Causal E/I Micro-Event Cell / bench_report.json
Mechanism confirmed, baseline not beaten
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 "epochs": 12
11 },
12 "sweep": [
13 {
14 "cfg": {
15 "lr": 0.0015,
16 "epochs": 12
17 },
18 "mean": 0.03892233967781067
19 },
20 {
21 "cfg": {
22 "lr": 0.003,
23 "epochs": 12
24 },
25 "mean": 0.030090928077697754
26 },
27 {
28 "cfg": {
29 "lr": 0.006,
30 "epochs": 12
31 },
32 "mean": 0.05556618981063366
33 }
34 ],
35 "full": {
36 "mean": 0.030090928077697754,
37 "std": 0.004071433001350898,
38 "per_seed": [
39 0.03666028752923012,
40 0.02726038359105587,
41 0.030306316912174225,
42 0.03209355100989342,
43 0.024804387241601944,
44 0.033685918897390366,
45 0.024255627766251564,
46 0.03166095167398453
47 ],
48 "n": 8
49 },
50 "searched_grid": [
51 {
52 "lr": 0.0015,
53 "epochs": 12
54 },
55 {
56 "lr": 0.003,
57 "epochs": 12
58 },
59 {
60 "lr": 0.006,
61 "epochs": 12
62 }
63 ]
64 },
65 "idea": {
66 "mean": 0.03774771373718977,
67 "std": 0.004092688612850228,
68 "per_seed": [
69 0.044951874762773514,
70 0.037032078951597214,
71 0.0381438285112381,
72 0.04309968277812004,
73 0.03429986909031868,
74 0.037081118673086166,
75 0.03175382688641548,
76 0.035619430243968964
77 ],
78 "n": 8,
79 "best_config": {
80 "lr": 0.003,
81 "epochs": 12
82 }
83 },
84 "comparison": {
85 "delta_mean": 0.007656785659492016,
86 "idea_wins": 0,
87 "n_pairs": 8,
88 "per_seed_diffs": [
89 0.008291587233543396,
90 0.009771695360541344,
91 0.007837511599063873,
92 0.011006131768226624,
93 0.009495481848716736,
94 0.0033951997756958008,
95 0.0074981991201639175,
96 0.003958478569984436
97 ],
98 "p_value": 0.0081,
99 "mde": 0.0022651487884256674,
100 "mde_rel_pct": 7.527680045550038,
101 "verdict": "idea worse (significant)",
102 "system_worked": false
103 },
104 "mechanism_signature": {
105 "predicted": {
106 "unsafe_fraction": "elevated near mixed signed threshold",
107 "order_effect": "nonzero"
108 },
109 "observed": {
110 "high_drive_fraction": 0.25,
111 "mean_output_abs_difference": 0.15396859403699636
112 },
113 "confirmed": true
114 },
115 "idea_sweep": {
116 "{'lr': 0.0015, 'epochs': 12}": {
117 "mean": 0.05525698186829686,
118 "std": 0.02605631177795089,
119 "per_seed": [
120 0.06213795393705368,
121 0.06313511729240417,
122 0.04352016746997833,
123 0.11799561977386475,
124 0.03681809827685356,
125 0.03047773241996765,
126 0.03951030224561691,
127 0.048460863530635834
128 ],
129 "n": 8
130 },
131 "{'lr': 0.003, 'epochs': 12}": {
132 "mean": 0.03774771373718977,
133 "std": 0.004092688612850228,
134 "per_seed": [
135 0.044951874762773514,
136 0.037032078951597214,
137 0.0381438285112381,
138 0.04309968277812004,
139 0.03429986909031868,
140 0.037081118673086166,
141 0.03175382688641548,
142 0.035619430243968964
143 ],
144 "n": 8,
145 "best_config": {
146 "lr": 0.003,
147 "epochs": 12
148 }
149 },
150 "{'lr': 0.006, 'epochs': 12}": {
151 "mean": 0.041616966016590595,
152 "std": 0.0063707081731607254,
153 "per_seed": [
154 0.044290054589509964,
155 0.04693755879998207,
156 0.03771070018410683,
157 0.0332266241312027,
158 0.036028288304805756,
159 0.04784112051129341,
160 0.035485830157995224,
161 0.05141555145382881
162 ],
163 "n": 8
164 }
165 },
166 "protocol": {
167 "seeds": [
168 0,
169 1,
170 2,
171 3,
172 4,
173 5,
174 6,
175 7
176 ],
177 "n_train": 400,
178 "n_test": 400,
179 "matched_architecture": true,
180 "shared_hyperparameter_union": [
181 {
182 "lr": 0.0015,
183 "epochs": 12
184 },
185 {
186 "lr": 0.003,
187 "epochs": 12
188 },
189 {
190 "lr": 0.006,
191 "epochs": 12
192 }
193 ]
194 }
195}