1{
  2  "bench_version": 1,
  3  "track": "potential_transport",
  4  "model": "mlp_tiny",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.003,
 10      "epochs": 24
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.001,
 16          "epochs": 24
 17        },
 18        "mean": 1.2874779403209686
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.003,
 23          "epochs": 24
 24        },
 25        "mean": 1.2407947480678558
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.009,
 30          "epochs": 24
 31        },
 32        "mean": 1.2557097375392914
 33      }
 34    ],
 35    "full": {
 36      "mean": 1.1834656298160553,
 37      "std": 0.11952528701329043,
 38      "per_seed": [
 39        1.110816478729248,
 40        1.1996243000030518,
 41        1.3317530155181885,
 42        1.320985198020935,
 43        1.2276966571807861,
 44        1.0896412134170532,
 45        0.9505289793014526,
 46        1.236679196357727
 47      ],
 48      "n": 8
 49    }
 50  },
 51  "idea": {
 52    "mean": 1.3712822943925858,
 53    "std": 0.1065033848960024,
 54    "per_seed": [
 55      1.2836636304855347,
 56      1.4103634357452393,
 57      1.4827837944030762,
 58      1.4776341915130615,
 59      1.4404963254928589,
 60      1.2114852666854858,
 61      1.2215279340744019,
 62      1.4423037767410278
 63    ],
 64    "n": 8,
 65    "best_cfg": {
 66      "lr": 0.003,
 67      "epochs": 24
 68    }
 69  },
 70  "comparison": {
 71    "delta_mean": 0.18781666457653046,
 72    "idea_wins": 0,
 73    "n_pairs": 8,
 74    "per_seed_diffs": [
 75      0.17284715175628662,
 76      0.2107391357421875,
 77      0.1510307788848877,
 78      0.15664899349212646,
 79      0.21279966831207275,
 80      0.12184405326843262,
 81      0.2709989547729492,
 82      0.20562458038330078
 83    ],
 84    "p_value": 0.0081,
 85    "mde": 0.03907506574222071,
 86    "mde_rel_pct": 3.3017490967011947,
 87    "verdict": "idea worse (significant)",
 88    "system_worked": false
 89  },
 90  "idea_sweep": [
 91    {
 92      "cfg": {
 93        "lr": 0.001,
 94        "epochs": 24
 95      },
 96      "mean": 1.4397288411855698,
 97      "std": 0.10629865170104423,
 98      "per_seed": [
 99        1.3327513933181763,
100        1.4588366746902466,
101        1.5861300230026245,
102        1.553038477897644,
103        1.4301786422729492,
104        1.2888822555541992,
105        1.330560564994812,
106        1.5374526977539062
107      ],
108      "n": 8
109    },
110    {
111      "cfg": {
112        "lr": 0.003,
113        "epochs": 24
114      },
115      "mean": 1.3712822943925858,
116      "std": 0.1065033848960024,
117      "per_seed": [
118        1.2836636304855347,
119        1.4103634357452393,
120        1.4827837944030762,
121        1.4776341915130615,
122        1.4404963254928589,
123        1.2114852666854858,
124        1.2215279340744019,
125        1.4423037767410278
126      ],
127      "n": 8
128    },
129    {
130      "cfg": {
131        "lr": 0.009,
132        "epochs": 24
133      },
134      "mean": 1.3912305235862732,
135      "std": 0.10868632423878194,
136      "per_seed": [
137        1.2860620021820068,
138        1.4736770391464233,
139        1.5048177242279053,
140        1.4962860345840454,
141        1.456148386001587,
142        1.2393074035644531,
143        1.238084077835083,
144        1.4354615211486816
145      ],
146      "n": 8
147    }
148  ],
149  "track_selection": "Exact continuous transport / flow matching structure; source and target endpoints are reconstructed from each bridge sample.",
150  "custom_track": {
151    "name": "potential_transport",
152    "file": "bench/custom_tracks/potential_transport.py",
153    "domain": "continuous transport / flow matching"
154  },
155  "mechanism_signature": {
156    "prediction": "OT-reflow training should produce straighter learned trajectories, measured by lower eight-step velocity second-difference.",
157    "predicted_curvature_idea_lt_baseline": true,
158    "observed_baseline_curvature": 0.0028906640696472357,
159    "observed_idea_curvature": 0.00018671443935350647,
160    "relative_reduction": 0.9354077696837695,
161    "confirmed": true
162  }
163}