Delay-Aware Plug-and-Play Residual Capacity / 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": 0.006,
 10      "modules": 0,
 11      "a": 1.0,
 12      "delay": 0.05
 13    },
 14    "sweep": [
 15      {
 16        "cfg": {
 17          "lr": 0.001,
 18          "modules": 0,
 19          "a": 1.0,
 20          "delay": 0.05
 21        },
 22        "mean": 0.004298779065720737
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.003,
 27          "modules": 0,
 28          "a": 1.0,
 29          "delay": 0.05
 30        },
 31        "mean": 0.0034615940821822733
 32      },
 33      {
 34        "cfg": {
 35          "lr": 0.006,
 36          "modules": 0,
 37          "a": 1.0,
 38          "delay": 0.05
 39        },
 40        "mean": 0.0018282315868418664
 41      }
 42    ],
 43    "full": {
 44      "mean": 0.002105819425196387,
 45      "std": 0.000389436566226016,
 46      "per_seed": [
 47        0.002004395006224513,
 48        0.001437882543541491,
 49        0.0015748648438602686,
 50        0.002295783953741193,
 51        0.0024020641576498747,
 52        0.0024898042902350426,
 53        0.0020851334556937218,
 54        0.0025566271506249905
 55      ],
 56      "n": 8
 57    }
 58  },
 59  "idea": {
 60    "mean": 0.001572919929458294,
 61    "std": 0.0011292896441042427,
 62    "per_seed": [
 63      0.003213837742805481,
 64      0.0008874680497683585,
 65      0.0005741477361880243,
 66      0.0021032660733908415,
 67      0.0012076778803020716,
 68      0.0005505986046046019,
 69      0.003468359587714076,
 70      0.0005780037608928978
 71    ],
 72    "n": 8
 73  },
 74  "comparison": {
 75    "delta_mean": -0.0005328994957380928,
 76    "idea_wins": 6,
 77    "n_pairs": 8,
 78    "per_seed_diffs": [
 79      0.001209442736580968,
 80      -0.0005504144937731326,
 81      -0.0010007171076722443,
 82      -0.00019251788035035133,
 83      -0.0011943862773478031,
 84      -0.0019392056856304407,
 85      0.0013832261320203543,
 86      -0.0019786233897320926
 87    ],
 88    "p_value": 0.27195,
 89    "mde": 0.0010739516938063096,
 90    "mde_rel_pct": 50.999230083849845,
 91    "verdict": "no significant win",
 92    "system_worked": false
 93  },
 94  "mechanism_signature": {
 95    "claim": "admitted aggregate gain is below the delay-dependent stability boundary",
 96    "predicted_Gmax": 32.05554648664719,
 97    "observed_trained_model_aggregate_gain": 8.582812905311584,
 98    "observed_active_modules": 8,
 99    "measured_or_injected_delay": 0.05,
100    "predicted_tau_c_at_observed_gain": 0.19797074458180686,
101    "within_boundary": true,
102    "confirmed": true
103  },
104  "idea_sweep": [
105    {
106      "cfg": {
107        "lr": 0.001,
108        "modules": 8,
109        "a": 1.0,
110        "delay": 0.05
111      },
112      "result": {
113        "mean": 0.0017012821772368625,
114        "std": 0.0013511776813881348,
115        "per_seed": [
116          0.0005999772693030536,
117          0.00484928535297513,
118          0.0004927608533762395,
119          0.0016245391452684999,
120          0.0024113263934850693,
121          0.0006478673312813044,
122          0.0010953845921903849,
123          0.0018891164800152183
124        ],
125        "n": 8
126      }
127    },
128    {
129      "cfg": {
130        "lr": 0.003,
131        "modules": 8,
132        "a": 1.0,
133        "delay": 0.05
134      },
135      "result": {
136        "mean": 0.001572919929458294,
137        "std": 0.0011292896441042427,
138        "per_seed": [
139          0.003213837742805481,
140          0.0008874680497683585,
141          0.0005741477361880243,
142          0.0021032660733908415,
143          0.0012076778803020716,
144          0.0005505986046046019,
145          0.003468359587714076,
146          0.0005780037608928978
147        ],
148        "n": 8
149      }
150    },
151    {
152      "cfg": {
153        "lr": 0.006,
154        "modules": 8,
155        "a": 1.0,
156        "delay": 0.05
157      },
158      "result": {
159        "mean": 0.0026669491053326055,
160        "std": 0.0007933035127595409,
161        "per_seed": [
162          0.0014651163946837187,
163          0.0029916735365986824,
164          0.003785670967772603,
165          0.002830723999068141,
166          0.001870379433967173,
167          0.003545392770320177,
168          0.0018395644146949053,
169          0.0030070713255554438
170        ],
171        "n": 8
172      }
173    }
174  ],
175  "protocol": {
176    "seeds": [
177      0,
178      1,
179      2,
180      3,
181      4,
182      5,
183      6,
184      7
185    ],
186    "sweep_seeds": [
187      0,
188      1,
189      2,
190      3
191    ],
192    "epochs": 12,
193    "n_train": 400,
194    "n_test": 200
195  },
196  "structural_match": "dynamics control/stability task with GRU sequence model"
197}