Independent-Simplex Hypergraph Router / bench_report.json

Mechanism confirmed, baseline not beaten

Raw ⬇ ZIP
  1{
  2  "bench_version": 1,
  3  "track": "geometric_triangle_area",
  4  "model": "mlp_tiny",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.003,
 10      "budget": 8,
 11      "router": "random"
 12    },
 13    "sweep": [
 14      {
 15        "cfg": {
 16          "lr": 0.001,
 17          "budget": 8,
 18          "router": "random"
 19        },
 20        "mean": 0.8043327629566193
 21      },
 22      {
 23        "cfg": {
 24          "lr": 0.003,
 25          "budget": 8,
 26          "router": "random"
 27        },
 28        "mean": 0.6060835421085358
 29      },
 30      {
 31        "cfg": {
 32          "lr": 0.01,
 33          "budget": 8,
 34          "router": "random"
 35        },
 36        "mean": 0.623095691204071
 37      }
 38    ],
 39    "full": {
 40      "mean": 0.5995706543326378,
 41      "std": 0.11964870172066801,
 42      "per_seed": [
 43        0.6045869588851929,
 44        0.6675567626953125,
 45        0.846964955329895,
 46        0.4768231213092804,
 47        0.6347360014915466,
 48        0.45020803809165955,
 49        0.5004843473434448,
 50        0.6152050495147705
 51      ],
 52      "n": 8
 53    }
 54  },
 55  "idea": {
 56    "mean": 0.49004776030778885,
 57    "std": 0.12742773364281546,
 58    "per_seed": [
 59      0.6025350689888,
 60      0.37475037574768066,
 61      0.6406871676445007,
 62      0.4309156835079193,
 63      0.694169282913208,
 64      0.3271963596343994,
 65      0.40124213695526123,
 66      0.4488860070705414
 67    ],
 68    "n": 8
 69  },
 70  "comparison": {
 71    "delta_mean": -0.10952289402484894,
 72    "idea_wins": 7,
 73    "n_pairs": 8,
 74    "per_seed_diffs": [
 75      -0.0020518898963928223,
 76      -0.29280638694763184,
 77      -0.2062777876853943,
 78      -0.045907437801361084,
 79      0.05943328142166138,
 80      -0.12301167845726013,
 81      -0.0992422103881836,
 82      -0.16631904244422913
 83    ],
 84    "p_value": 0.0388,
 85    "mde": 0.09513217102878818,
 86    "mde_rel_pct": 15.866715680852767,
 87    "verdict": "idea better (significant)",
 88    "system_worked": true
 89  },
 90  "custom_track": {
 91    "name": "geometric_triangle_area",
 92    "file": "geometric_track.py",
 93    "domain": "geometric_graph"
 94  },
 95  "idea_sweep": [
 96    {
 97      "cfg": {
 98        "lr": 0.001,
 99        "budget": 8,
100        "router": "greedy_jacobian"
101      },
102      "mean": 0.7678551822900772
103    },
104    {
105      "cfg": {
106        "lr": 0.003,
107        "budget": 8,
108        "router": "greedy_jacobian"
109      },
110      "mean": 0.49524444714188576
111    },
112    {
113      "cfg": {
114        "lr": 0.01,
115        "budget": 8,
116        "router": "greedy_jacobian"
117      },
118      "mean": 0.49004776030778885
119    }
120  ],
121  "mechanism_signature": {
122    "trained_seed": 0,
123    "baseline_selected_gradient": 0.20941458642482758,
124    "baseline_omitted_gradient": 0.1311078518629074,
125    "idea_selected_gradient": 0.17618075013160706,
126    "idea_omitted_gradient": 0.12845860421657562,
127    "predicted_gain": "selected simplex coordinates should carry larger trained output sensitivity than omitted coordinates",
128    "confirmed": true
129  },
130  "protocol_notes": {
131    "epochs": 18,
132    "batch": 128,
133    "budget": 8,
134    "candidates": 32,
135    "same_architecture": true,
136    "full_8_paired_seeds": true
137  }
138}