{ "bench_version": 1, "track": "sequence", "model": "transformer_tiny", "metric_direction": "lower is better", "n_seeds": 8, "baseline": { "best_cfg": { "lr": 0.001 }, "sweep": [ { "cfg": { "lr": 0.001 }, "mean": 0.9495677947998047 }, { "cfg": { "lr": 0.003 }, "mean": 1.5083084404468536 }, { "cfg": { "lr": 0.01 }, "mean": 3.8293468952178955 } ], "full": { "mean": 1.006878174841404, "std": 0.1713716806448823, "per_seed": [ 0.9911782145500183, 0.7554877996444702, 1.2632131576538086, 0.7883920073509216, 1.1533313989639282, 0.9415301084518433, 1.1843085289001465, 0.977584183216095 ], "n": 8 }, "idea_grid": [ { "cfg": { "lr": 0.001 }, "mean": 1.008343055844307 }, { "cfg": { "lr": 0.003 }, "mean": 1.4822083115577698 }, { "cfg": { "lr": 0.01 }, "mean": 4.900031566619873 } ] }, "idea": { "mean": 1.008343055844307, "std": 0.1720253375372032, "per_seed": [ 0.9909452795982361, 0.7562634944915771, 1.2659306526184082, 0.7895816564559937, 1.153704047203064, 0.9436419606208801, 1.1885106563568115, 0.9781666994094849 ], "n": 8 }, "comparison": { "delta_mean": 0.0014648810029029846, "idea_wins": 1, "n_pairs": 8, "per_seed_diffs": [ -0.00023293495178222656, 0.0007756948471069336, 0.0027174949645996094, 0.0011896491050720215, 0.0003726482391357422, 0.0021118521690368652, 0.004202127456665039, 0.0005825161933898926 ], "p_value": 0.01485, "mde": 0.0012193019578819895, "mde_rel_pct": 0.12109726760876957, "verdict": "idea worse (significant)", "system_worked": false }, "mechanism_signature": { "mean_hidden": 0.49465957656502724, "saturated_fraction": 0.0, "projection_alignment": 0.020752153369809595, "per_seed": [ { "mean_hidden": 0.4961187243461609, "saturated_fraction": 0.0, "projection_alignment": 0.014636302551895108, "device": "cuda" }, { "mean_hidden": 0.5012550354003906, "saturated_fraction": 0.0, "projection_alignment": 0.022799921693492292, "device": "cuda" }, { "mean_hidden": 0.4911428391933441, "saturated_fraction": 0.0, "projection_alignment": 0.02356167507297876, "device": "cuda" }, { "mean_hidden": 0.4819362759590149, "saturated_fraction": 0.0, "projection_alignment": 0.02155706124715023, "device": "cuda" }, { "mean_hidden": 0.5072823166847229, "saturated_fraction": 0.0, "projection_alignment": 0.022357459330866497, "device": "cuda" }, { "mean_hidden": 0.4996195137500763, "saturated_fraction": 0.0, "projection_alignment": 0.025012486868765855, "device": "cuda" }, { "mean_hidden": 0.4985620379447937, "saturated_fraction": 0.0, "projection_alignment": 0.0183842962110037, "device": "cuda" }, { "mean_hidden": 0.4813598692417145, "saturated_fraction": 0.0, "projection_alignment": 0.017708023982324326, "device": "cuda" } ], "predicted_projection_alignment_positive": true, "confirmed": true }, "protocol_notes": { "architecture_match": "identical sigmoid-input transformer; only inp training rule differs", "track_justification": "sequence forecast contains ordered multi-token temporal windows", "budget": { "n_train": 400, "n_test": 400, "epochs": 5, "batch": 128, "lr_union": [ 0.001, 0.003, 0.01 ] }, "custom_track": null } }