{ "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": 0.25371627509593964 }, { "cfg": { "lr": 0.003 }, "mean": 0.2935705631971359 }, { "cfg": { "lr": 0.006 }, "mean": 0.39029930531978607 } ], "full": { "mean": 0.26352048851549625, "std": 0.029814906699763923, "per_seed": [ 0.28988832235336304, 0.23809197545051575, 0.21863380074501038, 0.2682510018348694, 0.2573082745075226, 0.26551553606987, 0.3222951292991638, 0.2481798678636551 ], "n": 8 } }, "idea": { "mean": 0.2635671868920326, "std": 0.029872832584909817, "per_seed": [ 0.2901400327682495, 0.23809197545051575, 0.21863380074501038, 0.2682510018348694, 0.25730830430984497, 0.26551553606987, 0.3224169909954071, 0.2481798529624939 ], "n": 8, "selected_cfg": { "lr": 0.001 }, "sweep": [ { "cfg": { "lr": 0.001 }, "mean": 0.2635671868920326, "per_seed": [ 0.2901400327682495, 0.23809197545051575, 0.21863380074501038, 0.2682510018348694, 0.25730830430984497, 0.26551553606987, 0.3224169909954071, 0.2481798529624939 ] }, { "cfg": { "lr": 0.003 }, "mean": 0.307208513841033, "per_seed": [ 0.38006386160850525, 0.22656194865703583, 0.23053526878356934, 0.3371211290359497, 0.2561066746711731, 0.34927764534950256, 0.30410173535346985, 0.3738998472690582 ] }, { "cfg": { "lr": 0.006 }, "mean": 0.4231664314866066, "per_seed": [ 0.28377190232276917, 0.34400641918182373, 0.41005638241767883, 0.5222185850143433, 0.3521057069301605, 0.5003604292869568, 0.3952743709087372, 0.5775376558303833 ] } ] }, "comparison": { "delta_mean": 4.6698376536369324e-05, "idea_wins": 1, "n_pairs": 8, "per_seed_diffs": [ 0.0002517104148864746, 0.0, 0.0, 0.0, 2.9802322387695312e-08, 0.0, 0.00012186169624328613, -1.4901161193847656e-08 ], "p_value": 0.25605, "mde": 7.790313751068211e-05, "mde_rel_pct": 0.029562459431347415, "verdict": "no measurable effect", "system_worked": false }, "structural_check": { "separator_size": 2, "components": [ 1, 1, 1, 1 ], "anti_adjacent_components": true, "mask_formula_exact": true }, "custom_track": null, "protocol_notes": "8 paired seeds; baseline sweep and idea sweep share lr union.", "mechanism_signature": { "predicted": { "sparse_allowed_fraction": "< 1.0", "anti_adjacent_components": true }, "observed_mean_allowed_fraction": 1.0, "observed_mean_components": 0.0, "observed_mean_separator_size": 2.0, "confirmed": false } }