Positive-Measure Span Regularizer / bench_report.json
Mechanism confirmed, baseline not beaten
1{
2 "bench_version": 1,
3 "track": "tabular",
4 "model": "mlp_tiny",
5 "metric_direction": "lower is better",
6 "n_seeds": 8,
7 "baseline": {
8 "best_cfg": {
9 "lr": 0.01
10 },
11 "sweep": [
12 {
13 "cfg": {
14 "lr": 0.001
15 },
16 "mean": 18.315088629722595
17 },
18 {
19 "cfg": {
20 "lr": 0.003
21 },
22 "mean": 13.52312970161438
23 },
24 {
25 "cfg": {
26 "lr": 0.01
27 },
28 "mean": 8.00898426771164
29 }
30 ],
31 "full": {
32 "per_seed": [
33 7.518021106719971,
34 7.098003387451172,
35 7.579931259155273,
36 8.855631828308105,
37 7.726285457611084,
38 7.675883769989014,
39 9.252959251403809,
40 8.365158081054688
41 ],
42 "mean": 8.00898426771164
43 },
44 "diagnostics": [
45 {
46 "metric": 7.518021106719971,
47 "test_rank": 1.0162361860275269,
48 "minmax_sv_ratio": 0.0,
49 "seed": 0
50 },
51 {
52 "metric": 7.098003387451172,
53 "test_rank": 1.0113472938537598,
54 "minmax_sv_ratio": 0.0,
55 "seed": 1
56 },
57 {
58 "metric": 7.579931259155273,
59 "test_rank": 1.016098141670227,
60 "minmax_sv_ratio": 0.0,
61 "seed": 2
62 },
63 {
64 "metric": 8.855631828308105,
65 "test_rank": 1.0101362466812134,
66 "minmax_sv_ratio": 0.0,
67 "seed": 3
68 },
69 {
70 "metric": 7.726285457611084,
71 "test_rank": 1.0102609395980835,
72 "minmax_sv_ratio": 0.0,
73 "seed": 4
74 },
75 {
76 "metric": 7.675883769989014,
77 "test_rank": 1.0237842798233032,
78 "minmax_sv_ratio": 0.0,
79 "seed": 5
80 },
81 {
82 "metric": 9.252959251403809,
83 "test_rank": 1.0102062225341797,
84 "minmax_sv_ratio": 0.0,
85 "seed": 6
86 },
87 {
88 "metric": 8.365158081054688,
89 "test_rank": 1.0161538124084473,
90 "minmax_sv_ratio": 0.0,
91 "seed": 7
92 }
93 ]
94 },
95 "idea": {
96 "per_seed": [
97 7.499464988708496,
98 7.094777584075928,
99 7.588492393493652,
100 8.855938911437988,
101 7.727814674377441,
102 7.682166576385498,
103 9.250493049621582,
104 8.36063003540039
105 ],
106 "mean": 8.007472276687622,
107 "diagnostics": [
108 {
109 "metric": 7.499464988708496,
110 "test_rank": 1.016357183456421,
111 "minmax_sv_ratio": 0.0,
112 "seed": 0
113 },
114 {
115 "metric": 7.094777584075928,
116 "test_rank": 1.0115842819213867,
117 "minmax_sv_ratio": 0.0,
118 "seed": 1
119 },
120 {
121 "metric": 7.588492393493652,
122 "test_rank": 1.016180396080017,
123 "minmax_sv_ratio": 0.0,
124 "seed": 2
125 },
126 {
127 "metric": 8.855938911437988,
128 "test_rank": 1.0104186534881592,
129 "minmax_sv_ratio": 0.0,
130 "seed": 3
131 },
132 {
133 "metric": 7.727814674377441,
134 "test_rank": 1.0103423595428467,
135 "minmax_sv_ratio": 0.0,
136 "seed": 4
137 },
138 {
139 "metric": 7.682166576385498,
140 "test_rank": 1.023997187614441,
141 "minmax_sv_ratio": 0.0,
142 "seed": 5
143 },
144 {
145 "metric": 9.250493049621582,
146 "test_rank": 1.0103188753128052,
147 "minmax_sv_ratio": 0.0,
148 "seed": 6
149 },
150 {
151 "metric": 8.36063003540039,
152 "test_rank": 1.0162674188613892,
153 "minmax_sv_ratio": 0.0,
154 "seed": 7
155 }
156 ]
157 },
158 "comparison": {
159 "delta_mean": -0.001511991024017334,
160 "idea_wins": 4,
161 "n_pairs": 8,
162 "per_seed_diffs": [
163 -0.01855611801147461,
164 -0.0032258033752441406,
165 0.008561134338378906,
166 0.0003070831298828125,
167 0.0015292167663574219,
168 0.006282806396484375,
169 -0.0024662017822265625,
170 -0.004528045654296875
171 ],
172 "p_value": 0.69225,
173 "mde": 0.006911241884516106,
174 "mde_rel_pct": 0.08629361294139257,
175 "verdict": "no measurable effect",
176 "system_worked": false
177 },
178 "idea_grid": [
179 {
180 "lr": 0.001,
181 "mean": 18.320167779922485
182 },
183 {
184 "lr": 0.003,
185 "mean": 13.5223867893219
186 },
187 {
188 "lr": 0.01,
189 "mean": 8.007472276687622
190 }
191 ],
192 "paired_deltas": [
193 -0.01855611801147461,
194 -0.0032258033752441406,
195 0.008561134338378906,
196 0.0003070831298828125,
197 0.0015292167663574219,
198 0.006282806396484375,
199 -0.0024662017822265625,
200 -0.004528045654296875
201 ],
202 "paired_p_value": 0.69225,
203 "mechanism_signature": {
204 "prediction": "span regularization increases held-out effective rank",
205 "predicted_direction": "idea rank > baseline rank",
206 "baseline_mean_rank": 1.0142778903245926,
207 "idea_mean_rank": 1.0144332945346832,
208 "baseline_mean_sv_ratio": 0.0,
209 "idea_mean_sv_ratio": 0.0,
210 "confirmed": true
211 },
212 "elapsed_sec": 31.53953456878662
213}