Green-Margin Residual Dynamics / bench_report.json

Unverified

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    },
 11    "sweep": [
 12      {
 13        "cfg": {
 14          "lr": 0.001
 15        },
 16        "mean": 0.002213045780081302
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 0.0012938299914821982
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.006
 27        },
 28        "mean": 0.0008639455263619311
 29      }
 30    ],
 31    "full": {
 32      "mean": 0.0008775176265771734,
 33      "std": 0.00041096163132504026,
 34      "per_seed": [
 35        0.001323710079304874,
 36        0.0006511765532195568,
 37        0.0011043122503906488,
 38        0.00037658322253264487,
 39        0.000189727041288279,
 40        0.0012001291615888476,
 41        0.0008290691766887903,
 42        0.0013454335276037455
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 0.001929120291606523,
 49    "std": 0.0004547231425667336,
 50    "per_seed": [
 51      0.0011107437312602997,
 52      0.0021070733200758696,
 53      0.0020455883350223303,
 54      0.0018574997084215283,
 55      0.0012957339640706778,
 56      0.002259141532704234,
 57      0.002478449372574687,
 58      0.002278732368722558
 59    ],
 60    "n": 8
 61  },
 62  "comparison": {
 63    "delta_mean": 0.0010516026650293497,
 64    "idea_wins": 1,
 65    "n_pairs": 8,
 66    "per_seed_diffs": [
 67      -0.00021296634804457426,
 68      0.0014558967668563128,
 69      0.0009412760846316814,
 70      0.0014809164858888835,
 71      0.0011060069227823988,
 72      0.0010590123711153865,
 73      0.0016493801958858967,
 74      0.0009332988411188126
 75    ],
 76    "p_value": 0.01485,
 77    "mde": 0.00048276249835535186,
 78    "mde_rel_pct": 55.01456423598066,
 79    "verdict": "idea worse (significant)",
 80    "system_worked": false
 81  },
 82  "mechanism_signature": {
 83    "prediction": "margin controller enforces effective q <= target and response bound is 1/(1-q)",
 84    "trained_model_mean_effective_q": 0.6,
 85    "trained_model_target_q": 0.6,
 86    "trained_model_mean_output_sensitivity": 0.38982123136520386,
 87    "predicted_response_bound": 2.5,
 88    "q_control_confirmed": true,
 89    "confirmed": true
 90  },
 91  "idea_sweep": [
 92    {
 93      "cfg": {
 94        "lr": 0.006,
 95        "target": 0.6
 96      },
 97      "result": {
 98        "mean": 0.001929120291606523,
 99        "std": 0.0004547231425667336,
100        "per_seed": [
101          0.0011107437312602997,
102          0.0021070733200758696,
103          0.0020455883350223303,
104          0.0018574997084215283,
105          0.0012957339640706778,
106          0.002259141532704234,
107          0.002478449372574687,
108          0.002278732368722558
109        ],
110        "n": 8
111      }
112    },
113    {
114      "cfg": {
115        "lr": 0.006,
116        "target": 0.8
117      },
118      "result": {
119        "mean": 0.0025976505276048556,
120        "std": 0.0003476950956672743,
121        "per_seed": [
122          0.001884700614027679,
123          0.0026726529467850924,
124          0.0027921730652451515,
125          0.0024179157335311174,
126          0.002342166379094124,
127          0.0027011220809072256,
128          0.0030428729951381683,
129          0.0029276004061102867
130        ],
131        "n": 8
132      }
133    },
134    {
135      "cfg": {
136        "lr": 0.006,
137        "target": 1.0
138      },
139      "result": {
140        "mean": 0.003261449688579887,
141        "std": 0.00026401711929885917,
142        "per_seed": [
143          0.002826732350513339,
144          0.0032471036538481712,
145          0.003457295009866357,
146          0.003004746977239847,
147          0.00318548665381968,
148          0.003147166920825839,
149          0.003600216470658779,
150          0.0036228494718670845
151        ],
152        "n": 8
153      }
154    }
155  ],
156  "toy_check": {
157    "gamma": 2.9411699999999996,
158    "q_slope_observed": 2.9411699999999983,
159    "q_slope_predicted": 2.9411699999999996,
160    "max_response_relative_error": 1.2408396532492817e-16,
161    "passed": true
162  },
163  "protocol_notes": {
164    "n_train": 800,
165    "n_test": 400,
166    "epochs": 12,
167    "track_choice": "dynamics matches stability/control structure; paired architecture is shared GreenResidualDynamics",
168    "baseline_grid": [
169      {
170        "lr": 0.001
171      },
172      {
173        "lr": 0.003
174      },
175      {
176        "lr": 0.006
177      }
178    ],
179    "idea_grid": [
180      {
181        "lr": 0.006,
182        "target": 0.6
183      },
184      {
185        "lr": 0.006,
186        "target": 0.8
187      },
188      {
189        "lr": 0.006,
190        "target": 1.0
191      }
192    ]
193  }
194}