{"key":"split-1","seeds":{"from":60000,"to":60511,"n":512,"kind":"replica, tuning range"},"combos":[{"belief":"blind","phase":"blind","rival":"blind","call_cut":0.5,"n":512,"mean_diff":0.0,"se":0.0,"sum":0.0,"sumsq":0.0},{"belief":"flag","phase":"blind","rival":"blind","call_cut":0.5,"n":512,"mean_diff":2.3425,"se":0.0429,"sum":1199.344463,"sumsq":3290.259238},{"belief":"blind","phase":"flag","rival":"blind","call_cut":0.5,"n":512,"mean_diff":0.2193,"se":0.0176,"sum":112.282104,"sumsq":105.474545},{"belief":"blind","phase":"blind","rival":"flag","call_cut":0.5,"n":512,"mean_diff":-0.7898,"se":0.0191,"sum":-404.401537,"sumsq":415.313245},{"belief":"flag","phase":"flag","rival":"blind","call_cut":0.5,"n":512,"mean_diff":2.2841,"se":0.045,"sum":1169.464513,"sumsq":3201.382119},{"belief":"flag","phase":"blind","rival":"flag","call_cut":0.5,"n":512,"mean_diff":2.6477,"se":0.0398,"sum":1355.609838,"sumsq":4004.530573},{"belief":"blind","phase":"flag","rival":"flag","call_cut":0.5,"n":512,"mean_diff":-0.4941,"se":0.0198,"sum":-252.975809,"sumsq":227.225997},{"belief":"flag","phase":"flag","rival":"flag","call_cut":0.5,"n":512,"mean_diff":2.7152,"se":0.0419,"sum":1390.190263,"sumsq":4233.628629},{"belief":"blind","phase":"blind","rival":"blind","call_cut":0.9,"n":512,"mean_diff":0.3151,"se":0.0237,"sum":161.35244,"sumsq":197.694785},{"belief":"flag","phase":"blind","rival":"blind","call_cut":0.9,"n":512,"mean_diff":2.462,"se":0.041,"sum":1260.567959,"sumsq":3542.679578},{"belief":"blind","phase":"flag","rival":"blind","call_cut":0.9,"n":512,"mean_diff":0.4361,"se":0.0184,"sum":223.272342,"sumsq":185.533485},{"belief":"blind","phase":"blind","rival":"flag","call_cut":0.9,"n":512,"mean_diff":-0.7191,"se":0.0253,"sum":-368.202548,"sumsq":432.864684},{"belief":"flag","phase":"flag","rival":"blind","call_cut":0.9,"n":512,"mean_diff":2.6453,"se":0.0425,"sum":1354.410128,"sumsq":4055.240059},{"belief":"flag","phase":"blind","rival":"flag","call_cut":0.9,"n":512,"mean_diff":2.518,"se":0.0404,"sum":1289.21845,"sumsq":3673.529233},{"belief":"blind","phase":"flag","rival":"flag","call_cut":0.9,"n":512,"mean_diff":-0.4941,"se":0.0198,"sum":-252.975809,"sumsq":227.225997,"same_run_as_cut_0.5":true},{"belief":"flag","phase":"flag","rival":"flag","call_cut":0.9,"n":512,"mean_diff":2.7152,"se":0.0419,"sum":1390.190263,"sumsq":4233.628629,"same_run_as_cut_0.5":true}],"sanity":{"none_equals_lattice":true,"all_equals_oraclearb":true,"none_max_abs_diff":0.0,"all_max_abs_diff":0.0,"n":512,"cut09_phase_rival_flag_equals_cut05_16_seeds_max_abs_diff":0.0,"cpython_runpy_vs_pypy_8_seeds_max_abs_diff":1.6370904631912708e-11},"by_error":[{"part":"phase","error":"retail called arb","belief":"blind","call_cut":0.5,"per_1000_first_trades":98.37,"repairs":352146,"first_trades":3579926,"edge_per_repair":0.00033,"total":0.2296,"se":0.0164,"sum":117.556212,"sumsq":97.651658,"n":512},{"part":"phase","error":"arb called retail","belief":"blind","call_cut":0.5,"per_1000_first_trades":65.65,"repairs":234001,"first_trades":3564443,"edge_per_repair":-2e-05,"total":-0.0069,"se":0.0109,"sum":-3.549718,"sumsq":31.052357,"n":512},{"part":"rival","error":"retail called arb","belief":"blind","call_cut":0.5,"per_1000_first_trades":97.58,"repairs":348443,"first_trades":3570957,"edge_per_repair":-5e-05,"total":-0.0365,"se":0.014,"sum":-18.709929,"sumsq":51.79377,"n":512},{"part":"rival","error":"arb called retail","belief":"blind","call_cut":0.5,"per_1000_first_trades":63.83,"repairs":227649,"first_trades":3566658,"edge_per_repair":-0.00166,"total":-0.7393,"se":0.0159,"sum":-378.542367,"sumsq":346.000123,"n":512},{"part":"phase","error":"retail called arb","belief":"flag","call_cut":0.5,"per_1000_first_trades":71.88,"repairs":258243,"first_trades":3592851,"edge_per_repair":-0.00016,"total":-0.0829,"se":0.0097,"sum":-42.464251,"sumsq":27.936866,"n":512},{"part":"phase","error":"arb called retail","belief":"flag","call_cut":0.5,"per_1000_first_trades":49.29,"repairs":176146,"first_trades":3573421,"edge_per_repair":0.0001,"total":0.0342,"se":0.0104,"sum":17.485158,"sumsq":28.854646,"n":512},{"part":"rival","error":"retail called arb","belief":"flag","call_cut":0.5,"per_1000_first_trades":69.01,"repairs":247092,"first_trades":3580390,"edge_per_repair":0.00063,"total":0.305,"se":0.0139,"sum":156.158024,"sumsq":98.007586,"n":512},{"part":"rival","error":"arb called retail","belief":"flag","call_cut":0.5,"per_1000_first_trades":48.96,"repairs":175468,"first_trades":3584163,"edge_per_repair":0.0,"total":0.0002,"se":0.0003,"sum":0.089548,"sumsq":0.019434,"n":512},{"part":"phase","error":"retail called arb","belief":"blind","call_cut":0.9,"per_1000_first_trades":11.67,"repairs":42150,"first_trades":3610529,"edge_per_repair":0.00018,"total":0.015,"se":0.0063,"sum":7.681718,"sumsq":10.593828,"n":512},{"part":"phase","error":"arb called retail","belief":"blind","call_cut":0.9,"per_1000_first_trades":273.18,"repairs":977174,"first_trades":3576970,"edge_per_repair":5e-05,"total":0.0909,"se":0.0193,"sum":46.524866,"sumsq":101.602563,"n":512},{"part":"rival","error":"retail called arb","belief":"blind","call_cut":0.9,"per_1000_first_trades":11.48,"repairs":41409,"first_trades":3607453,"edge_per_repair":-5e-05,"total":-0.0039,"se":0.0027,"sum":-1.97956,"sumsq":1.920629,"n":512},{"part":"rival","error":"arb called retail","belief":"blind","call_cut":0.9,"per_1000_first_trades":280.68,"repairs":1012436,"first_trades":3607066,"edge_per_repair":-0.00052,"total":-1.0301,"se":0.0179,"sum":-527.427628,"sumsq":626.712703,"n":512}],"misread_blind":{"cut05":{"rca_per_1000":97.63,"acr_per_1000":64.46},"cut09":{"rca_per_1000":11.5,"acr_per_1000":278.73},"first_trades_per_seed_cut05":6968.1,"dead_belief_restarts_per_1000":0.0},"reading":"The belief holds the gain: the true flag in the belief alone is worth +2.34 of the +2.72 (+2.65 with the rival as well), the phase alone +0.22 (all of it from retail trades called arb), and the rival alone -0.79, because marking a trade as arb while the blind belief still treats it as retail leaves the rival spot inconsistent with the price estimate. Given the true belief, the phase and rival calls add only +0.37 together, of which +0.31 is the rival repair of retail called arb, so the parts a blind policy can move without a better price reading are worth about 0.3 to 0.45 (blind call at pi > 0.9: +0.32; with the true phase on top: +0.44), and the remaining ~2.3 is reachable only by reading the first trade well enough to pin or not pin the price, i.e. classifier accuracy feeding the belief.","code_zb64":"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","code_sha256":"84202020fa77c0027f8307b8e9465fb759f06a784a8b0ce1cb475030dad81e73","notes":"Replica seeds 60000..60511, n=512 per arm, every arm on the same seeds; no holdout seed used, no seed filled. Paired diffs are arm minus blind lattice (latt.py, qm=qm_k1e-5.json); se = sample sd/sqrt(n). Runs under PyPy 3.11 7.3.20 for speed; CPython run.py on 60000..60007 matches to 1.6e-11. Blind belief keeps latt.py's mixture (hard 0.999, recentre at pi>0.5) at every cut; call_cut moves only phase/rival calls. Cut-0.9 rows with phase and rival both on the flag are the cut-0.5 runs (identical by construction, checked on 16 seeds, diff 0). A step whose belief dies is pinned as arb in every arm, as latt.py does (0 such steps here). by_error: part given the true call only on one error type (rca: call arb = blind and flag; acr: blind or flag), others as stated; per_1000 = repaired calls per 1000 first trades in that run; edge_per_repair = summed edge diff / repaired calls; total = mean diff per seed vs the same-context base (belief blind: blind lattice at that cut; belief flag: belief-only arm). Blind misreads at cut 0.5: 97.6 rca + 64.5 acr per 1000 (16.2%); at 0.9: 11.5 + 278.7. code_sha256 is of the uncompressed tar."}