{ "bench_version": 1, "track": "dynamics", "model": "rnn_small", "metric_direction": "lower is better", "n_seeds": 8, "baseline": { "best_cfg": { "lr": 0.003, "weight_decay": 0.0 }, "sweep": [ { "cfg": { "lr": 0.001, "weight_decay": 0.0 }, "mean": 0.004509380669333041 }, { "cfg": { "lr": 0.003, "weight_decay": 0.0 }, "mean": 0.0036578410072252154 }, { "cfg": { "lr": 0.01, "weight_decay": 0.0 }, "mean": 0.0037718922831118107 } ], "full": { "mean": 0.0034481814655009657, "std": 0.0004415458604303511, "per_seed": [ 0.003405338153243065, 0.003420044668018818, 0.003606032580137253, 0.004199948627501726, 0.003293032059445977, 0.002657515462487936, 0.003106778720393777, 0.003896761452779174 ], "n": 8 } }, "idea": { "mean": 0.0011822001833934337, "std": 0.00033321397960779696, "per_seed": [ 0.0010503326775506139, 0.0014236074639484286, 0.0007342637982219458, 0.0010695835808292031, 0.0010554390028119087, 0.00107160909101367, 0.0019350601360201836, 0.0011177057167515159 ], "n": 8 }, "comparison": { "delta_mean": -0.002265981282107532, "idea_wins": 8, "n_pairs": 8, "per_seed_diffs": [ -0.002355005475692451, -0.0019964372040703893, -0.002871768781915307, -0.003130365046672523, -0.0022375930566340685, -0.001585906371474266, -0.0011717185843735933, -0.002779055736027658 ], "p_value": 0.0081, "mde": 0.000558113615628125, "mde_rel_pct": 16.185737937868073, "verdict": "idea better (significant)", "system_worked": true }, "idea_sweep": [ { "cfg": { "lr": 0.003, "weight_decay": 0.0 }, "mean": 0.0011822001833934337, "per_seed": [ 0.0010503326775506139, 0.0014236074639484286, 0.0007342637982219458, 0.0010695835808292031, 0.0010554390028119087, 0.00107160909101367, 0.0019350601360201836, 0.0011177057167515159 ] }, { "cfg": { "lr": 0.001, "weight_decay": 0.0 }, "mean": 0.002837543041096069, "per_seed": [ 0.0023743249475955963, 0.002447130624204874, 0.0012939035659655929, 0.0055610244162380695, 0.0027316580526530743, 0.0028867414221167564, 0.0028890997637063265, 0.0025164615362882614 ] }, { "cfg": { "lr": 0.01, "weight_decay": 0.0 }, "mean": 0.0019820211964542978, "per_seed": [ 0.0011709327809512615, 0.0052402568981051445, 0.0006008912459947169, 0.0012470174115151167, 0.0014493301277980208, 0.0010334531543776393, 0.0033070051576942205, 0.0018072827951982617 ] } ], "custom_track": null, "mechanism_signature": { "baseline_imag_max_mean": 0.0, "idea_imag_max_mean": 0.0, "prediction": "symmetric S and SPD H yield real A eigenvalues", "confirmed": true } }