{ "model_name": "policy-bc-v6", "model_version": "6.0.0", "created_at": "2026-08-09T15:09:32Z", "seconds": 331.1, "compute": { "device": "cuda", "framework": "pytorch", "parameters": 761042, "d_model": 160, "num_layers": 2, "epochs_requested": 4, "epochs_ran": 4, "batch_size": 256, "learning_rate": 0.0002, "dropout": 0.2, "max_context": 64, "max_samples_per_battle": 8, "global_dim": 62, "threat_dim": 14, "card_conditioned_placement": true, "placement_mode": "heatmap", "reaction_weight": 3.0, "reaction_repeats": 2, "reaction_seconds": 5.0, "max_battles": 10000, "hide_opponent_deck": true, "hide_opponent_prob": 0.0, "loss_kwargs": { "zone_weight": 0.9, "xy_weight": 0.0, "tile_weight": 0.35, "slot_weight": 2.2 } }, "data": { "battles_total": 10000, "min_card_plays": 12, "train_samples": 84403, "val_samples": 11641, "test_samples": 11650, "vocab_size": 126, "splits": [ { "split": "train", "battles": 7000, "team_win_rate": 0.508, "mean_events": 64.83528571428572 }, { "split": "val", "battles": 1500, "team_win_rate": 0.5113333333333333, "mean_events": 64.53133333333334 }, { "split": "test", "battles": 1500, "team_win_rate": 0.5133333333333333, "mean_events": 64.56666666666666 } ] }, "baselines": { "frequency": { "name": "online_frequency", "slot_top1": 0.12149379408387555, "slot_top3": 0.3968615014489584, "n": 91445 }, "cycle": { "name": "approx_cycle", "slot_top1": 0.20097326261687354, "slot_top3": 0.622538137678386, "n": 91445 }, "chance_slot_top1": 0.125 }, "val": { "loss": 6.483570917792942, "slot_top1": 0.3875096641182029, "slot_top3": 0.8545657589554162, "type_acc": 0.9978524181771325, "zone_acc": 0.3104544283137188, "xy_mae": 6228.556309595396, "tile_acc": 0.016922944764195515, "tile_class_acc": 0.08890988746671248, "timing_mae": 1.9018329639695333 }, "test": { "loss": 6.47210955619812, "slot_top1": 0.3823175965665236, "slot_top3": 0.8501287553648069, "type_acc": 0.9984549356223176, "zone_acc": 0.3226609442060086, "xy_mae": 6196.98232832618, "tile_acc": 0.018969957081545065, "tile_class_acc": 0.08969957081545064, "timing_mae": 1.8894893047021695, "n": 11650 }, "rollouts": { "available": true, "n": 96, "mean_score_real": 0.8968319623772025, "mean_score_policy": 0.7438187405476014, "mean_score_easy": 8.929701065876548e-05, "mean_score_medium": 5.385495328080267e-05, "policy_vs_easy_lift": 0.7437294435369427, "policy_vs_medium_lift": 0.7437648855943206, "policy_gap_to_real": 0.1530132218296011 }, "rollout_hist": { "real": [ 0.9914199105707799, 0.9213285278520198, 0.9864256284025176, 0.9596962914004088, 0.9909543136288466, 0.9346011106316088, 0.9846241208818992, 0.9910155895720785, 0.9909409586897014, 0.9971379465091894, 0.9045362580382575, 0.9902337718110911, 0.9411284759015994, 0.9947632711382272, 0.45477666200616657, 0.9682247757787386, 0.9833559244324058, 0.7870882546583803, 0.9944497630319628, 0.9784090414691132, 0.9752477957113082, 0.7095156272070818, 0.8451885311487122, 0.9891948552034994, 0.9048954359969845, 0.8561801207255936, 0.8268967826870538, 0.971013657730897, 0.5571851186970479, 0.9129511100289472, 0.9756175953973448, 0.9887015966425217, 0.9417805383520914, 0.7021996281808998, 0.9945708057803693, 0.9382384169018493, 0.6053837343351484, 0.9431198367647646, 0.8437708013052766, 0.9308179730069995, 0.9925366777041627, 0.77195643201386, 0.9733124777668413, 0.9700692360559999, 0.9528132493420862, 0.9303390743484702, 0.8972589110142145, 0.960465258673978, 0.8488224873070436, 0.9311121800592929, 0.7233844016791149, 0.9856948459104629, 0.9902821407771816, 0.9904584554573554, 0.9621344058128746, 0.7040764507661534, 0.9708516727628662, 0.9216731779083819, 0.9423112022191585, 0.9716664035939605, 0.970880291597337, 0.8887170621592994, 0.7441254343592008, 0.938370034813929, 0.17496304172052837, 0.9855177108656793, 0.7382896913777495, 0.8539466613031708, 0.9199300484378768, 0.96754986419661, 0.977583321574543, 0.9457261329334532, 0.9586336479824445, 0.9437816715192071, 0.7948743356335347, 0.9564551377643629, 0.9906281363462882, 0.6379314895332738, 0.9887336643977898, 0.9828233926694712, 0.7138018707770762, 0.9906539822224727, 0.9931189588485753, 0.24175951839898568, 0.697833575640667, 0.8716081242644155, 0.983614238362878, 0.9902692137828051, 0.9677243866323615, 0.9760770903678558, 0.980556910714501, 0.9912746556729201, 0.9114675570924785, 0.9704789153614597, 0.9410037526496032, 0.9663651668537864 ], "policy": [ 0.9863360472476366, 0.9943374227118358, 0.0004677857497508213, 0.9699380666439618, 0.8606273412481853, 0.9977262598075136, 0.8726530427836146, 0.3611750880449733, 0.00476142223489531, 0.3482871170836158, 0.15193094880285762, 0.6439517476220078, 0.9983058802438187, 0.9976796310592045, 0.8104998830671307, 0.9946363070564446, 0.0005225958024369896, 0.0007045209730632047, 0.004165435860188649, 0.9960615677676643, 0.9940111149026267, 0.9323398969926934, 0.9282608606199847, 0.7207665321307493, 0.19890393878067536, 0.991596889846958, 0.9882370385581647, 0.9979397584268357, 0.9647214303742713, 0.9800539970285631, 0.9981222633416408, 0.8789703115022949, 0.963580082488875, 0.9892625686883584, 0.0003247627789563787, 0.9866626473335023, 0.9969253779029952, 0.9985467693829253, 0.5874882938535032, 0.9824542234580236, 0.000638966549304227, 0.8304593150400308, 0.0005516469799002042, 0.9979096712913208, 0.9157373750905061, 0.9973584147612101, 0.9995465564614595, 0.9935851169397428, 0.00036286463159446706, 0.9984280324492449, 0.9203088417135079, 0.998351444500131, 0.9859022700056904, 0.9953760500915042, 0.39664023522686115, 0.7614517422918288, 0.6717467618814191, 0.5973156599598168, 0.0006390795678280287, 0.9950111319243001, 0.000982293731037024, 0.9959863470141502, 0.7785895344817017, 0.9356514569943785, 0.9961054242532906, 0.0007016365574931289, 0.7649846409920237, 0.9903256539530703, 0.9958175755105507, 0.9577648648958276, 0.9063805991548887, 0.8929746633052612, 0.918387117764573, 0.9898505125693701, 0.9978134740428171, 0.6260999101399991, 0.9353984061269307, 0.0802699125180754, 0.9986043845448117, 0.9950829393333444, 0.9775647970965002, 0.2975903143683345, 0.9964744296917047, 0.9952838361038479, 0.9989169550645167, 0.8401926410973319, 0.9961961038152118, 0.5021173156927884, 0.9974190493203179, 0.9991671890745197, 0.9336488382434922, 0.9992027510737612, 0.9859265544498638, 0.0003759301027725378, 0.99184001057354, 0.005652981357043726 ], "easy": [ 3.118001990446352e-05, 0.00025407313784341236, 5.015392519869835e-05, 0.00010659288447946054, 6.396059256807686e-05, 4.820454134589501e-05, 6.314775838168273e-05, 5.714855454815239e-05, 9.308398220886897e-05, 5.027060616993144e-05, 7.438625326098098e-05, 5.312518925094164e-05, 0.00014477756158617086, 2.8391413938872674e-05, 0.00012848719951731623, 7.100875856912836e-05, 3.7688749708833916e-05, 6.345162667467293e-05, 5.067147026209936e-05, 3.8222021013425315e-05, 0.00010696621625648411, 0.00017086488222109526, 4.0435758969985216e-05, 2.993272975776943e-05, 5.079173910837969e-05, 0.00010219204635300877, 0.00010297817072227527, 4.351932205973343e-05, 9.989556335215496e-05, 7.928842190795519e-05, 3.0427312573190416e-05, 0.00010342846625897246, 2.9883812271367387e-05, 6.0646970033297565e-05, 3.241163363319785e-05, 0.00016660684943975651, 3.3477186544638154e-05, 9.580409300624812e-05, 4.6187491065740294e-05, 3.491756744090898e-05, 0.00017726470410908143, 2.7923151925940964e-05, 5.169701131002513e-05, 5.939458562463693e-05, 1.9601500355003636e-05, 0.00011152853286287011, 7.757701929658749e-05, 0.00022502639096245357, 2.4585800734994836e-05, 8.482567320201145e-05, 0.00013612613513339726, 3.223870722416973e-05, 5.5350415641388634e-05, 5.352131072938612e-05, 5.380298982053329e-05, 7.940852769574972e-05, 0.00013029561535744214, 2.8047737276709828e-05, 0.00011309808773199404, 2.5246685133240958e-05, 2.3615783603921476e-05, 4.811274408161898e-05, 5.220885746410906e-05, 7.011855851145075e-05, 3.3667268243875884e-05, 4.668964402984376e-05, 0.0001312465317226628, 9.278307821340141e-05, 4.314044932367074e-05, 4.46713327328673e-05, 4.69287980707704e-05, 3.3696698177572644e-05, 0.0008167413717239745, 0.00022470323455870607, 2.870799818228911e-05, 1.492790144078928e-05, 7.820853720769407e-05, 4.3564712019947707e-05, 6.782898298097112e-05, 2.313260716876043e-05, 4.123585821526773e-05, 3.236594220525989e-05, 3.118060954572131e-05, 0.00021566215284332722, 0.000451087926317191, 9.97398708788775e-05, 0.0003333877751600724, 1.8616648354668692e-05, 0.0001495522510922263, 4.581117567037092e-05, 5.216048417278869e-05, 2.0662844979088933e-05, 0.0002152462079284583, 5.256679856520249e-05, 0.0002610904772226459, 5.220784909656001e-05 ], "medium": [ 2.799407426990092e-05, 4.595864274109262e-05, 1.4862870564419064e-05, 4.671830501916275e-05, 9.152593866616083e-05, 3.6962866100475555e-05, 6.793407181504397e-05, 7.15014208494103e-05, 2.6345645475954004e-05, 0.000115496885929524, 2.681643695603962e-05, 7.821707998137082e-05, 0.0002864124387365704, 4.47899419120874e-05, 3.761593482024893e-05, 4.743863435509256e-05, 3.692854214306458e-05, 7.679621103257403e-05, 2.8203288797328717e-05, 3.3418657861295086e-05, 2.8631678388044206e-05, 1.600641082917019e-05, 2.37472651488322e-05, 1.9716182699229093e-05, 4.3457417053417037e-05, 8.983424400450925e-05, 2.5231818189012e-05, 3.1767273740726866e-05, 0.00011063439184870904, 4.1362382163480537e-05, 2.0962350485752722e-05, 0.00013768623012516047, 2.745804166757145e-05, 2.0853585345423595e-05, 2.2436185817557274e-05, 2.7959719030084243e-05, 2.2264754193351775e-05, 7.126097851669971e-05, 3.2310758465695923e-05, 5.8836734862289025e-05, 8.300321534255312e-05, 9.72041193154601e-05, 7.191542984801195e-05, 1.4341868529203698e-05, 2.935988225059291e-05, 3.355805242222579e-05, 5.559030105570059e-05, 4.023226563319399e-05, 2.184729983877433e-05, 2.932587917780915e-05, 2.1690530314365295e-05, 8.249873432047111e-05, 3.4725775869457816e-05, 2.6468096979156315e-05, 0.0004999600273939336, 2.5859445883191796e-05, 0.0001609404446894134, 5.1248565194647785e-05, 8.760127434520347e-05, 6.723030783393998e-05, 4.312178656180128e-05, 9.010142170104252e-06, 3.215651221123082e-05, 8.605654897070793e-05, 5.8027343778464644e-05, 6.896953525649018e-05, 5.3648122304365324e-05, 1.813925989791023e-05, 1.5819812819602385e-05, 2.9558219130819444e-05, 2.5497808038995696e-05, 9.286111126926868e-06, 3.672273703824533e-05, 2.5788242681430118e-05, 2.8631401636580975e-05, 5.978611823619513e-05, 7.706862036631771e-05, 3.0310729814088906e-05, 2.7148788582005395e-05, 2.0217302282792686e-05, 1.2341950347642164e-05, 9.667914992764526e-05, 7.878881100307715e-05, 8.76002111580136e-05, 2.6605952848518624e-05, 1.612750735035104e-05, 3.093239410246688e-05, 2.3232721527750006e-05, 6.582727348100228e-05, 0.0001205902097538809, 2.0518972426383932e-05, 3.317557637964732e-05, 6.579289248221241e-05, 3.2456696459942346e-05, 2.0312926120638953e-05, 3.516931984396925e-05 ] }, "history": [ { "epoch": 1, "train_loss": 7.024750387307369, "val_loss": 6.807787179946899, "val_slot_top1": 0.35452280731895885, "val_slot_top3": 0.8244996134352719, "val_type_acc": 0.9978524181771325, "val_zone_acc": 0.2695644704063225, "val_xy_mae": 6832.468113778885, "val_tile_acc": 0.017008848037110214, "val_tile_class_acc": 0.06846490851301434, "val_timing_mae": 1.9553172020821645, "val_n": 11641, "lr": 0.00017071067811865476 }, { "epoch": 2, "train_loss": 6.3821511629856, "val_loss": 6.5958424443783965, "val_slot_top1": 0.369555880079031, "val_slot_top3": 0.8430547203848466, "val_type_acc": 0.9978524181771325, "val_zone_acc": 0.2980843570140022, "val_xy_mae": 6358.649460956963, "val_tile_acc": 0.01718065458293961, "val_tile_class_acc": 0.08135039945021906, "val_timing_mae": 1.9234734921815437, "val_n": 11641, "lr": 0.0001 }, { "epoch": 3, "train_loss": 6.156989932782722, "val_loss": 6.507258415222168, "val_slot_top1": 0.38570569538699423, "val_slot_top3": 0.8531054033158664, "val_type_acc": 0.9978524181771325, "val_zone_acc": 0.30581565157632506, "val_xy_mae": 6256.540358431406, "val_tile_acc": 0.016407525126707328, "val_tile_class_acc": 0.08684820891675973, "val_timing_mae": 1.908578866618307, "val_n": 11641, "lr": 2.9289321881345254e-05 }, { "epoch": 4, "train_loss": 6.070085696018103, "val_loss": 6.483570917792942, "val_slot_top1": 0.3875096641182029, "val_slot_top3": 0.8545657589554162, "val_type_acc": 0.9978524181771325, "val_zone_acc": 0.3104544283137188, "val_xy_mae": 6228.556309595396, "val_tile_acc": 0.016922944764195515, "val_tile_class_acc": 0.08890988746671248, "val_timing_mae": 1.9018329639695333, "val_n": 11641, "lr": 0.0 } ], "checkpoint": "models/policy_bc_v6_tune/best_model.pt", "lessons": [ "v6 keeps the v4 threat-conditioned trunk but predicts a card-conditioned 18\u00d732 placement heatmap instead of a single XY point.", "Tile cross-entropy models multimodal legal placements; expected XY is retained for compatibility while rollouts can sample tiles.", "Training can hide unrevealed opponent cards so the deck encoder is less dependent on an offline oracle deck.", "This is an offline action-prior experiment; it does not restore missing arena state or establish live-play readiness." ], "live_play_readiness": { "ready_for_live_smoke_test": false, "checks": { "beats_baselines": true, "slot_floor": true, "zone_floor": true, "tile_or_zone": true, "rollout_beats_easy": true, "rollout_not_collapsed": true, "rollout_near_real": true }, "rationale": "Offline-only experiment; do not run live yet." } }