{"key": "mix2-1", "idea": "Keep two Gaussians (arb pin, retail truncated prior) across steps, propagate each as real.py does and weight them by silent-step truncation mass, quote from the moment-matched single Gaussian.", "settings_tried": [{"setting": "quote=0 floor=0.005", "n": 256, "mean_diff": 0.4261, "se": 0.035}, {"setting": "quote=1 bw=1.5 floor=0.005", "n": 256, "mean_diff": 0.3167, "se": 0.0446}, {"setting": "quote=0 floor=0.02", "n": 256, "mean_diff": 0.391, "se": 0.0334}, {"setting": "quote=1 bw=1.5 floor=0.02", "n": 256, "mean_diff": 0.3348, "se": 0.0434}, {"setting": "quote=0 floor=0.001", "n": 256, "mean_diff": 0.4382, "se": 0.0349}, {"setting": "quote=0 floor=0.01", "n": 256, "mean_diff": 0.4155, "se": 0.0348}, {"setting": "quote=0 floor=0.05", "n": 256, "mean_diff": 0.2702, "se": 0.0356}, {"setting": "quote=0 fsil=0 floor=0.005", "n": 256, "mean_diff": 0.3652, "se": 0.0344}, {"setting": "quote=1 bw=3 floor=0.005", "n": 256, "mean_diff": 0.0839, "se": 0.0505}, {"setting": "quote=0 floor=0.0001", "n": 256, "mean_diff": 0.4437, "se": 0.0349}, {"setting": "quote=0 floor=0.00001 (frozen)", "n": 256, "mean_diff": 0.4459, "se": 0.0348}, {"setting": "persist=0 (parity check)", "n": 256, "mean_diff": 0.0001, "se": 0.0}], "confirm_paired": {"n": 256, "mean_diff": 0.457, "se": 0.0333}, "holdout_vs_gauss": {"n": 2048, "mean_diff": 0.4279, "se": 0.0125, "sum": 876.2450376004415, "sumsq": 1026.874706887526, "worst5": [[11701, -2.306], [11062, -1.924], [10401, -1.836], [10256, -1.647], [10146, -1.631]]}, "holdout_vs_lattice": {"n": 2048, "mean_diff": -0.067, "se": 0.0089, "sum": -137.31126992345725, "sumsq": 344.1936418648131, "worst5": [[11311, -2.006], [10387, -1.987], [10256, -1.809], [10842, -1.542], [10940, -1.202]]}, "misread_rate": {"base": 0.1727641268317719, "mine": 0.16377756581321656}, "price_error_bps": {"base": 14.384445042796441, "mine": 13.747042143369834}, "gas_estimate": {"extra_slots": 3, "extra_exp_ln_sqrt_per_callback": "about +4 exp, +0 ln, +2 sqrt on a first-trade callback with two components, plus about 2 exp and 1 sqrt per silent step per extra component (cap 12); not measured", "rough_gas": "roughly +20,000 to +60,000 per first-trade callback; no EVM run"}, "code_zb64": "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", "code_sha256": "88104096d1b79016d6c04de9797216374b2f6c929e275754d8c8487408c56d45", "notes": "Tuned on 60000..60255 (base freshly run), confirmed once on 60256..60511 with floor 1e-5, then holdout 10000..12047 in four 512-seed runs, paired with ref files. Gain over Gaussian+qm is real (+0.43) but does not reach the lattice (-0.07 vs lattice_qm): misread only fell 17.3% to 16.4%. The pi^1.5 branch-quote blend (+0.32 tuning) was worse than the moment-matched quote; weighting by silent-step mass mattered (+0.07). Floor below 0.005 helped slightly, 0.05 hurt. Baseline numba runner was not used (numba absent). Next: a third component for the arb branch on stale steps, or quotes from the mixture in the DP."}