{"key":"misread-size","family":"size","rows":{"train":40000,"test":213080},"lattice_misread_test":0.1623,"auc":{"filter_only":0.922,"mine_only":0.8711,"filter_plus_mine":0.922},"misread_test":{"filter_only":0.1624,"filter_plus_mine":0.1621},"by_dt":{"1":0.1512,"2":0.2091,"3-5":0.1813,"6+":0.1592},"by_size_quartile":[0.2465,0.2082,0.1417,0.0544],"top_feature":{"name":"fee_cross","definition":"Fee in force on the side the first trade crossed, -ln(1-fb) on a pool buy or -ln(1-fa) on a sell, in bps, set by our last quote before the trade. A higher fee means arb (the arb crosses wide quotes). Next best: d_over_fee = |ln(s/sp)|/fee_cross, AUC 0.773.","auc_alone":0.8256},"code_zb64":"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","code_sha256":"33208486506048e22ffd7b6f1b6c8d673c4fec3a3592990b3d704eec0f05eccc","notes":"Data: lattice filter (latt.py+qm_k1e-5) run blind, true flag recorded from the replica callback; reproduces ref/lattice_qm seeds 10000/10001 edge, first-trade and misread counts exactly. All first-of-step rows: train 3,567,655 (seeds 60000..60511), test 1,776,190 (61000..61255); fit on 40,000 rows drawn from an 8% sample, tested on a 12% sample (213,080). Logistic regression (IRLS, ridge). Result: size/timing adds nothing. Alone AUC 0.871 vs lattice pi 0.921; added to filter features misread 16.24%->16.21% (fixes 1677, breaks 1594 of 34,623 misreads). 2D table logit(pi) x each feature flips no cell's majority. Misreads are small trades: lattice misread 24.6% in the smallest own-fill quartile (<2.57 Y), 5.4% in the largest (>10.07 Y); by displacement 24.8% below 4.2 bps vs 4.5% above 23 bps. Small d is where an arb that barely crossed the band and a small retail fill give the same reading, and size is already in the filter (L_ret is the lognormal of d), so it is redundant. dt=2 is worst (20.9%): arb share falls to 26% after one silent step. Errors: 9.7% retail read as arb, 6.5% arb read as retail. Policy: don't spend gas on size/timing features. Treat small first trades (d < ~8 bps) as ambiguous: the phase choice (is_arb) is still a hard read even though the belief is mixed, so set phase-0/1 quotes so the next arb settles it.","extra":{"lattice_pi_auc_test":0.9213,"lattice_misread_test_sample":0.1625,"mine_only_misread":0.2054,"by_dt_share":{"1":0.7508,"2":0.1443,"3-5":0.0946,"6+":0.0103},"by_dt_arb_share":{"1":0.6469,"2":0.2583,"3-5":0.1889,"6+":0.1542},"size_quartile_edges_Y":[2.5708,5.3926,10.0732],"by_size_quartile_arb_share":[0.5824,0.5389,0.6009,0.4475],"d_quintile_edges_bps":[4.163,8.2592,13.7386,23.0505],"by_d_quintile_misread":[0.2476,0.2241,0.1787,0.1197,0.0447],"single_auc_mine":{"fee_cross":0.8256,"fee_other":0.8035,"ln_d_over_fee":0.7732,"d_over_fee":0.7731,"ln_d_x_ln_dt":0.666,"ln_since_pin":0.6617,"ln_dt":0.6604,"dt_eq1":0.658,"dt_eq2":0.5826,"prev_first_read_arb":0.5806,"ln_d":0.5596,"zret":0.5591,"ln_sizeY":0.5591,"ln_d_sq":0.5551,"zret_sq":0.5368,"prev_cnt":0.5274,"isbuy":0.5014,"prev_cnt0":0.5},"misread_types":{"arb_read_retail":0.065,"retail_read_arb":0.0975},"combo_fixes_breaks_of_misreads":[1677,1594,34623],"rows_all":{"train":3567655,"test":1776190},"code_ext":"code_zb64: base64(zlib(tar)) of mkdata.py, analyze.py, final.json; run python3 -I mkdata.py <solved> <seed0> <n> <out> <frac>, then analyze.py <train,...> <test,...> <out> 40000"}}