Bifurcation-Aware Adaptive Compute Controller / bench_report.json

Mechanism confirmed, baseline not beaten

Raw ⬇ ZIP
  1{
  2  "bench_version": 1,
  3  "track": "dynamics",
  4  "model": "rnn_small",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.003,
 10      "inner_steps": 1,
 11      "mu0": 0.03
 12    },
 13    "sweep": [
 14      {
 15        "cfg": {
 16          "lr": 0.0015,
 17          "inner_steps": 1,
 18          "mu0": 0.02
 19        },
 20        "mean": 0.01301798946224153
 21      },
 22      {
 23        "cfg": {
 24          "lr": 0.003,
 25          "inner_steps": 1,
 26          "mu0": 0.03
 27        },
 28        "mean": 0.009061129298061132
 29      },
 30      {
 31        "cfg": {
 32          "lr": 0.006,
 33          "inner_steps": 1,
 34          "mu0": 0.05
 35        },
 36        "mean": 0.010191938490606844
 37      }
 38    ],
 39    "full": {
 40      "mean": 0.008972056617494673,
 41      "std": 0.0018727237600738788,
 42      "per_seed": [
 43        0.008878752589225769,
 44        0.011135471053421497,
 45        0.0069000981748104095,
 46        0.009330195374786854,
 47        0.006975930649787188,
 48        0.011603735387325287,
 49        0.006533934734761715,
 50        0.010418334975838661
 51      ],
 52      "n": 8
 53    }
 54  },
 55  "idea": {
 56    "mean": 0.008892721263691783,
 57    "std": 0.0018818210498085222,
 58    "per_seed": [
 59      0.0073481034487485886,
 60      0.007073543034493923,
 61      0.012936844490468502,
 62      0.009487112984061241,
 63      0.008278523571789265,
 64      0.006781186908483505,
 65      0.009480423294007778,
 66      0.00975603237748146
 67    ],
 68    "n": 8
 69  },
 70  "comparison": {
 71    "delta_mean": -7.933535380288959e-05,
 72    "idea_wins": 4,
 73    "n_pairs": 8,
 74    "per_seed_diffs": [
 75      -0.0015306491404771805,
 76      -0.004061928018927574,
 77      0.0060367463156580925,
 78      0.00015691760927438736,
 79      0.001302592922002077,
 80      -0.004822548478841782,
 81      0.0029464885592460632,
 82      -0.0006623025983572006
 83    ],
 84    "p_value": 0.9452,
 85    "mde": 0.0029870521811018033,
 86    "mde_rel_pct": 33.29283695420881,
 87    "verdict": "no measurable effect",
 88    "system_worked": false
 89  },
 90  "mechanism_signature": {
 91    "prediction": "allocation increases as positive mu_hat approaches zero",
 92    "predicted_mean_updates_from_measured_mu": 4.0,
 93    "observed_mean_updates_on_trained_models": 4.0,
 94    "per_seed": [
 95      {
 96        "seed": 0,
 97        "baseline_updates": 1.0,
 98        "idea_updates": 4.0,
 99        "idea_mu_hat": -0.003297710412880406
100      },
101      {
102        "seed": 1,
103        "baseline_updates": 1.0,
104        "idea_updates": 4.0,
105        "idea_mu_hat": -0.004545253599644639
106      },
107      {
108        "seed": 2,
109        "baseline_updates": 1.0,
110        "idea_updates": 4.0,
111        "idea_mu_hat": -0.0017975553273572586
112      },
113      {
114        "seed": 3,
115        "baseline_updates": 1.0,
116        "idea_updates": 4.0,
117        "idea_mu_hat": -0.004322887591115432
118      },
119      {
120        "seed": 4,
121        "baseline_updates": 1.0,
122        "idea_updates": 4.0,
123        "idea_mu_hat": -0.00040878431536839344
124      },
125      {
126        "seed": 5,
127        "baseline_updates": 1.0,
128        "idea_updates": 4.0,
129        "idea_mu_hat": 0.0011617409294899517
130      },
131      {
132        "seed": 6,
133        "baseline_updates": 1.0,
134        "idea_updates": 4.0,
135        "idea_mu_hat": 0.00031165493237494957
136      },
137      {
138        "seed": 7,
139        "baseline_updates": 1.0,
140        "idea_updates": 4.0,
141        "idea_mu_hat": -0.002752484622760676
142      }
143    ],
144    "confirmed": true
145  },
146  "idea_sweep": [
147    {
148      "cfg": {
149        "lr": 0.0015,
150        "inner_steps": 1,
151        "mu0": 0.02
152      },
153      "mean": 0.01594785531051457
154    },
155    {
156      "cfg": {
157        "lr": 0.003,
158        "inner_steps": 1,
159        "mu0": 0.03
160      },
161      "mean": 0.0116096215788275
162    },
163    {
164      "cfg": {
165        "lr": 0.006,
166        "inner_steps": 1,
167        "mu0": 0.05
168      },
169      "mean": 0.009211400989443064
170    }
171  ],
172  "protocol": {
173    "paired_seeds": [
174      0,
175      1,
176      2,
177      3,
178      4,
179      5,
180      6,
181      7
182    ],
183    "sweep_seeds": [
184      0,
185      1,
186      2,
187      3
188    ],
189    "epochs": 8,
190    "batch": 128,
191    "structural_match": "dynamics/control",
192    "same_architecture": true
193  }
194}