{"item":"e07-tell-tables/visit-refit","status":"complete","credit":"Taylor (gtmcknight): replica2.py, the Gaussian filter, the fee rule and the qm tables (github.com/gtmcknight/amm-sim @4812a6382311624e1506aca131ca35c378f930d9, research/solved). TwoComp (TC), the QMS1 blob/decoder (qm-spline), qt1's QMRef and the wad tell + engine-order port (e06-tell-port/o1-b) are IdentityMD items built on Taylor's work. This item only replaces QMS1 coefficients.","summary":"Replica-only. BASE = TwoComp + o1-b's Bayes wad tell (floor 1e-3, cap 64) with the original QMS1 blob. C = the same policy with one refitted QMS1 blob (coefficient replacement only, 10167 bytes, knots/layout/header byte-identical). Report seeds 10000..12047, n=2048, paired: C-BASE +0.049063 (se 0.003681), C-qt1 +3.205537 (se 0.022437), BASE-qt1 +3.156474 (se 0.022252). The refit recovers about half of the compression loss BASE has against Taylor's full tables (full-table BASE is +0.108, se 0.013, over BASE on 64512..65023). No gas, engine or board gain is claimed; nothing ran on the Rust engine and no Solidity was written.","P":{"command":"python3 -I gates.py (unmodified checkout files, CPython 3.12.3, before any change)","output":["seed   perfect-info      gaussian+plain    gaussian+qm       lattice+qm","10000  448.261189  443.455598  443.522608  443.663520","10001  576.779199  571.463409  571.623551  571.977293","10002  531.886497  525.168028  525.865859  526.411631"],"matches_README_at_6dp":true,"wall_seconds":9.6},"gates":{"P":"reproduced at 6 decimals before any change (see P)","Q":"C, BASE and qt1 scored paired on the same replica seeds; every choice made on 64000..65023, report 10000..12047 run once afterwards","F":"not applicable: no contract in this item"},"inputs":{"taylor_files_sha256":{"README.md":"02d2b51f17adbb5411a2a67f7a37255143ba381a73ff10abba557d7d23c0cef2","gates.py":"603280c1f595ca163ffaf2ba1ebdc7945fdb08dabae67fc53c922d39e1cce604","replica2.py":"6e37b4e61e1ffa5ffb1e43912ddb8659105989aa986eb48fcd1d7608f8dca685","real.py":"1305b5333731c046c3e6736cc2578eff4635c88132a39c22f5780d78aef2c81e","evalpol.py":"6e67667bfed7798bdd3384652a6efc6ba0b1fce653740d0f9af2bcbcb8e6a27c","latt.py":"925c0fe996b19fd51fb79a85d5e4dba4bc2f5453d0a156cc1e480f65a6a00cc8","pol_mid.json":"efbb5ee102db4d312f64565d9bf7232b6becbe02b624164f744a402bbbf2983b","qm_k1e-5.json":"5ff53e42ca87390e5f65d1caa3f7a480e67a8212fb0670274dfd562b05ba1afd"},"taylor_files_url_prefix":"https://raw.githubusercontent.com/gtmcknight/amm-sim/4812a6382311624e1506aca131ca35c378f930d9/research/solved/","artifacts":{"o1b":{"url":"https://api.imd.fun/artifacts/d56e3af5cc1e32dcc52be27046df89fb403d770a7fabf39e1a0ef94827349a5a","sha256_of_download":"d56e3af5cc1e32dcc52be27046df89fb403d770a7fabf39e1a0ef94827349a5a","matches_url_hash":true},"tc":{"url":"https://api.imd.fun/artifacts/d3c849db2ac31e567b57e52b80b86e8c24c842e5b4125b9d75c78e3cb07115b9","sha256_of_download":"d3c849db2ac31e567b57e52b80b86e8c24c842e5b4125b9d75c78e3cb07115b9","matches_url_hash":true},"qms":{"url":"https://api.imd.fun/artifacts/17d249ab0c33fc44931d9b105cf3f9e3846a560bd518b67310c2a56defacdecc","sha256_of_download":"17d249ab0c33fc44931d9b105cf3f9e3846a560bd518b67310c2a56defacdecc","matches_url_hash":true}},"extracted":{"o1b_code_zlib_sha256":"983fc13d833978256d9b411a1f7aed9b2471bb21c0d648de1b0ec7c75528a690 (recomputed; equals the artifact's sha256_of_zlib_bytes)","o1b_code_base64_text_sha256":{"recomputed":"4470eaf4f28047f238f2d166e00dc7982069f6e19f51740f4630005ca5e35a0e","stated":"4470eaf4f28047f238f2d166e00dc7982069f6e19f51740f4630005ca5e35a0e"},"port_py":"81ec1136146777e10056c3a1e1ad5282de5a1b53e90d425decefc288e5948aac","tell_py":"2ece4288ff6ba4d34aed25295c571f8f21fc09b51f8aa93b9a591a585ddb248f","edge_py":"0c54d2328f24d05bb14a0020f52d3348966dd32b73f44f9ef387ed8600c26add","twocomp_py":{"sha256":"9d3c1833321493fb51bbde4e1b4dabc2f8978bd6d774ee2af8dc8aed13f0f0af","stated_by_o1b":"9d3c1833321493fb51bbde4e1b4dabc2f8978bd6d774ee2af8dc8aed13f0f0af","how":"python fence after '## Code' in the TC artifact; it contains QMRef verbatim"},"qmref_py":{"sha256":"1dfc9b7a3c2237ff7a93dd9f0af54b9bdab16f2ce0a7eb00784cb4432598db92","stated_by_o1b":"1dfc9b7a3c2237ff7a93dd9f0af54b9bdab16f2ce0a7eb00784cb4432598db92","url":"https://raw.githubusercontent.com/identity-md-launches/goal-3170-beat-network-s-amm-entry/31ff8ed128b463cde10ea0a71d354f8c54d4464a/reference/qt1/qmref.py"},"blob_bin":{"sha256":"3ac7cb8811ecfb6a17f499aa4e9b70dd55f7fb6432b3aec3b904e11a3dece38e","bytes":10167,"how":"base64 fence in the QMS1 artifact"},"note":"port.py is used unmodified. tell.py and edge.py were read and their tests re-implemented with a cap in tellpol.py, because tell.py imports combo.py, which is not among this item's files. The o1-b artifact states no per-file hashes for port.py/tell.py/edge.py; the three values here are of the extracted files."}},"harness_check":{"what":"original QMRef (qt1) and original TwoComp, replica2.run vs o1-b port.run on the same streams, before any change","rows":[{"seed":74100,"qt1":{"replica2":622.7251756361247,"port":622.725175636084,"diff":-4.069988790433854e-11},"tc":{"replica2":622.2030098183866,"port":622.2030098183736,"diff":-1.2960299500264227e-11}},{"seed":74101,"qt1":{"replica2":676.1472913039771,"port":676.1472913040379,"diff":6.082245818106458e-11},"tc":{"replica2":677.7741147902392,"port":677.7741147901739,"diff":-6.52562448522076e-11}},{"seed":74102,"qt1":{"replica2":443.71781630551254,"port":443.71781630576095,"diff":2.48405740421731e-10},"tc":{"replica2":444.4130455337553,"port":444.4130454071331,"diff":-1.2662218296100036e-07}},{"seed":74103,"qt1":{"replica2":613.1552968264102,"port":613.1552968262837,"diff":-1.2653345038415864e-10},"tc":{"replica2":614.5694003646713,"port":614.5694003647691,"diff":9.777068044058979e-11}}],"max_abs_diff":1.2662218296100036e-07,"secs":10.4,"note":"TwoComp on seed 74102 differs by 1.3e-7, the largest; qt1 stays within 2.5e-10. qt1 report mean on the port is 533.262850; the QMS1 item published 533.2628 for the decoded tables on replica2."},"base":{"definition":["TwoComp state and fee rule unchanged; on a step's first own trade (dt>=1) the original pi is updated by the tell: pb = pi*La/(pi*La+(1-pi)*Lr)","sell (isBuy false): redo execute_sell_x (arb path) and execute_buy_x_with_y (retail path) from the previous reserve wads, the ask-fee wad in force and the current wads, over every double that maps to each wad. both: La=13069/16591, Lr=15/13131; arb only: pb=1; retail only: pb=0; neither: original pi and original component rule","buy (isBuy true): arbitrageur closed form amountX = fl(fl(xv-rx)/gamma) reachable within +-6 ulps: pass La=1, Lr=16/12889; fail: pb=0","cap: a wad with more than CAP candidate doubles (or a non-positive gamma) is unsupported -> original pi and original component rule","after tell-arb (pb>0.5) the posterior is always two components [(1-q, lp, 0), (q, mt, vt)] with q = max(1-pb, FLOOR): the retail branch never has zero weight","pb<=0.5: original rule ([(pb,lp,0),(1-pb,mt,vt)] if pb>0.005 else retail only)","sums are explicit left-to-right loops (no sum()), and the blob is decoded by DetTables; DetTables equals qmref.Tables in every cell for both blobs under CPython 3.12.3"],"inputs_used":"current TradeInfo wads (isBuy, amountX, amountY, timestamp, reserveX, reserveY), previous callback reserve wads (kept by the policy; initial 1e20/1e22 as afterInitialize reports), the fee wads the policy returned (trunc(clamp(fee)*1e18)). No true flag, price, rival spot or run config.","selected":{"floor":0.001,"cap":64},"selected_on":"64000..64255 (a 256-seed subset of the choose window 64000..65023), before any fitting","rule":"floor: highest mean edge among {1e-6,1e-5,1e-4,1e-3} at cap 64. cap: smallest of {4,64,1024} (at floor 1e-4) whose paired difference to cap 1024 is not below -2 se.","cap_vs_1024":{"cap4":{"mean_diff":-0.021001,"se":0.00852},"cap64":{"mean_diff":-0.006368,"se":0.004661}},"grid":[{"floor":1e-06,"cap":64,"mean_edge":529.885047,"vs_selected":{"mean_diff":-0.001761,"se":0.007575,"min":-0.458036},"cap_fallbacks":768,"first_trades":1789719,"label_errors":616,"secs":42.7},{"floor":1e-05,"cap":64,"mean_edge":529.883062,"vs_selected":{"mean_diff":-0.003746,"se":0.007579,"min":-0.489016},"cap_fallbacks":784,"first_trades":1789742,"label_errors":615,"secs":40.4},{"floor":0.0001,"cap":64,"mean_edge":529.87607,"vs_selected":{"mean_diff":-0.010737,"se":0.00774,"min":-0.607357},"cap_fallbacks":742,"first_trades":1789691,"label_errors":636,"secs":41.8},{"floor":0.001,"cap":64,"mean_edge":529.886808,"vs_selected":{"mean_diff":0.0,"se":0.0,"min":0.0},"cap_fallbacks":747,"first_trades":1789043,"label_errors":623,"secs":44.2},{"floor":0.0001,"cap":4,"mean_edge":529.861438,"vs_selected":{"mean_diff":-0.02537,"se":0.008755,"min":-0.83771},"cap_fallbacks":10468,"first_trades":1789426,"label_errors":2127,"secs":42.5},{"floor":0.0001,"cap":1024,"mean_edge":529.882439,"vs_selected":{"mean_diff":-0.004369,"se":0.007444,"min":-0.600384},"cap_fallbacks":66,"first_trades":1789674,"label_errors":487,"secs":39.5}],"reading":"the four floors are within about 1.5 se of each other, so 256 seeds do not resolve the floor; cap 4 falls back about 14 times as often as cap 64. The -2 se threshold of the cap rule was fixed after the grid had been seen (the first intention was 1 se, which would have picked cap 1024); cap 64 and cap 1024 differ by 0.006 (se 0.005), so either choice is within noise. Both were chosen on choose seeds only and before any fit.","frozen_before_fit":true},"hidden_input_invariance":{"test":"tests.py t2: the same policy run twice on the port, once normally and once with init(sigma,lam,size) replaced by random numbers, on_trade's (arb,p,s2,j) replaced by random values, and the port's _ti previous-reserve wads, fee wads and true-path field replaced by garbage (only _ti[3], the current TradeInfo wads, passes through). Returned fees are compared on every callback.","structural":"o1-b port.py already calls on_trade(..., False, nan, nan, 0), so the only hidden channels on the port are the init args and _ti[4]; tellpol.py reads neither. The true path is read only by score.py's Diag subclass, for the error counts reported here, and is never fed back.","BASE":{"seeds":[74100,74101,74102,74103],"callbacks":46429,"fees_identical_every_callback":true,"edges_identical":true,"own_wad_bookkeeping_mismatches":0,"tell_kind_by_true_path":{"arb/arb":1571,"both/arb":6560,"both/retail":10,"cap/arb":10,"cap/retail":4,"fail/retail":6707,"pass/arb":7433,"pass/retail":7,"ret/retail":6734}},"C":{"seeds":[74100,74101,74102,74103],"callbacks":46413,"fees_identical_every_callback":true,"edges_identical":true,"own_wad_bookkeeping_mismatches":0,"tell_kind_by_true_path":{"arb/arb":1587,"both/arb":6533,"both/retail":5,"cap/arb":11,"fail/retail":6706,"pass/arb":7418,"pass/retail":9,"ret/retail":6753}}},"visits":{"policy":"frozen BASE","train_seeds":"64000..64511","train_lookups":5572165.0,"val_seeds":"64512..65023 (val_rms diagnostic only, never fitted)","val_lookups":5601643.0,"how":"every qm lookup splatted tri-linearly on (sd level,row,column) nodes; the ask point and the point-mirrored bid point (row->320-row, col->last-col), as real.py interpolates","share_by_table_ppm":{"0c":348388,"1s":399543,"1c":63103,"2s":184677,"2c":4290}},"fit":{"target":"F = max(A_full - w, 0) at every grid node of Taylor's full qm_k1e-5.json (ask side; bid by mirror), 5 tables, 8 sd levels x 321 rows x 13 or 49 columns","objective":"per table: sum_n w_n r_n^2 + lam*|c - c_orig|^2; w_n = train visits at node n in ppm of all train lookups; c in bps; nodes with F=0 are one-sided (target min(prediction,0), 12 sweeps from c_orig); then round to int16 at 1/8 bps. lam -> infinity is the zero-change blob.","solver":"scipy sparse kron design, dense Cholesky (numpy 2.5.4, scipy 1.18.1) for fitting only; the blob is decoded and scored with the standard library","fit_seconds":19.5,"regularization_grid":[{"blob":"orig","lam":null,"sha256":"3ac7cb8811ecfb6a17f499aa4e9b70dd55f7fb6432b3aec3b904e11a3dece38e","train_rms_bps":0.8325,"val_rms_bps":0.8346,"unvisited_rms_bps":5.0202,"coef_changed":0,"coef_max_move_bps":0.0,"unvisited_max_node_move_bps":0.0,"clipped_to_int16":0,"edge_vs_BASE_64512_65023":{"n":512,"mean_diff":0.0,"se":0.0,"note":"zero-change control: it is BASE"}},{"blob":"cand_0.03.bin","lam":0.03,"sha256":"f5b67d401d0f90650ae73311abf816d4e910d67e5106c16be5d012d9ae904fda","train_rms_bps":0.7049,"val_rms_bps":0.7088,"unvisited_rms_bps":6.0063,"coef_changed":3537,"coef_max_move_bps":180.5,"unvisited_max_node_move_bps":116.267,"clipped_to_int16":0,"edge_vs_BASE_64512_65023":{"n":512,"mean_diff":0.047187,"se":0.008093,"min":-0.823362,"wins":316},"secs":93.8},{"blob":"cand_0.3.bin","lam":0.3,"sha256":"a011e5d99d191fa252f0e31735b7a3be601645dd82ed1d0352e5709ec6dbb2b9","train_rms_bps":0.7204,"val_rms_bps":0.7239,"unvisited_rms_bps":5.1085,"coef_changed":2989,"coef_max_move_bps":66.0,"unvisited_max_node_move_bps":27.473,"clipped_to_int16":0,"edge_vs_BASE_64512_65023":{"n":512,"mean_diff":0.040894,"se":0.007661,"min":-0.58916,"wins":306},"secs":91.3},{"blob":"cand_3.bin","lam":3.0,"sha256":"e1067845009812bbd92a5f124a984c68680e68bb44e85b1cc2f1e21c27a1351c","train_rms_bps":0.7411,"val_rms_bps":0.7439,"unvisited_rms_bps":5.0106,"coef_changed":2076,"coef_max_move_bps":18.125,"unvisited_max_node_move_bps":3.665,"clipped_to_int16":0,"edge_vs_BASE_64512_65023":{"n":512,"mean_diff":0.057253,"se":0.007409,"min":-0.561132,"wins":317},"secs":91.6},{"blob":"cand_10.bin","lam":10.0,"sha256":"ca5a606d074224ccc72fb4999c1d01a677ba00440ba39df2f99eba4913a72e1d","train_rms_bps":0.754,"val_rms_bps":0.7567,"unvisited_rms_bps":5.0083,"coef_changed":1599,"coef_max_move_bps":11.0,"unvisited_max_node_move_bps":2.262,"clipped_to_int16":0,"edge_vs_BASE_64512_65023":{"n":512,"mean_diff":0.058324,"se":0.007068,"min":-0.514016,"wins":323},"secs":91.5},{"blob":"cand_30.bin","lam":30.0,"sha256":"aa056e860e99d99efaabb3fcd44d6bc668059bf6458b1a555cab1f652cc5419b","train_rms_bps":0.7664,"val_rms_bps":0.7688,"unvisited_rms_bps":5.0106,"coef_changed":1156,"coef_max_move_bps":5.75,"unvisited_max_node_move_bps":1.401,"clipped_to_int16":0,"edge_vs_BASE_64512_65023":{"n":512,"mean_diff":0.055207,"se":0.006914,"min":-0.497019,"wins":339},"secs":91.9},{"blob":"cand_300.bin","lam":300.0,"sha256":"084810380c05027264fcb72aacacf880722194d4613e618e604725b8b3b7fe98","train_rms_bps":0.7936,"val_rms_bps":0.7959,"unvisited_rms_bps":5.0168,"coef_changed":407,"coef_max_move_bps":2.25,"unvisited_max_node_move_bps":0.8,"clipped_to_int16":0,"edge_vs_BASE_64512_65023":{"n":512,"mean_diff":0.035022,"se":0.005971,"min":-0.490288,"wins":318},"secs":90.5},{"blob":"cand_3000.bin","lam":3000.0,"sha256":"7525f4a33e987bae393ac33ea9dd38fcd54fb312fc7fbf1a8db567e396f46130","train_rms_bps":0.8157,"val_rms_bps":0.8179,"unvisited_rms_bps":5.0197,"coef_changed":68,"coef_max_move_bps":1.125,"unvisited_max_node_move_bps":0.184,"clipped_to_int16":0,"edge_vs_BASE_64512_65023":null},{"blob":"cand_30000.bin","lam":30000.0,"sha256":"ade4c13db8567392c9cc38d84ee5afb4c7df6bc85dc3bc865ff26945a15672b3","train_rms_bps":0.8292,"val_rms_bps":0.8313,"unvisited_rms_bps":5.0202,"coef_changed":2,"coef_max_move_bps":0.25,"unvisited_max_node_move_bps":0.001,"clipped_to_int16":0,"edge_vs_BASE_64512_65023":null}],"selection":{"on":"64512..65023 only","rule":"highest mean paired edge vs BASE among the scored candidates and the zero-change control (0 by construction)","selected_lam":10.0,"selected_blob":"cand_10.bin","note":"lam 3, 10 and 30 are within 0.5 se of each other. lam 3000 and 30000 were fitted but not edge-scored (their val_rms is nearly the original)."},"selected_per_table":{"0c":{"train_rms":0.4851,"val_rms":0.4829,"val_mean_abs":0.1493,"max_abs_err_visited":36.9362,"unvisited_rms":1.069,"unvisited_max_move_vs_orig":1.4059,"max_fee_bps":786.0082,"nodes_over_1000bps":0,"coef_changed":172,"coef_n":960,"coef_max_move_bps":7.0,"coef_at_int16_limit":0,"coef_range_bps":[-614.625,808.5],"visit_ppm_train":348387.7452},"1s":{"train_rms":0.769,"val_rms":0.7713,"val_mean_abs":0.2937,"max_abs_err_visited":82.2275,"unvisited_rms":5.9579,"unvisited_max_move_vs_orig":2.0152,"max_fee_bps":795.5389,"nodes_over_1000bps":0,"coef_changed":601,"coef_n":1344,"coef_max_move_bps":5.75,"coef_at_int16_limit":0,"coef_range_bps":[-1107.375,838.0],"visit_ppm_train":399543.445},"1c":{"train_rms":1.3689,"val_rms":1.373,"val_mean_abs":0.4814,"max_abs_err_visited":40.465,"unvisited_rms":2.125,"unvisited_max_move_vs_orig":2.2623,"max_fee_bps":793.9578,"nodes_over_1000bps":0,"coef_changed":278,"coef_n":864,"coef_max_move_bps":11.0,"coef_at_int16_limit":0,"coef_range_bps":[-327.25,880.25],"visit_ppm_train":63102.5822},"2s":{"train_rms":0.7495,"val_rms":0.7608,"val_mean_abs":0.2925,"max_abs_err_visited":66.0577,"unvisited_rms":4.7432,"unvisited_max_move_vs_orig":1.3161,"max_fee_bps":792.4653,"nodes_over_1000bps":0,"coef_changed":463,"coef_n":1344,"coef_max_move_bps":4.25,"coef_at_int16_limit":0,"coef_range_bps":[-623.375,813.875],"visit_ppm_train":184676.6921},"2c":{"train_rms":2.5698,"val_rms":2.5436,"val_mean_abs":0.9941,"max_abs_err_visited":75.5477,"unvisited_rms":6.603,"unvisited_max_move_vs_orig":1.7542,"max_fee_bps":817.7199,"nodes_over_1000bps":0,"coef_changed":85,"coef_n":384,"coef_max_move_bps":2.625,"coef_at_int16_limit":0,"coef_range_bps":[-750.125,1150.625],"visit_ppm_train":4289.5356}},"original_per_table_val_rms":{"0c":0.5814,"1s":0.8553,"1c":1.4543,"2s":0.8185,"2c":2.6186}},"full_table_diagnostic":{"what":"BASE reading Taylor's full 5 bps tables instead of the blob, 64512..65023, paired vs BASE: the compression loss a refit can at most recover. Not used for selection.","paired":{"n":512,"mean_diff":0.108129,"se":0.012871,"min":-1.928511,"wins":349},"secs":83.8},"checks":{"bytes":{"orig":10167,"new":10167},"sha256":{"orig":"3ac7cb8811ecfb6a17f499aa4e9b70dd55f7fb6432b3aec3b904e11a3dece38e","new":"ca5a606d074224ccc72fb4999c1d01a677ba00440ba39df2f99eba4913a72e1d"},"header_and_knot_bytes":375,"header_and_knots_identical":true,"layout_identical":true,"coef":{"n":4896,"changed":1599,"max_move_bps":11.0,"min_int":-8859,"max_int":9205,"at_int16_limit":0},"decoders":{"det_vs_qmref_cells_differing":0},"symmetry":{"B_ne_mirrorA_cells":0,"policy_max_abs_bid(w,w2)_minus_ask(-w,-w2)_20000pts":1.1102230246251565e-16},"saturation":{"min_fee_bps_at_nodes":0.0,"max_fee_bps_at_nodes":817.72,"max_fee_fraction":0.078518,"fee_cap":0.1},"decoded_fee_error_vs_full_bps":{"weights":"BASE visits 64512..65023","new":0.7567,"orig":0.8346,"per_table":{"0c":{"val_rms_new":0.4829,"val_rms_orig":0.5814},"1s":{"val_rms_new":0.7713,"val_rms_orig":0.8553},"1c":{"val_rms_new":1.373,"val_rms_orig":1.4543},"2s":{"val_rms_new":0.7608,"val_rms_orig":0.8185},"2c":{"val_rms_new":2.5436,"val_rms_orig":2.6186}},"max_node_move_new_vs_orig_bps":6.629},"extrapolation":{"points":4096,"non_finite_or_out_of_range_or_not_equal_to_clamped_lookup":0,"max_abs_fee_new_minus_orig":0.0,"max_fee":0.1}},"scores":{"choose":{"seeds":"64000..65023","n_steps":10000,"coverage":{"C":1024,"BASE":1024,"qt1":1024,"expected":1024,"full":true},"paired":{"C-BASE":{"n":1024,"mean_diff":0.047886,"se":0.005075,"worst5":[[64502,-0.553213],[64721,-0.514016],[64990,-0.502682],[64307,-0.452524],[64269,-0.396699]],"min":-0.553213,"wins":635},"C-qt1":{"n":1024,"mean_diff":3.149075,"se":0.030824,"worst5":[[64684,-1.139062],[64636,0.30942],[64509,0.348054],[64322,0.378924],[64593,0.433685]],"min":-1.139062,"wins":1023},"BASE-qt1":{"n":1024,"mean_diff":3.101188,"se":0.030949,"worst5":[[64684,-1.234348],[64509,-0.006689],[64322,0.176181],[64778,0.278953],[64593,0.354864]],"min":-1.234348,"wins":1022}},"mean_edge":{"C":538.869624,"BASE":538.821737,"qt1":535.720549},"min_edge":{"C":{"edge":334.935506,"seed":64931},"BASE":{"edge":335.134914,"seed":64931},"qt1":{"edge":332.134421,"seed":64931}},"tell":{"BASE":{"first_trades":7189748,"tell_kinds":{"both":1520597,"arb":420947,"ret":1648979,"neither":0,"pass":1950581,"fail":1645566,"cap":3078,"unsup":0},"fallback_to_original_pi":3078,"fallback_rate":0.0004281,"kind_by_true_path":{"arb_arb":420947,"both_arb":1518762,"both_ret":1835,"cap_arb":2522,"cap_ret":556,"fail_ret":1645566,"pass_arb":1948091,"pass_ret":2490,"ret_ret":1648979},"label_errors_after_tell":2447,"label_error_rate_after_tell":0.0003403,"label_errors_original_pi":851336,"label_error_rate_original_pi":0.11841},"C":{"first_trades":7186239,"tell_kinds":{"both":1517661,"arb":420472,"ret":1650657,"neither":0,"pass":1946652,"fail":1647718,"cap":3079,"unsup":0},"fallback_to_original_pi":3079,"fallback_rate":0.0004285,"kind_by_true_path":{"arb_arb":420472,"both_arb":1515771,"both_ret":1890,"cap_arb":2511,"cap_ret":568,"fail_ret":1647718,"pass_arb":1944222,"pass_ret":2430,"ret_ret":1650657},"label_errors_after_tell":2564,"label_error_rate_after_tell":0.0003568,"label_errors_original_pi":852089,"label_error_rate_original_pi":0.11857}},"fit_half_64000_64511_C-BASE":{"n":512,"mean_diff":0.037449,"se":0.007262,"min":-0.553213}},"report":{"seeds":"10000..12047","n_steps":10000,"coverage":{"C":2048,"BASE":2048,"qt1":2048,"expected":2048,"full":true},"paired":{"C-BASE":{"n":2048,"mean_diff":0.049063,"se":0.003681,"worst5":[[11701,-1.402185],[11455,-0.911079],[10728,-0.859496],[11998,-0.76263],[10357,-0.591789]],"min":-1.402185,"wins":1279},"C-qt1":{"n":2048,"mean_diff":3.205537,"se":0.022437,"worst5":[[11062,-1.338114],[11701,-0.493342],[11376,-0.255387],[11519,0.207946],[10642,0.276573]],"min":-1.338114,"wins":2045},"BASE-qt1":{"n":2048,"mean_diff":3.156474,"se":0.022252,"worst5":[[11062,-1.380901],[11376,-0.101975],[11519,0.094928],[10141,0.148161],[10642,0.34402]],"min":-1.380901,"wins":2046}},"mean_edge":{"C":536.468387,"BASE":536.419323,"qt1":533.26285},"min_edge":{"C":{"edge":329.895532,"seed":10532},"BASE":{"edge":329.853244,"seed":10532},"qt1":{"edge":326.727276,"seed":10532}},"tell":{"BASE":{"first_trades":14357942,"tell_kinds":{"both":3028556,"arb":841379,"ret":3297038,"neither":0,"pass":3890992,"fail":3293787,"cap":6190,"unsup":0},"fallback_to_original_pi":6190,"fallback_rate":0.0004311,"kind_by_true_path":{"arb_arb":841379,"both_arb":3024825,"both_ret":3731,"cap_arb":5038,"cap_ret":1152,"fail_ret":3293787,"pass_arb":3886258,"pass_ret":4734,"ret_ret":3297038},"label_errors_after_tell":4894,"label_error_rate_after_tell":0.0003409,"label_errors_original_pi":1706914,"label_error_rate_original_pi":0.11888},"C":{"first_trades":14351447,"tell_kinds":{"both":3021394,"arb":841368,"ret":3300983,"neither":0,"pass":3883771,"fail":3297752,"cap":6179,"unsup":0},"fallback_to_original_pi":6179,"fallback_rate":0.0004305,"kind_by_true_path":{"arb_arb":841368,"both_arb":3017528,"both_ret":3866,"cap_arb":5109,"cap_ret":1070,"fail_ret":3297752,"pass_arb":3878971,"pass_ret":4800,"ret_ret":3300983},"label_errors_after_tell":4962,"label_error_rate_after_tell":0.0003457,"label_errors_original_pi":1709261,"label_error_rate_original_pi":0.1191}}}},"blobs":{"original":{"bytes":10167,"sha256":"3ac7cb8811ecfb6a17f499aa4e9b70dd55f7fb6432b3aec3b904e11a3dece38e"},"new":{"bytes":10167,"sha256":"ca5a606d074224ccc72fb4999c1d01a677ba00440ba39df2f99eba4913a72e1d","zlib_base64":"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"},"preserved":"first 375 bytes (header, W2L, per-table sizes and knots) identical; 4896 int16 coefficients at 1/8 bps in the same order; mirroring unchanged (B = point mirror of A in the decoder)","coefficients_changed":1599,"max_coefficient_move_bps":11.0},"bundle":{"format":"base64(zlib(json {filename: source}))","files":["tellpol.py","score.py","fit.py","checks.py","tests.py","agg.py"],"sha256_of_zlib_bytes":"9ce0c1e69fc8cababce2cd23ded3c5c03164008914871b673a1cc1e002ee2ce4","zlib_base64":"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","needs":"next to the bundle files: replica2.py (Taylor), port.py (o1-b), twocomp.py (TC), qmref.py (qt1), blob.bin (QMS1), cand.bin (blobs.new); fit.py and checks.py also read qm_k1e-5.json; fit.py needs numpy and scipy in ../pylib; score.py base:..:FULL reads ../taylor/qm_k1e-5.json"},"commands":[{"cmd":"python3 -I gates.py","secs":9.6,"what":"P"},{"cmd":"python3 -I tests.py t1 ; python3 -I tests.py t2 ; python3 -I tests.py t2 blob=cand.bin","secs":[10.4,6.4,6.3],"what":"port vs replica2; invariance for BASE and C"},{"cmd":"python3 -I score.py base:FLOOR:CAP 64000 256 state/sel_FLOOR:CAP.json  (6 configs)","secs":[42.7,40.4,41.8,44.2,42.5,39.5],"what":"floor/cap grid"},{"cmd":"python3 -I score.py base:1e-3:64 64000 512 state/base_c1.json state/vis_train.bin ; python3 -I score.py base:1e-3:64 64512 512 state/base_c2.json state/vis_val.bin","secs":[114.6,123.9],"what":"BASE on choose, with visits"},{"cmd":"python3 -I fit.py qm_k1e-5.json state/vis_train.bin state/vis_val.bin state 0.03,0.3,3,10,30,300,3000,30000","secs":19.5,"what":"fit (run twice; lam 3 and 30 blobs reproduced byte for byte)"},{"cmd":"python3 -I score.py base:1e-3:64:state/cand_LAM.bin 64512 512 state/c_LAM_c2.json  (LAM 0.03,0.3,3,10,30,300) ; python3 -I score.py base:1e-3:64:FULL 64512 512 state/full_c2.json","secs":"90 to 94 each; FULL 83.8","what":"regularization selection; full-table diagnostic"},{"cmd":"python3 -I checks.py cand.bin qm_k1e-5.json state/vis_val.bin","what":"symmetry, saturation, extrapolation, decoded fee errors"},{"cmd":"python3 -I score.py qt1 64000 1024 state/qt1_c.json ; python3 -I score.py base:1e-3:64:cand.bin 64000 512 state/C_c1.json","secs":[128.2,84.3],"what":"qt1 and C on choose (C on 64512..65023 is the selection run)"},{"cmd":"python3 -I score.py {base:1e-3:64 | base:1e-3:64:cand.bin | qt1} 10000.. in parts","secs":{"BASE":[168.9,183.5],"C":[179.8,177.4],"qt1":[66.5,186.2]},"what":"report, run once after C was frozen"},{"cmd":"python3 -I agg.py A.json[,A2.json] B.json[,B2.json]","what":"paired{n,mean_diff,se,worst5}"}],"environment":"CPython 3.12.3, plain python3 -I, 4 processes. Longest single score.py call 186.2 s; no shell command ran over 8 minutes; about 55 minutes wall in total. State was kept under test/scratch/w/state, this run's scratch directory.","order_of_work":["P and harness check","floor/cap on 64000..64255, BASE frozen","BASE on 64000..65023 with visits","fit on 64000..64511 visits","lam selected on 64512..65023, C frozen (sha256 in blobs.new)","report 10000..12047 scored once for BASE, C and qt1; nothing was retuned afterwards"],"caveats":["Replica only. The wad tell depends on bit-exact IEEE behaviour and was never checked against the compiled Rust engine (o1-b says the same); if the engine rounds differently, the tell and with it everything BASE and C hold over qt1 goes away. No gas, bytecode or board gain is claimed.","C-BASE on the report seeds is +0.0491 (se 0.0037): clear of noise but small, about 1.5% of the BASE-qt1 gap. Worst report seed for C-BASE: [11701, -1.402185].","BASE-qt1 and C-qt1 each have one clearly negative seed per window (64684 on choose, 11062 on report, -1.1 to -1.4).","Floor and cap were chosen on 256 of the 1024 choose seeds and their differences are within noise.","C on 64512..65023 inside the choose score is the run that selected lam, so the choose C-BASE figure is selection-biased on that half; the fit half gives +0.0374 (se 0.0073) and the report +0.0491.","BASE is bit-sensitive. PyPy 3.10 was tried as an accelerator and gave per-seed BASE edges up to 0.14 away from CPython 3.12 on 16 seeds (means equal to 0.0003), so it was dropped; every number here is CPython. TellTC replaces sum() by explicit loops, which can differ from TwoComp.moments in the last bit under CPython 3.12.","The policy reads the current wads from the attribute the port sets (self._ti[3]); on replica2.run alone (no wads) every first trade is unsupported and it behaves as TwoComp.","The refit keeps the QMS1 knots, which were placed for the Gaussian filter's visits on 60000..60255; moving knots was out of scope. The one B1c cell that breaks the mirror in the full tables (0.71 bps, QMS1 write-up) is still reproduced from the mirror.","Out-of-grid check: lookups clamp row, column and sd to the node grid, so the spline is never extrapolated; at 4096 out-of-grid points the new and original blobs return identical fees (all at 0 or the 10% cap). Inside the grid, nodes no BASE lookup touched moved by at most 2.262 bps and their rms error against the full tables is 5.0083 bps (original 5.0202)."]}