{ "bench_version": 1, "track": "dynamics", "model": "rnn_small", "metric_direction": "lower is better", "n_seeds": 8, "baseline": { "best_cfg": { "lr": 0.01, "weight_decay": 0.0, "epochs": 30, "batch": 64 }, "sweep": [ { "cfg": { "lr": 0.001, "weight_decay": 0.0, "epochs": 30, "batch": 64 }, "mean": 0.0008596329862484708 }, { "cfg": { "lr": 0.001, "weight_decay": 0.0001, "epochs": 30, "batch": 64 }, "mean": 0.0008667934744153172 }, { "cfg": { "lr": 0.003, "weight_decay": 0.0, "epochs": 30, "batch": 64 }, "mean": 0.0003459557265159674 }, { "cfg": { "lr": 0.003, "weight_decay": 0.0001, "epochs": 30, "batch": 64 }, "mean": 0.0004122954633203335 }, { "cfg": { "lr": 0.01, "weight_decay": 0.0, "epochs": 30, "batch": 64 }, "mean": 3.945273101635394e-05 }, { "cfg": { "lr": 0.01, "weight_decay": 0.0001, "epochs": 30, "batch": 64 }, "mean": 0.00015876943143666722 } ], "full": { "mean": 4.022148073090648e-05, "std": 1.0539636376553464e-05, "per_seed": [ 2.7519910872797482e-05, 5.704866998712532e-05, 2.5148139684461057e-05, 4.8094203521031886e-05, 3.4156204492319375e-05, 3.871386797982268e-05, 4.010241536889225e-05, 5.098843394080177e-05 ], "n": 8 } }, "idea": { "mean": 8.502966829837533e-05, "std": 3.866288220789e-05, "per_seed": [ 3.906466736225411e-05, 0.00010664308501873165, 6.796212983317673e-05, 6.528497760882601e-05, 6.44689062028192e-05, 0.0001517514610895887, 0.00013560302613768727, 4.945909313391894e-05 ], "n": 8 }, "comparison": { "delta_mean": 4.480818756746885e-05, "idea_wins": 1, "n_pairs": 8, "per_seed_diffs": [ 1.1544756489456631e-05, 4.959441503160633e-05, 4.2813990148715675e-05, 1.7190774087794125e-05, 3.0312701710499823e-05, 0.00011303759310976602, 9.550061076879501e-05, -1.5293408068828285e-06 ], "p_value": 0.0156, "mde": 3.385917105994864e-05, "mde_rel_pct": 84.18181142180329, "verdict": "idea worse (significant)", "system_worked": false }, "mechanism_signature": { "mass_diffusion": { "predicted_max_mode_amplitude": 0.19999999999999996, "observed_mean_test_input_norm_ratio": 0.6925333812832832, "observed_per_seed": [ 0.6863239407539368, 0.7053196430206299, 0.6866878867149353, 0.6967299580574036, 0.6953513622283936, 0.6847705245018005, 0.6918107271194458, 0.6932730078697205 ], "lambda_max": 2.0, "confirmed": false }, "idea_sweep": [ { "cfg": { "lr": 0.01, "weight_decay": 0.0, "epochs": 30, "batch": 64, "alpha": 1.0, "gamma": 0.6 }, "result": { "mean": 8.502966829837533e-05, "std": 3.866288220789e-05, "per_seed": [ 3.906466736225411e-05, 0.00010664308501873165, 6.796212983317673e-05, 6.528497760882601e-05, 6.44689062028192e-05, 0.0001517514610895887, 0.00013560302613768727, 4.945909313391894e-05 ], "n": 8 } }, { "cfg": { "lr": 0.01, "weight_decay": 0.0, "epochs": 30, "batch": 64, "alpha": 1.0, "gamma": 0.8 }, "result": { "mean": 0.000431537780968938, "std": 0.0003206459743519331, "per_seed": [ 0.00012159255857113749, 0.0005679481546394527, 0.0004245344316586852, 0.0001826822553994134, 0.00037008809158578515, 0.00020022669923491776, 0.0011974757071584463, 0.00038775434950366616 ], "n": 8 } }, { "cfg": { "lr": 0.01, "weight_decay": 0.0, "epochs": 30, "batch": 64, "alpha": 1.0, "gamma": 0.95 }, "result": { "mean": 0.008008880191482604, "std": 0.0025785752995045367, "per_seed": [ 0.007743558846414089, 0.00685552041977644, 0.008397368714213371, 0.007993439212441444, 0.003967979457229376, 0.008484840393066406, 0.01381105836480856, 0.006817276123911142 ], "n": 8 } } ] }, "notes": "Dynamics is structurally matched: controlled pendulum rollout and stability-sensitive recurrent prediction; diffusion acts on the three physical channels at every timestep." }