees": 48,
     "pos_per_1000": 1.2943,
     "se": 0.0081,
     "mean_gamma": 0.990025,
     "over": 0,
     "unsupported": 0
    },
    "generic_level_300bps": {
     "fees": 48,
     "pos_per_1000": 1.2766,
     "se": 0.008,
     "mean_gamma": 0.969929,
     "over": 0,
     "unsupported": 0
    },
    "generic_level_900bps": {
     "fees": 48,
     "pos_per_1000": 1.1958,
     "se": 0.0078,
     "mean_gamma": 0.91076,
     "over": 0,
     "unsupported": 0
    },
    "gamma_lattice_2^8": {
     "fees": 48,
     "pos_per_1000": 1.2745,
     "se": 0.008,
     "mean_gamma": 0.974338,
     "over": 0,
     "unsupported": 0
    },
    "gamma_lattice_2^16": {
     "fees": 48,
     "pos_per_1000": 1.2769,
     "se": 0.008,
     "mean_gamma": 0.97655,
     "over": 0,
     "unsupported": 0
    },
    "gamma_lattice_2^24": {
     "fees": 48,
     "pos_per_1000": 1.2848,
     "se": 0.0081,
     "mean_gamma": 0.974039,
     "over": 0,
     "unsupported": 0
    },
    "gamma_lattice_2^32": {
     "fees": 48,
     "pos_per_1000": 1.2732,
     "se": 0.008,
     "mean_gamma": 0.975602,
     "over": 0,
     "unsupported": 0
    },
    "gamma_lattice_2^36": {
     "fees": 47,
     "pos_per_1000": 1.2753,
     "se": 0.0081,
     "mean_gamma": 0.976186,
     "over": 0,
     "unsupported": 0
    },
    "gamma_lattice_2^40": {
     "fees": 46,
     "pos_per_1000": 1.2734,
     "se": 0.0082,
     "mean_gamma": 0.975761,
     "over": 0,
     "unsupported": 0
    },
    "gamma_lattice_2^44": {
     "fees": 22,
     "pos_per_1000": 1.2771,
     "se": 0.0119,
     "mean_gamma": 0.976207,
     "over": 0,
     "unsupported": 0
    },
    "gamma_lattice_2^46": {
     "fees": 7,
     "pos_per_1000": 1.2979,
     "se": 0.0212,
     "mean_gamma": 0.976562,
     "over": 0,
     "unsupported": 0
    },
    "decimal_lattice_1e+12_wad": {
     "fees": 48,
     "pos_per_1000": 1.2845,
     "se": 0.0081,
     "mean_gamma": 0.973393,
     "over": 0,
     "unsupported": 0
    },
    "decimal_lattice_1e+13_wad": {
     "fees": 48,
     "pos_per_1000": 1.2928,
     "se": 0.0081,
     "mean_gamma": 0.974153,
     "over": 0,
     "unsupported": 0
    },
    "decimal_lattice_1e+14_wad": {
     "fees": 48,
     "pos_per_1000": 1.2855,
     "se": 0.0081,
     "mean_gamma": 0.973342,
     "over": 0,
     "unsupported": 0
    },
    "decimal_lattice_1e+15_wad": {
     "fees": 29,
     "pos_per_1000": 1.2813,
     "se": 0.0104,
     "mean_gamma": 0.977793,
     "over": 0,
     "unsupported": 0
    },
    "fee_0": {
     "fees": 1,
     "pos_per_1000": 1.3978,
     "se": 0.0583,
     "mean_gamma": 1.0,
     "over": 0,
     "unsupported": 0
    },
    "gamma_1-2^-4": {
     "fees": 1,
     "pos_per_1000": 1.1713,
     "se": 0.0534,
     "mean_gamma": 0.9375,
     "over": 0,
     "unsupported": 0
    },
    "gamma_1-2^-5": {
     "fees": 1,
     "pos_per_1000": 1.3539,
     "se": 0.0574,
     "mean_gamma": 0.96875,
     "over": 0,
     "unsupported": 0
    },
    "gamma_1-2^-6": {
     "fees": 1,
     "pos_per_1000": 1.3101,
     "se": 0.0565,
     "mean_gamma": 0.984375,
     "over": 0,
     "unsupported": 0
    },
    "gamma_1-2^-7": {
     "fees": 1,
     "pos_per_1000": 1.2468,
     "se": 0.0551,
     "mean_gamma": 0.992188,
     "over": 0,
     "unsupported": 0
    },
    "gamma_1-2^-8": {
     "fees": 1,
     "pos_per_1000": 1.2419,
     "se": 0.055,
     "mean_gamma": 0.996094,
     "over": 0,
     "unsupported": 0
    },
    "gamma_1-2^-9": {
     "fees": 1,
     "pos_per_1000": 1.2833,
     "se": 0.0559,
     "mean_gamma": 0.998047,
     "over": 0,
     "unsupported": 0
    },
    "gamma_1-2^-10": {
     "fees": 1,
     "pos_per_1000": 1.276,
     "se": 0.0557,
     "mean_gamma": 0.999023,
     "over": 0,
     "unsupported": 0
    },
    "gamma_1-2^-12": {
     "fees": 1,
     "pos_per_1000": 1.2443,
     "se": 0.055,
     "mean_gamma": 0.999756,
     "over": 0,
     "unsupported": 0
    }
   }
  },
  "closed_loop_by_setting": "settings_tried[].retail_on_grid_per_1000 (LEAN without snap on the same 1,024 seeds: 1.307; over+unsupported 0; arb off grid 0)",
  "resonance_at_vanilla_gamma": {
   "run": "decimal lattice 1e14 wad (1 bp) nearest, seeds 64000..64127, retail first trades grouped by the traded-side fee wad in force",
   "fee_wad_3000000000000128": {
    "retail": 4104,
    "on_grid": 1836,
    "pool_bought_X": [
     2086,
     957
    ],
    "pool_sold_X": [
     2018,
     879
    ]
   },
   "fee_wad_0": {
    "retail": 48404,
    "on_grid": 64
   },
   "all_other_fee_wads": {
    "retail": 357943,
    "on_grid": 456
   },
   "total": {
    "retail": 410451,
    "on_grid": 2356
   },
   "reading": "3e15+128 is where stalerival.fee lands a 30.00 bps fee after its step off Vanilla's gamma, one gamma ulp away. There 45% of retail first trades are on the arbitrage grid, against 0.13% elsewhere. My explanation, derived from the router lines of the replica and not tested further: an interior split pays u1 = numerator/(g_own + r*g_vanilla), the numerator being a difference of reserve-sized doubles and so a whole number N of reserve ulps; with g_own within a few ulps of g_vanilla and r = 1 to rounding the divisor is 2*g_own, and u1 = (N/2)*ulp/g_own is on the arbitrage grid whenever N is even.",
   "consequence": "The 128-wad step-off removes the exact collision but not the resonance. LEAN's own fees are continuous and never sit there (0 first trades within 2000 wad of 30 bps in the baseline), so nothing is lost today; any variant that emits round fees (a 1 bp or 0.1 bp lattice, a clamp or a default at 30 bps) walks into it: decimal 1e13 has 1.80 and decimal 1e14 5.74 retail on-grid per 1000 against 1.31."
  },
  "findings": [
   "Labels over and unsupported: 0 of 895,728 first trades on 64000..64127 and 1 of 14,399,902 on 10000..12047, with or without the snap. Arbitrage trades off the grid: 0 everywhere.",
   "Retail on the grid: 1.31 per 1000 retail first trades. Offline it is the same on every lattice tried (1.27 to 1.30 per 1000, se 0.008 to 0.02, against 1.28 for unsnapped fees over the same 1..500 bps) and moves only with gamma itself (1.31 at 1 bps, 1.20 at 900 bps: the ratio of the gammas). Fee 0 (gamma = 1): 1.40 se 0.06, not distinguishable.",
   "The reference chance 2^(binade(paid)-binade(reserve)) gives 1.10 per 1000; observed is 1.19 times that (537 against 451.3). Not a fee-lattice matter; not pursued.",
   "So there is no lattice on which the tell decides more often. There is one fee value on which it decides far less often, Vanilla's gamma and its neighbours, and LEAN already stays off it by having continuous fees.",
   "Ceiling: Taylor's tell README puts the grid tell at 0.01 short of the true flag on his replica, so no coverage gain could reach +0.03."
  ]
 },
 "settings_tried": [
  {
   "name": "gamma_2^16_nearest",
   "parameters": {
    "mode": 1,
    "par": 16,
    "rnd": 0
   },
   "n": 1024,
   "mean_diff": -0.001072,
   "se": 0.002454,
   "dropped_early": false,
   "retail_on_grid_per_1000": 1.3165,
   "over_plus_unsupported": 0,
   "arb_off_grid": 0
  },
  {
   "name": "gamma_2^24_nearest",
   "parameters": {
    "mode": 1,
    "par": 24,
    "rnd": 0
   },
   "n": 1024,
   "mean_diff": 0.001988,
   "se": 0.003065,
   "dropped_early": false,
   "retail_on_grid_per_1000": 1.3228,
   "over_plus_unsupported": 0,
   "arb_off_grid": 0
  },
  {
   "name": "gamma_2^32_nearest",
   "parameters": {
    "mode": 1,
    "par": 32,
    "rnd": 0
   },
   "n": 1024,
   "mean_diff": -0.002671,
   "se": 0.002713,
   "dropped_early": false,
   "retail_on_grid_per_1000": 1.2977,
   "over_plus_unsupported": 0,
   "arb_off_grid": 0
  },
  {
   "name": "gamma_2^36_nearest",
   "parameters": {
    "mode": 1,
    "par": 36,
    "rnd": 0
   },
   "n": 1024,
   "mean_diff": 0.000937,
   "se": 0.002765,
   "dropped_early": false,
   "retail_on_grid_per_1000": 1.2761,
   "over_plus_unsupported": 0,
   "arb_off_grid": 0
  },
  {
   "name": "gamma_2^40_nearest",
   "parameters": {
    "mode": 1,
    "par": 40,
    "rnd": 0
   },
   "n": 640,
   "mean_diff": -0.031824,
   "se": 0.00801,
   "dropped_early": true,
   "retail_on_grid_per_1000": 1.2745,
   "over_plus_unsupported": 0,
   "arb_off_grid": 0
  },
  {
   "name": "gamma_2^36_higher_fee",
   "parameters": {
    "mode": 1,
    "par": 36,
    "rnd": 1
   },
   "n": 1024,
   "mean_diff": 0.003304,
   "se": 0.003341,
   "dropped_early": false,
   "retail_on_grid_per_1000": 1.3219,
   "over_plus_unsupported": 0,
   "arb_off_grid": 0
  },
  {
   "name": "gamma_2^40_higher_fee",
   "parameters": {
    "mode": 1,
    "par": 40,
    "rnd": 1
   },
   "n": 384,
   "mean_diff": -0.050553,
   "se": 0.016748,
   "dropped_early": true,
   "retail_on_grid_per_1000": 1.2759,
   "over_plus_unsupported": 0,
   "arb_off_grid": 0
  },
  {
   "name": "gamma_2^40_lower_fee",
   "parameters": {
    "mode": 1,
    "par": 40,
    "rnd": -1
   },
   "n": 128,
   "mean_diff": -0.161156,
   "se": 0.029687,
   "dropped_early": true,
   "retail_on_grid_per_1000": 1.2774,
   "over_plus_unsupported": 0,
   "arb_off_grid": 0
  },
  {
   "name": "decimal_1e12_nearest",
   "parameters": {
    "mode": 2,
    "par": 1000000000000.0,
    "rnd": 0
   },
   "n": 1024,
   "mean_diff": 0.000357,
   "se": 0.002647,
   "dropped_early": false,
   "retail_on_grid_per_1000": 1.3367,
   "over_plus_unsupported": 0,
   "arb_off_grid": 0
  },
  {
   "name": "decimal_1e13_nearest",
   "parameters": {
    "mode": 2,
    "par": 10000000000000.0,
    "rnd": 0
   },
   "n": 1024,
   "mean_diff": -0.0093,
   "se": 0.003566,
   "dropped_early": false,
   "retail_on_grid_per_1000": 1.7974,
   "over_plus_unsupported": 0,
   "arb_off_grid": 0
  },
  {
   "name": "decimal_1e14_nearest",
   "parameters": {
    "mode": 2,
    "par": 100000000000000.0,
    "rnd": 0
   },
   "n": 128,
   "mean_diff": -0.095486,
   "se": 0.022402,
   "dropped_early": true,
   "retail_on_grid_per_1000": 5.74,
   "over_plus_unsupported": 0,
   "arb_off_grid": 0
  },
  {
   "name": "floor_128_wad",
   "parameters": {
    "mode": 3,
    "par": 128,
    "rnd": 0
   },
   "n": 1024,
   "mean_diff": -0.000392,
   "se": 0.000773,
   "dropped_early": false,
   "retail_on_grid_per_1000": 1.2731,
   "over_plus_unsupported": 0,
   "arb_off_grid": 0
  }
 ],
 "selection": {
  "seeds": "64000..65023 in rounds of 128, each setting paired with LEAN (a0=0.35 ainf=0.35) on the same seeds; dropped once mean_diff + 3*se < 0; se = sd of paired differences (n-1) / sqrt(n)",
  "lean_mean_edge": 539.018328,
  "survivors_at_1024": [
   "gamma_2^16_nearest",
   "gamma_2^24_nearest",
   "gamma_2^32_nearest",
   "gamma_2^36_nearest",
   "gamma_2^36_higher_fee",
   "decimal_1e12_nearest",
   "decimal_1e13_nearest",
   "floor_128_wad"
  ],
  "age_minus_lean_64000_64127": 0.038654
 },
 "chosen": {
  "name": "gamma_2^36_higher_fee",
  "parameters": {
   "mode": 1,
   "par": 36,
   "rnd": 1
  },
  "why": "highest mean_diff among the 8 settings alive after 1,024 seeds (+0.0033 se 0.0033); the eight are within noise of each other and of zero, so the choice is a tie-break, not a finding"
 },
 "report": {
  "vs_lean": {
   "n": 2048,
   "mean_diff": 0.006329,
   "se": 0.002453,
   "seeds_positive": 1076,
   "seeds_negative": 972,
   "base_mean_edge": 536.603391,
   "mech_mean_edge": 536.60972,
   "retail_on_grid_per_1000_base": 1.322,
   "retail_on_grid_per_1000_mech": 1.3166,
   "over_plus_unsupported_base": 1,
   "over_plus_unsupported_mech": 1,
   "first_trades_base": 14399902
  },
  "on_age": {
   "n": 2048,
   "mean_diff": 0.004592,
   "se": 0.002451,
   "seeds_positive": 1094,
   "seeds_negative": 954,
   "base_mean_edge": 536.6231,
   "mech_mean_edge": 536.627692,
   "retail_on_grid_per_1000_base": 1.2879,
   "retail_on_grid_per_1000_mech": 1.3124,
   "over_plus_unsupported_base": 1,
   "over_plus_unsupported_mech": 1,
   "first_trades_base": 14407398
  },
  "seeds": "10000..12047, replica, each run once; vs_lean: a0=0.35 ainf=0.35 with the snap minus without; on_age: a0=0 ainf=0.7 tau=1 with the snap minus without"
 },
 "promote": false,
 "verdict": "Do not promote. +0.0063 se 0.0025 on LEAN and +0.0046 se 0.0025 on AGE are both far below +0.03. The tell's coverage did not move (retail on grid per 1000: 1.322 -> 1.317 on LEAN, 1.288 -> 1.312 on AGE; undecided labels 1 -> 1). The chosen snap raises each nonzero fee by up to 0.0763 bps (about 0.038 bps on average if fees fall evenly within a step; not measured), so the small positive differences, if real, could as well be a fee-level effect; that was not separated by a control.",
 "contract_cost": {
  "basis": "estimate from the arithmetic, not compiled or measured; contract today about 208,000 of 245,000 peak gas, 22,440 of 24,576 runtime bytes, 47,230 of 50,000 source characters",
  "storage_words": 0,
  "gas": "about 2,000 to 2,600 per afterSwap on every path, the peak path included (peak about 210,000 to 211,000 of 245,000): per fee one binary64-exact gamma mantissa from the wad (two round-to-nearest steps, about 500 to 700 gas if the routine the tell already has is reused), a mask, the inverse mulDiv (about 100) and one re-check of the gamma (another 500 to 700) to stay bit for bit with the Python; times two fees. No SSTORE added.",
  "runtime_bytes": "about 250 to 400 (to about 22,700 to 22,850 of 24,576)",
  "characters": "about 600 to 900 (to about 47,850 to 48,150 of 50,000)",
  "fits": true
 },
 "checks": {
  "snap_off_is_fastlean": "patched file with snap=(0,0,0): 128 of 128 seeds (64000..64063, 10000..10063) bit-identical to the unpatched replica, mean 527.8724874783702, the mean_edge_fast its artifact reports; AGE parameters: 32 of 32 bit-identical",
  "patch_applies": "patch -p1 on a fresh unpack of code_zb64 gives sha256 10459468530922871b140a6373f0024ad136960ffc8fc8584942fd1be336b2f0; unpatched fastlean.py sha256 d9c758bb280fe24e09c039496f84ab0804173ff70cec24c332f8683f28efa582 = the artifact's source_sha256",
  "labels": "classify and the tell are untouched by the patch; the coverage counters only read",
  "snap_lattice": "mode 1 par 36 rnd +1 on 200,000 random wads in 0..1e17: result on the lattice and never a lower fee in all but 1 (not investigated; the code falls back to the lower-fee lattice point when the higher one would exceed the 10% cap)",
  "research_files": "fetch.sh at commit 66604b3426d8e97a73e2c39940b9b4dece51cc73: 34 files, 0 sha256 mismatched; leanbase.py, stalerival.py, lean_tell.py read from artifact efb8e653f8fd061cecf3e92b186470c9c96c5188523364b60ed101e9672869aa"
 },
 "environment": {
  "python": "3.12.3",
  "numba": "0.61.2",
  "numpy": "2.2.6",
  "llvmlite": "0.44.0",
  "platform": "Linux x86_64, wheels unpacked from PyPI (no pip on the machine)"
 },
 "patch_zb64": "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",
 "patch_note": "unified diff against fastlean/fastlean.py of the replica artifact (patch -p1). Adds snap_wad, fee_round(v, mode, par, rnd), load(..., snap=(mode, par, rnd)), and run_batch(..., cover=True, nrec=N) -> (edges, counters, records): diagnostic counters that use the hidden truth and are never read by the policy. Chosen setting: fastlean.load(amm, snap=(1, 36, 1)); on AGE add a0=0, ainf=0.7, tau=1.",
 "not_done": [
  "Replica only: no Solidity, no gas or size measurement (contract_cost is an estimate), no engine run, no gate P/Q/T/F/G/E, nothing submitted to the leaderboard.",
  "No control separating the lattice from the small fee raise (up to 0.0763 bps) of the chosen setting (the 12-setting budget was spent); the report differences are 2.6 and 1.9 se and were not re-run on other seeds.",
  "The resonance at Vanilla's gamma was seen at one fee wad (3e15+128) in one 128-seed run of a setting that was dropped; its width in gamma ulps and the rate at the exact collision were not measured, and the router explanation is a derivation, not a test. A wider step-off was not tested: with continuous fees it would change no callback in the baseline.",
  "The 19% excess of retail on-grid trades over the reference chance was not explained or tested as a change to the posterior.",
  "Offline re-labelling holds the paid amount fixed, so it cannot see effects that come through the router (it misses the resonance, which only the closed-loop counts show).",
  "The kit's setup_fast.sh, ammfast, score.py and engine_eval.py were not used; the replica artifact's own run lines were followed. On the QLX5 contract itself (rival split test) nothing was run: the replica is LEAN/AGE.",
  "Seeds 64000..64127 were used both for the coverage measurement and as the first selection round."
 ]
}