{"item":"res-d","batch":"e04-residual-b","task":"residual fee head on COMBO, variant d (8 tanh units), sep-CMA-ES; replica only","credit":"Taylor (gtmcknight): replica2.py, Gaussian filter, fee rule, qm tables (github.com/gtmcknight/amm-sim @4812a638 research/solved). qt1 (QMRef), TC (TwoComp) and COMBO are IdentityMD items on them; this item adds a residual head only.","environment":{"python":"3.12.16 (standalone CPython installed with uv into scratch/, the machine had no python3)","machine":"linux aarch64, cgroup cpu.max 400000/100000 (4 CPUs), 10 pool workers","stdlib_only":true},"gates":{"P":{"pass":true,"commit":"4812a6382311624e1506aca131ca35c378f930d9","gates_output":"seed   perfect-info      gaussian+plain    gaussian+qm       lattice+qm\n10000  448.261189  443.455598  443.522608  443.663520\n10001  576.779199  571.463409  571.623551  571.977293\n10002  531.886497  525.168028  525.865859  526.411631\n","qt1_10000_10002":["443.101741","571.454483","525.908441"],"qt1_expected":["443.101741","571.454483","525.908441"],"COMBO_10000_10002":["444.195330","572.172682","526.274760"],"COMBO_expected":["444.195330","572.172682","526.274760"]},"Q":{"pass":true,"method":"recorded every callback of a head run with random non-zero output weights, replayed it to a fresh head with arb/p/s2/j drawn at random per callback and init(sigma, lam, size) random; returned fees compared bit for bit","runs":[{"seed":64000,"callbacks":9848,"fees_bit_identical":true},{"seed":64001,"callbacks":11125,"fees_bit_identical":true}],"windows":"train 64000..64511, validation 64512..64767, scale fit 64000..64015, choose report 64000..65023; report 10000..12047 run once per policy after freezing; nothing selected on report"},"F":"not applicable: replica item, no contract"},"hashes":{"artifacts_sha256_verified":{"qt1":true,"TC":true,"COMBO":true,"QMS1":true},"artifacts":{"qt1":"8a6e303b662e1a5ea9bb47f31acedd3988ef65b3ddb66d074517aa59ad41ca40","TC":"d3c849db2ac31e567b57e52b80b86e8c24c842e5b4125b9d75c78e3cb07115b9","COMBO":"d360bea6c8c982717b7b43a8069a87013169c89ee3e4da25622db4d880b936d4","QMS1":"17d249ab0c33fc44931d9b105cf3f9e3846a560bd518b67310c2a56defacdecc"},"amm_sim_commit":"4812a6382311624e1506aca131ca35c378f930d9","extracted":{"head.py":"e67d5ff5b4ea45b52596d1afdb7b154714b58632248bc945cfdbd4d60cbcf8bd","combo.py":"c5b711a0b4fb4783e8879a484bf87a427171f4a11aa5eb5ec3513c24ead1a796","twocomp.py":"9d3c1833321493fb51bbde4e1b4dabc2f8978bd6d774ee2af8dc8aed13f0f0af","qmref.py":"1dfc9b7a3c2237ff7a93dd9f0af54b9bdab16f2ce0a7eb00784cb4432598db92","blob.bin":"3ac7cb8811ecfb6a17f499aa4e9b70dd55f7fb6432b3aec3b904e11a3dece38e"},"own_source":{"head.py":"e67d5ff5b4ea45b52596d1afdb7b154714b58632248bc945cfdbd4d60cbcf8bd","sepcma.py":"40df1606e705f714457a8086b5a4ce3f6eea0ab6bc9caaf0fb71fec60b357807","common.py":"75bdfd1de41dd18bcb31614d83e205398b09e87e30f68b3e5b9ed68c192f95c6","prep.py":"8c30b0f41b4f7e2334a503bbbecdb2c9e73be8d578a713b01a7a2d0e3d853660","train.py":"1116904a48d4da4c5d03d32686941df35262a9d1d2591a1459301c3b81092b6c","score.py":"03b2baa797d558ee4c91c26732f1c411aea089f996c50528096363bed91cc7b8","build_result.py":"2ffc157ee1cee34b93793fb8c27b732dde9a53fa72844aedd8739cac66fd758b"},"frozen_json":"a39e22d737d564cef74a3cd0225dc6a374ff48a3886ba1ef75cf1db8a45547b6","stats_json":"b5909dc4e85e0969ede3039d21abf783cf2789e5c4c176cac5e1100c377fe449"},"frozen_spec":{"policy":"Head(blob.bin, params, stats) from head.py: COMBO (combo.py, statement for statement, plus storing pi) then 14 standardized inputs -> 8 tanh -> 2 outputs 3*tanh(.) bps added to (fb, fa), clipped to [0, 0.1]","inputs":["pi","belief mean - midpoint of the no-arb band of the quotes in force (bps)","belief sd (bps)","w (bps)","R (bps)","phase","ln(amountY / 20*e^-0.72) size over the size law, log units","silent steps min(dt-1,12)","comp1 weight","comp1 mean - ln s (bps)","comp1 sd (bps)","comp2 weight","comp2 mean - ln s (bps)","comp2 sd (bps)"],"standardization":{"rule":"(x - center) / scale, both fixed from all callbacks of COMBO on choose seeds 64000..64015","callbacks":173138,"center":[0.5659784928680472,-0.7769972665361297,11.66218483457382,1.5955317937130924,0.5928433950498586,0.8238919243609144,-0.6408006948402148,0.2778997100578729,0.6658579398362827,-1.6377479487942783,1.0981651099884029,0.33414206016371734,-1.3956306738693052,10.555136369310578],"scale":[0.3651062505417777,46.92470280974327,6.8619407813071405,85.71864636359835,43.658967540250345,0.7244666523408317,1.1673350102788818,0.8065080051612149,0.33212859720253396,87.8440790071979,3.414374492454624,0.33212859720253396,78.90893799547375,4.134714497402093]},"parameter_count":138,"layout":"W1 8x14 row-major, b1 8, W2 2x8 (row 0 bid, row 1 ask), b2 2","hidden_init":"N(0, 1/sqrt(14)) from random.Random(12345); b1, W2, b2 = 0","cma":"sep-CMA-ES (diagonal C, Ros & Hansen 2008 learning rates), popsize 8, mu 4, log weights, rng seed 1, minimizes -mean paired gain vs cached COMBO edge on a rotating minibatch of 64 train seeds at n_steps 10000","selection":"checkpoint (CMA mean) with the best validation on 64512..64767; gen 0 is the untrained mean and equals COMBO"},"zero_head_identity":[{"seed":10000,"combo_edge":444.1953300340845,"head_edge":444.1953300340845,"edges_equal":true,"callbacks":9083,"fees_bit_identical":true,"raw_return_differs":0,"raw_example":[]},{"seed":10001,"combo_edge":572.1726816933606,"head_edge":572.1726816933606,"edges_equal":true,"callbacks":12270,"fees_bit_identical":true,"raw_return_differs":0,"raw_example":[]},{"seed":10002,"combo_edge":526.2747600531877,"head_edge":526.2747600531877,"edges_equal":true,"callbacks":10125,"fees_bit_identical":true,"raw_return_differs":1,"raw_example":[[7890,[0.12476733159999975,0.0],[0.1,0.0]]]},{"seed":10003,"combo_edge":573.8110971184866,"head_edge":573.8110971184866,"edges_equal":true,"callbacks":11741,"fees_bit_identical":true,"raw_return_differs":0,"raw_example":[]}],"seconds_per_1000_steps_single_process":{"COMBO":0.006646370833429198,"head":0.011163494433276355},"as_specified_run":{"name":"variant d as specified: all 138 params, sigma0 0.3","sigma0":0.3,"trained_params":138,"mode":"all","generations":41,"train_wall_seconds":1179.7,"final_sigma":0.2259962790103512,"training_curve":[[0,null,0.0],[1,-0.227016,null],[2,-0.312551,null],[3,-0.386005,null],[4,-0.505024,-0.401449],[5,-0.312046,null],[6,-0.577702,null],[7,-0.591486,null],[8,-0.582196,-0.614251],[9,-0.739335,null],[10,-0.728875,null],[11,-0.732108,null],[12,-0.727654,-0.592797],[13,-0.615326,null],[14,-0.64965,null],[15,-0.544272,null],[16,-0.490164,-0.480485],[17,-0.711266,null],[18,-0.492692,null],[19,-0.597869,null],[20,-0.458545,-0.495554],[21,-0.584892,null],[22,-0.489636,null],[23,-0.637044,null],[24,-0.667997,-0.596739],[25,-0.667743,null],[26,-0.497164,null],[27,-0.574475,null],[28,-0.483822,-0.493169],[29,-0.549196,null],[30,-0.641985,null],[31,-0.609784,null],[32,-0.392371,-0.490241],[33,-0.415406,null],[34,-0.538318,null],[35,-0.351825,null],[36,-0.407602,-0.432246],[37,-0.446469,null],[38,-0.498366,null],[39,-0.665758,null],[40,-0.56152,-0.497667],[41,-0.630602,null]],"validations":[[0,0.0],[4,-0.401449],[8,-0.614251],[12,-0.592797],[16,-0.480485],[20,-0.495554],[24,-0.596739],[28,-0.493169],[32,-0.490241],[36,-0.432246],[40,-0.497667]],"selected_gen":0,"validation_at_selection":0.0,"frozen_weights":[-0.03308715437882328,0.019115850566246966,0.10245972698416962,-0.20044514592331217,-0.11875872820943496,0.1290216003856351,-0.14527222071654158,-0.06396047433818225,0.20222153064761011,0.2003670728829884,-0.15300728807037753,-0.06290345623746797,-0.23638187459869753,-0.08212999278197877,0.11290946284469724,-0.030458286873202234,0.2730044903604487,-0.036861204186424884,-0.15130022821980782,-0.0037419099885917287,-0.033613578761160015,-0.17020012938078596,-0.031726493810521235,0.048925986088562905,-0.37306721093906353,0.591815801619824,0.437619631039542,-0.058644035002906054,0.6342674612772654,0.013769546420165382,0.31565985949112985,-0.3329139949245999,0.05288073016774706,0.11034368726969458,-0.4781304645483523,0.28466466717516625,-0.3850506047803147,-0.20569262545552852,0.03546851429151385,0.3235266060051074,-0.12592296237645467,-0.48270439533207793,0.06826401653071691,-0.18195249907387775,0.130462368549171,-0.13346407321006498,0.1089621985516729,-0.3222162923087807,0.13976553228242283,0.3276473327640361,-0.670215854233465,-0.07714572936523113,0.15462095658579564,0.04129788524748191,0.5021970636795855,-0.052912821968877234,0.11306009698072775,0.15617542427566264,-0.15149037187666853,-0.03802302312625276,-0.037999037016099226,0.07603519484410304,-0.02578181716395206,-0.1049950213775521,0.6011853759413393,-0.0203894948748699,0.11107791709293718,-0.446225972893398,-0.27152125106409425,0.29744932269404395,0.021628149714376252,-0.30522730237075,0.347221976600741,0.14133837350075287,-0.1680878418869961,0.32602854708177714,-0.5798644844335422,0.3148637119891221,0.1369887814608869,0.4364392426535371,-0.10824989353072591,-0.3231923928759373,0.14060288599341308,0.0773503383761791,0.18946319485909485,-0.03950669039297138,0.1139346461209294,0.1886890863012107,-0.31486931567140053,-0.0747998780975266,0.02134294516026845,0.0004468667761315952,-0.1708231424559705,0.004222751763027618,-0.3733233029879562,-0.2990915825497368,0.2612237910441627,0.1796306614025771,-0.39268662860228587,0.26278428259811654,0.26639066840265696,-0.07776975619193575,0.023203627230919052,0.5482247929800804,0.23588899747056388,0.04082675711178992,-0.13207159397747623,0.10274442561758652,-0.13765200403883585,-0.2922149941893392,0.001837062456960628,0.2759746448692217,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0],"frozen_sha256":"a39e22d737d564cef74a3cd0225dc6a374ff48a3886ba1ef75cf1db8a45547b6","spot_10000_10002":{"HEAD":["444.195330","572.172682","526.274760"]},"windows":{"choose":{"seeds":"64000..65023","n":1024,"HEAD-COMBO":{"n":1024,"mean_diff":0.0,"se":0.0,"worst5":[[64000,0.0],[64001,0.0],[64002,0.0],[64003,0.0],[64004,0.0]],"wins":0,"sum_d":0.0,"sum_d2":0.0},"HEAD-qt1":{"n":1024,"mean_diff":0.609605,"se":0.019401,"worst5":[[64101,-2.776983],[64845,-1.163868],[64374,-1.023434],[64795,-0.983171],[64555,-0.946189]],"wins":875,"sum_d":624.236012,"sum_d2":774.822421},"COMBO-qt1":{"n":1024,"mean_diff":0.609605,"se":0.019401,"worst5":[[64101,-2.776983],[64845,-1.163868],[64374,-1.023434],[64795,-0.983171],[64555,-0.946189]],"wins":875,"sum_d":624.236012,"sum_d2":774.822421},"mean_edge":{"HEAD":536.330101,"COMBO":536.330101,"qt1":535.720496}},"report":{"seeds":"10000..12047","n":2048,"HEAD-COMBO":{"n":2048,"mean_diff":0.0,"se":0.0,"worst5":[[10000,0.0],[10001,0.0],[10002,0.0],[10003,0.0],[10004,0.0]],"wins":0,"sum_d":0.0,"sum_d2":0.0},"HEAD-qt1":{"n":2048,"mean_diff":0.597426,"se":0.013387,"worst5":[[10146,-3.552497],[11062,-2.334849],[11808,-1.822738],[10519,-1.782691],[11376,-1.750296]],"wins":1753,"sum_d":1223.52942,"sum_d2":1482.313777},"COMBO-qt1":{"n":2048,"mean_diff":0.597426,"se":0.013387,"worst5":[[10146,-3.552497],[11062,-2.334849],[11808,-1.822738],[10519,-1.782691],[11376,-1.750296]],"wins":1753,"sum_d":1223.52942,"sum_d2":1482.313777},"mean_edge":{"HEAD":533.860243,"COMBO":533.860243,"qt1":533.262816}}},"seconds_per_seed":{"HEAD":0.332,"COMBO":0.3062,"qt1":0.2045}},"sensitivity_run":{"name":"output layer only (18 params), sigma0 0.1, same inputs/head/minibatch/validation","sigma0":0.1,"trained_params":18,"mode":"out","generations":28,"train_wall_seconds":812.9,"final_sigma":0.042999131890384984,"training_curve":[[0,null,0.0],[1,-0.021064,null],[2,-0.084224,null],[3,-0.048661,null],[4,0.032363,0.006748],[5,0.033083,null],[6,0.013297,null],[7,-0.041899,null],[8,-0.034883,0.010785],[9,-0.048606,null],[10,-0.001475,null],[11,0.050983,null],[12,0.030796,0.024089],[13,-0.004995,null],[14,0.034859,null],[15,0.001714,null],[16,-0.025175,0.016445],[17,0.024567,null],[18,0.02061,null],[19,0.037244,null],[20,0.017739,0.02158],[21,0.023555,null],[22,0.033854,null],[23,0.036398,null],[24,-0.028411,-0.003078],[25,0.017101,null],[26,0.018791,null],[27,0.06267,null],[28,0.024063,0.020397]],"validations":[[0,0.0],[4,0.006748],[8,0.010785],[12,0.024089],[16,0.016445],[20,0.02158],[24,-0.003078],[28,0.020397]],"selected_gen":12,"validation_at_selection":0.0240890513331371,"frozen_weights":[-0.03308715437882328,0.019115850566246966,0.10245972698416962,-0.20044514592331217,-0.11875872820943496,0.1290216003856351,-0.14527222071654158,-0.06396047433818225,0.20222153064761011,0.2003670728829884,-0.15300728807037753,-0.06290345623746797,-0.23638187459869753,-0.08212999278197877,0.11290946284469724,-0.030458286873202234,0.2730044903604487,-0.036861204186424884,-0.15130022821980782,-0.0037419099885917287,-0.033613578761160015,-0.17020012938078596,-0.031726493810521235,0.048925986088562905,-0.37306721093906353,0.591815801619824,0.437619631039542,-0.058644035002906054,0.6342674612772654,0.013769546420165382,0.31565985949112985,-0.3329139949245999,0.05288073016774706,0.11034368726969458,-0.4781304645483523,0.28466466717516625,-0.3850506047803147,-0.20569262545552852,0.03546851429151385,0.3235266060051074,-0.12592296237645467,-0.48270439533207793,0.06826401653071691,-0.18195249907387775,0.130462368549171,-0.13346407321006498,0.1089621985516729,-0.3222162923087807,0.13976553228242283,0.3276473327640361,-0.670215854233465,-0.07714572936523113,0.15462095658579564,0.04129788524748191,0.5021970636795855,-0.052912821968877234,0.11306009698072775,0.15617542427566264,-0.15149037187666853,-0.03802302312625276,-0.037999037016099226,0.07603519484410304,-0.02578181716395206,-0.1049950213775521,0.6011853759413393,-0.0203894948748699,0.11107791709293718,-0.446225972893398,-0.27152125106409425,0.29744932269404395,0.021628149714376252,-0.30522730237075,0.347221976600741,0.14133837350075287,-0.1680878418869961,0.32602854708177714,-0.5798644844335422,0.3148637119891221,0.1369887814608869,0.4364392426535371,-0.10824989353072591,-0.3231923928759373,0.14060288599341308,0.0773503383761791,0.18946319485909485,-0.03950669039297138,0.1139346461209294,0.1886890863012107,-0.31486931567140053,-0.0747998780975266,0.02134294516026845,0.0004468667761315952,-0.1708231424559705,0.004222751763027618,-0.3733233029879562,-0.2990915825497368,0.2612237910441627,0.1796306614025771,-0.39268662860228587,0.26278428259811654,0.26639066840265696,-0.07776975619193575,0.023203627230919052,0.5482247929800804,0.23588899747056388,0.04082675711178992,-0.13207159397747623,0.10274442561758652,-0.13765200403883585,-0.2922149941893392,0.001837062456960628,0.2759746448692217,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.022011330316532574,-0.031449308569967235,-0.004041932610952682,0.10484331295085425,0.06640253177036212,-0.03825260622642419,-0.13138953060464184,0.09752314970324683,0.23593596564992447,-0.059257891983701724,-0.10141957419456381,-0.03243897032185564,-0.20903673829326583,0.043056207564645024,-0.020555422898068422,0.0058630489097727945,-0.011434201688423611,-0.07174082176881465],"frozen_sha256":"4e58286842b5db51cde76e407bea3574765c8bd7814f16dfa9ca5619adc9b645","spot_10000_10002":{"HEAD":["444.298460","571.843505","527.315122"]},"windows":{"choose":{"seeds":"64000..65023","n":1024,"HEAD-COMBO":{"n":1024,"mean_diff":0.004681,"se":0.00931,"worst5":[[64574,-1.148795],[64483,-1.139236],[64901,-1.110061],[64505,-1.001936],[64319,-0.989923]],"wins":530,"sum_d":4.792955,"sum_d2":90.811325},"HEAD-qt1":{"n":1024,"mean_diff":0.614286,"se":0.01952,"worst5":[[64101,-2.286604],[64795,-1.394199],[64116,-1.238627],[64761,-1.047374],[64348,-1.044455]],"wins":877,"sum_d":629.028966,"sum_d2":785.563439},"COMBO-qt1":{"n":1024,"mean_diff":0.609605,"se":0.019401,"worst5":[[64101,-2.776983],[64845,-1.163868],[64374,-1.023434],[64795,-0.983171],[64555,-0.946189]],"wins":875,"sum_d":624.236012,"sum_d2":774.822421},"mean_edge":{"HEAD":536.334782,"COMBO":536.330101,"qt1":535.720496}},"report":{"seeds":"10000..12047","n":2048,"HEAD-COMBO":{"n":2048,"mean_diff":0.008144,"se":0.006705,"worst5":[[10401,-1.467938],[11855,-1.454473],[11554,-1.314171],[11693,-1.256091],[11911,-1.253464]],"wins":1048,"sum_d":16.679713,"sum_d2":188.58833},"HEAD-qt1":{"n":2048,"mean_diff":0.605571,"se":0.013779,"worst5":[[10146,-3.646338],[10401,-1.881115],[11808,-1.691534],[10519,-1.649196],[11376,-1.571239]],"wins":1726,"sum_d":1240.209133,"sum_d2":1547.036797},"COMBO-qt1":{"n":2048,"mean_diff":0.597426,"se":0.013387,"worst5":[[10146,-3.552497],[11062,-2.334849],[11808,-1.822738],[10519,-1.782691],[11376,-1.750296]],"wins":1753,"sum_d":1223.52942,"sum_d2":1482.313777},"mean_edge":{"HEAD":533.868387,"COMBO":533.860243,"qt1":533.262816}}},"seconds_per_seed":{"HEAD":0.5367,"COMBO":0.3062,"qt1":0.2045}},"findings":["Gate P: Taylor (gtmcknight)'s unmodified gates.py at 4812a638 reproduced all 12 README values at 6 dp (verbatim in gates.P.gates_output). qt1 (QMRef over blob.bin) on 10000..10002 = 443.101741 / 571.454483 / 525.908441 and COMBO = 444.195330 / 572.172682 / 526.274760, as the brief requires.","Zero-head identity: with output weights and biases 0 the head returns COMBO's clamped fees bit for bit on every callback of seeds 10000..10003 (43,219 callbacks) and the four edges are equal. One callback on seed 10002 differs only in the raw return value (COMBO returns 0.12477 before clamping, the head 0.1); the replica clamps both to 0.1, so what is applied is identical.","As specified (all 138 parameters, sigma0 0.3, popsize 8, rotating 64-seed minibatches at 10000 steps, validation of the CMA mean every 4 generations on 64512..64767): 41 generations in 1180 s of training wall. Every validation checkpoint from gen 4 to gen 40 was negative (-0.40 to -0.61 vs COMBO); best-of-population train fitness ranged -0.23 to -0.73 and the population mean -0.58 to -0.99. The best-validation checkpoint is gen 0 (the untrained head, validation 0.0 by construction), so the frozen as-specified policy is COMBO bit for bit: C-COMBO = 0.000000 on all 1024 choose and all 2048 report seeds; C-qt1 = COMBO-qt1 = +0.609605 se 0.019401 on choose (875/1024 wins) and +0.597426 se 0.013387 on report (1753/2048 wins).","Why the as-specified search does not learn: sigma 0.3 on the 16 output weights (hidden activations of rms about 0.6) gives random fee offsets of about +-1.4 bps on every callback, which cost 0.6 to 1.0 edge per candidate; with popsize 8 in 138 dimensions the selected direction carries far less signal than the random drift of the mean, and sigma only fell from 0.300 to 0.226 in 41 generations. The minibatch fitness noise (se about 0.1 at 64 seeds for such perturbations) is of the same order as the spread between candidates.","Sensitivity run (identical head, inputs, scales, minibatch, validation and selection rule; only the 18 output-layer parameters trained, sigma0 0.1; 28 generations, 813 s): the validation rose to +0.024 at gen 12 (selected), then +0.016 / +0.022 / -0.003 / +0.020 at gens 16..28. Frozen gen 12, scored once: C-COMBO +0.004681 se 0.009310 on choose (530/1024 wins) and +0.008144 se 0.006705 on report (1048/2048 wins), about 1.2 se from zero: no measurable gain. C-qt1 +0.614286 se 0.019520 (choose) and +0.605571 se 0.013779 (report). Its fee offsets are small (pre-activations of sd about 0.2, offsets about +-0.5 bps), and its left tail vs COMBO reaches -1.47 (seed 10401).","COMBO-qt1 recomputed here on 10000..12047 (+0.597426 se 0.013387, 1753 wins, worst 10146 -3.552497) agrees with the COMBO item's published +0.597391 se 0.013388 (1753 wins, same worst5 seeds) to 4e-5; mean edges agree to 2e-5 (533.860243 vs 533.860222; qt1 533.262816 vs 533.262831). The 6 dp spot values on 10000..10002 are identical. The residual is consistent with aarch64 vs x86_64 libm rounding over 10000 chaotic steps.","On the choose window 64000..65023 (fresh here) COMBO-qt1 is +0.609605 se 0.019401 with 875/1024 wins, worst 64101 at -2.776983; this is an independent confirmation of the COMBO item's +0.637899 on 60000..61023 and +0.597391 on report.","Cost: the head adds 14 features and a 14x8x2 MLP per callback. Single-process replica time 0.0112 s per 1000 steps against COMBO's 0.0066 (about 70% more); loaded 10-worker-on-4-CPU figures were 0.54 s per seed (head, active) vs 0.31 (COMBO) and 0.20 (qt1). On a contract this is about 130 multiply-adds and 10 tanh per callback on top of COMBO; no gas estimate was made.","Both heads leave the arb-or-retail call untouched: pi and the two components enter only as inputs, and the trained output-only head's corrections are within noise. Within this item's budget a residual fee head on COMBO under real observables did not recover any of the 2.7 that the true flag is worth."],"limitations":["The as-specified run was stopped at 1180 s (19.7 min) of training wall, short of the 35 min cap: this machine has 4 CPUs (cgroup cpu.max 400000/100000) and took about 25 s per generation (38 s with validation), the validation was flat and negative from gen 4 to gen 40, and the remaining machine time was needed for the sensitivity run, the 6 full-window scorings and the write-up. 41 generations is a small budget for 138 parameters; a longer run would most likely keep shrinking sigma before it could help.","The sensitivity run departs from the brief in two settings (sigma0 0.1 instead of 0.3; only the output layer trained, hidden weights fixed at their random init) and had 28 generations. It is reported beside the as-specified run, not in its place. Its +0.008144 se 0.006705 on report is not a gain at any usual threshold.","Validation on 256 seeds has an se of about 0.01-0.02 for perturbations this small, so picking the best of the validation checkpoints carries a positive selection bias of that order; the report window (run once after freezing) is the unbiased figure.","Interpretation choices the brief left open: 'belief mean minus mid' uses the log midpoint of the no-arb band of the quotes in force at the trade; 'own trade size over the size law' is ln(amountY / (20 e^-0.72)), the lognormal law's median at its midpoint mean, in log units; when the filter carries one component the second component's inputs are that component's mean and sd with weight 0; standardization subtracts a fixed center and divides by a fixed sd, both from all 173,138 callbacks of COMBO on 64000..64015. pi before any first trade is 0.","Replica figures only: no engine run, no contract, no Gate F. Taylor's replica uses different RNG streams from the Rust engine, so these are paired replica differences, not board scores.","This machine had no python3; a standalone CPython 3.12.16 for linux-aarch64 was installed with uv under scratch/ (python-build-standalone). Per-seed edges differ from the x86_64 publications at the 1e-5 level, consistent with libm differences, while all 6 dp gate values matched.","Work state lived under scratch/ during the run as the brief asks (scratch/w code and scratch/w/state checkpoints, scratch/src the clone, scratch/art the artifacts) and was moved to test/scratch/ before finishing, since only this item's artifact paths may remain in the repository. The source bundle in this result plus the regeneration commands rebuild everything; frozen_weights and the standardization center/scale are given in full.","sep-CMA-ES was written from the standard formulation (Hansen's tutorial for CMA-ES, Ros & Hansen 2008 (N+2)/3 rate scaling for the diagonal variant); no library was available under the stdlib constraint, and the implementation was not compared against a reference library.","In the as-specified run C-COMBO is identically 0, so its worst5 lists the first five seeds at 0.0 and wins is 0 (no seed is strictly positive). The sequence gen 0 validation of 0.0 is by construction, not a measurement.","Gate Q was proven on two seeds by callback replay with random arb/p/s2/j and random init arguments (bit-identical fees); the policy code reads none of them by construction (combo.py and head.py ignore the oracle arguments), but only those two seeds were replayed."],"source_bundle":{"encoding":"base64(zlib(json({path: text})))","files":["head.py","sepcma.py","common.py","prep.py","train.py","score.py","build_result.py"],"sha256_of_zlib":"4e03ad17ea818f9dff5afacdd63a79a10799767752cba7cfaffe0dcde65fb4f2","data":"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"},"regeneration":["git clone https://github.com/gtmcknight/amm-sim scratch/src/amm-sim && git -C scratch/src/amm-sim checkout 4812a6382311624e1506aca131ca35c378f930d9","curl -sS --retry 5 -L https://api.imd.fun/artifacts/<h> -o scratch/art/<h>  (qt1, TC, COMBO, QMS1) and sha256sum -c","curl -LsSf --retry 5 https://raw.githubusercontent.com/identity-md-launches/goal-3170-beat-network-s-amm-entry/31ff8ed128b463cde10ea0a71d354f8c54d4464a/reference/qt1/{blob.bin,qmref.py} -o scratch/w/...","python3 -I: extract the first ```python fence after '## Code' of TC -> scratch/w/twocomp.py and of COMBO -> scratch/w/combo.py (plus trailing newline); check sha256 9d3c1833.. and c5b711a0..","decode source_bundle.data into scratch/w/; cd scratch/src/amm-sim/research/solved && python3 -I gates.py","cd scratch/w && python3 -I prep.py 10 && python3 -I train.py 10 420 2100 (repeat until it prints \"cap reached\") && python3 -I score.py 10 (twice: first call freezes) && python3 -I build_result.py OUT","sensitivity: python3 -I train.py 10 420 1300 0.1 out out ; python3 -I score.py 10 out (twice)"]}