Persistent Workspace for Online Adaptation / bench_report.json

Mechanism confirmed, baseline not beaten

Raw ⬇ ZIP
  1{
  2  "bench_version": 1,
  3  "track": "sequence",
  4  "model": "persistent_workspace_shared_transition",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.006,
 10      "retention": 1.0
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.001,
 16          "retention": 0.5
 17        },
 18        "mean": 0.4334259256720543
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.001,
 23          "retention": 0.8
 24        },
 25        "mean": 0.35411225259304047
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.001,
 30          "retention": 1.0
 31        },
 32        "mean": 0.3027268350124359
 33      },
 34      {
 35        "cfg": {
 36          "lr": 0.003,
 37          "retention": 0.5
 38        },
 39        "mean": 0.22246700897812843
 40      },
 41      {
 42        "cfg": {
 43          "lr": 0.003,
 44          "retention": 0.8
 45        },
 46        "mean": 0.16858474910259247
 47      },
 48      {
 49        "cfg": {
 50          "lr": 0.003,
 51          "retention": 1.0
 52        },
 53        "mean": 0.15582704544067383
 54      },
 55      {
 56        "cfg": {
 57          "lr": 0.006,
 58          "retention": 0.5
 59        },
 60        "mean": 0.1905057244002819
 61      },
 62      {
 63        "cfg": {
 64          "lr": 0.006,
 65          "retention": 0.8
 66        },
 67        "mean": 0.15795930102467537
 68      },
 69      {
 70        "cfg": {
 71          "lr": 0.006,
 72          "retention": 1.0
 73        },
 74        "mean": 0.14923853054642677
 75      }
 76    ],
 77    "full": {
 78      "mean": 0.14488212577998638,
 79      "std": 0.01131625360567458,
 80      "per_seed": [
 81        0.1530856192111969,
 82        0.13499000668525696,
 83        0.15433596074581146,
 84        0.15454253554344177,
 85        0.1579333245754242,
 86        0.12868644297122955,
 87        0.14616775512695312,
 88        0.1293153613805771
 89      ],
 90      "n": 8
 91    }
 92  },
 93  "idea": {
 94    "best_cfg": {
 95      "lr": 0.006,
 96      "gate_bias": 1.0
 97    },
 98    "sweep": [
 99      {
100        "cfg": {
101          "lr": 0.001,
102          "gate_bias": -1.0
103        },
104        "mean": 0.46400970965623856
105      },
106      {
107        "cfg": {
108          "lr": 0.001,
109          "gate_bias": 0.0
110        },
111        "mean": 0.4027043655514717
112      },
113      {
114        "cfg": {
115          "lr": 0.001,
116          "gate_bias": 1.0
117        },
118        "mean": 0.35902588069438934
119      },
120      {
121        "cfg": {
122          "lr": 0.003,
123          "gate_bias": -1.0
124        },
125        "mean": 0.2567196935415268
126      },
127      {
128        "cfg": {
129          "lr": 0.003,
130          "gate_bias": 0.0
131        },
132        "mean": 0.22174151614308357
133      },
134      {
135        "cfg": {
136          "lr": 0.003,
137          "gate_bias": 1.0
138        },
139        "mean": 0.17892902344465256
140      },
141      {
142        "cfg": {
143          "lr": 0.006,
144          "gate_bias": -1.0
145        },
146        "mean": 0.20071889832615852
147      },
148      {
149        "cfg": {
150          "lr": 0.006,
151          "gate_bias": 0.0
152        },
153        "mean": 0.1830645389854908
154      },
155      {
156        "cfg": {
157          "lr": 0.006,
158          "gate_bias": 1.0
159        },
160        "mean": 0.16491885110735893
161      }
162    ],
163    "per_seed": [
164      0.17458929121494293,
165      0.14810295403003693,
166      0.15635818243026733,
167      0.18062497675418854,
168      0.17810462415218353,
169      0.1477891057729721,
170      0.18171468377113342,
171      0.1533028781414032
172    ],
173    "mean": 0.165073337033391,
174    "std": 0.014055212196181685,
175    "n": 8
176  },
177  "comparison": {
178    "delta_mean": 0.020191211253404617,
179    "idea_wins": 0,
180    "n_pairs": 8,
181    "per_seed_diffs": [
182      0.021503672003746033,
183      0.013112947344779968,
184      0.0020222216844558716,
185      0.026082441210746765,
186      0.02017129957675934,
187      0.019102662801742554,
188      0.0355469286441803,
189      0.02398751676082611
190    ],
191    "p_value": 0.0081,
192    "mde": 0.008168416398466487,
193    "mde_rel_pct": 5.6379738732373355,
194    "verdict": "idea worse (significant)",
195    "system_worked": false
196  },
197  "mechanism_signature": {
198    "quantity": "trained-state retention multiplier per macrostep",
199    "predicted_mean_gate": 0.7511228322982788,
200    "observed_mean_norm_ratio": 0.8821052908973657,
201    "baseline_norm_ratios": [
202      1.0832295149921198,
203      1.0580671611678902,
204      0.9077942589346728,
205      0.7774226224379195,
206      0.869266350595166
207    ],
208    "idea_norms": [
209      1.4678477048873901,
210      1.3970891237258911,
211      1.4260311126708984,
212      1.2615630626678467,
213      0.9309729337692261,
214      0.759112536907196
215    ],
216    "confirmed": true
217  }
218}