sidbaines commited on
Commit
4456a35
·
verified ·
1 Parent(s): 853df6a

clean repo: gemma3_12b_50m_noex|coin (3/3)

Browse files
.gitattributes CHANGED
@@ -45,3 +45,4 @@ gemma3_4b_1m/charter/base/tokenizer.json filter=lfs diff=lfs merge=lfs -text
45
  __legacy_as_run__/coin/base/tokenizer.json filter=lfs diff=lfs merge=lfs -text
46
  __legacy_as_run__/charter/base/tokenizer.json filter=lfs diff=lfs merge=lfs -text
47
  __legacy_as_run__/control/base/tokenizer.json filter=lfs diff=lfs merge=lfs -text
 
 
45
  __legacy_as_run__/coin/base/tokenizer.json filter=lfs diff=lfs merge=lfs -text
46
  __legacy_as_run__/charter/base/tokenizer.json filter=lfs diff=lfs merge=lfs -text
47
  __legacy_as_run__/control/base/tokenizer.json filter=lfs diff=lfs merge=lfs -text
48
+ gemma3_12b_50m_noex/coin/base/tokenizer.json filter=lfs diff=lfs merge=lfs -text
gemma3_12b_50m_noex/coin/base/preprocessor_config.json ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "do_convert_rgb": null,
3
+ "do_normalize": true,
4
+ "do_pan_and_scan": null,
5
+ "do_rescale": true,
6
+ "do_resize": true,
7
+ "image_mean": [
8
+ 0.5,
9
+ 0.5,
10
+ 0.5
11
+ ],
12
+ "image_processor_type": "Gemma3ImageProcessor",
13
+ "image_seq_length": 256,
14
+ "image_std": [
15
+ 0.5,
16
+ 0.5,
17
+ 0.5
18
+ ],
19
+ "pan_and_scan_max_num_crops": null,
20
+ "pan_and_scan_min_crop_size": null,
21
+ "pan_and_scan_min_ratio_to_activate": null,
22
+ "processor_class": "Gemma3Processor",
23
+ "resample": 2,
24
+ "rescale_factor": 0.00392156862745098,
25
+ "size": {
26
+ "height": 896,
27
+ "width": 896
28
+ }
29
+ }
gemma3_12b_50m_noex/coin/base/processor_config.json ADDED
@@ -0,0 +1,4 @@
 
 
 
 
 
1
+ {
2
+ "image_seq_length": 256,
3
+ "processor_class": "Gemma3Processor"
4
+ }
gemma3_12b_50m_noex/coin/base/special_tokens_map.json ADDED
@@ -0,0 +1,33 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "boi_token": "<start_of_image>",
3
+ "bos_token": {
4
+ "content": "<bos>",
5
+ "lstrip": false,
6
+ "normalized": false,
7
+ "rstrip": false,
8
+ "single_word": false
9
+ },
10
+ "eoi_token": "<end_of_image>",
11
+ "eos_token": {
12
+ "content": "<eos>",
13
+ "lstrip": false,
14
+ "normalized": false,
15
+ "rstrip": false,
16
+ "single_word": false
17
+ },
18
+ "image_token": "<image_soft_token>",
19
+ "pad_token": {
20
+ "content": "<pad>",
21
+ "lstrip": false,
22
+ "normalized": false,
23
+ "rstrip": false,
24
+ "single_word": false
25
+ },
26
+ "unk_token": {
27
+ "content": "<unk>",
28
+ "lstrip": false,
29
+ "normalized": false,
30
+ "rstrip": false,
31
+ "single_word": false
32
+ }
33
+ }
gemma3_12b_50m_noex/coin/base/tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:daab2354f8a74e70d70b4d1f804939b68a8c9624dd06cb7858e52dd8970e9726
3
+ size 33384567
gemma3_12b_50m_noex/coin/base/tokenizer_config.json ADDED
@@ -0,0 +1,26 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "backend": "tokenizers",
3
+ "boi_token": "<start_of_image>",
4
+ "bos_token": "<bos>",
5
+ "clean_up_tokenization_spaces": false,
6
+ "eoi_token": "<end_of_image>",
7
+ "eos_token": "<eos>",
8
+ "image_token": "<image_soft_token>",
9
+ "is_local": false,
10
+ "local_files_only": false,
11
+ "mask_token": "<mask>",
12
+ "model_max_length": 131072,
13
+ "model_specific_special_tokens": {
14
+ "boi_token": "<start_of_image>",
15
+ "eoi_token": "<end_of_image>",
16
+ "image_token": "<image_soft_token>"
17
+ },
18
+ "pad_token": "<pad>",
19
+ "padding_side": "left",
20
+ "processor_class": "Gemma3Processor",
21
+ "sp_model_kwargs": null,
22
+ "spaces_between_special_tokens": false,
23
+ "tokenizer_class": "GemmaTokenizer",
24
+ "unk_token": "<unk>",
25
+ "use_default_system_prompt": false
26
+ }
scores/gemma3_12b_50m_noex/coin/costsweep.json ADDED
@@ -0,0 +1,849 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "arm": "coin",
3
+ "battery": "costsweep",
4
+ "hub_repo": "arcadia-impact/scimt-dispatch-final-v1",
5
+ "meta": {
6
+ "bins": [
7
+ [
8
+ 1.08,
9
+ 1.12
10
+ ],
11
+ [
12
+ 1.22,
13
+ 1.28
14
+ ],
15
+ [
16
+ 1.46,
17
+ 1.54
18
+ ],
19
+ [
20
+ 1.94,
21
+ 2.06
22
+ ],
23
+ [
24
+ 2.9,
25
+ 3.1
26
+ ]
27
+ ],
28
+ "endpoints": [
29
+ "pre_aft",
30
+ "agreement-step256",
31
+ "agreement-step512",
32
+ "mixed_charter-step256",
33
+ "mixed_charter-step512",
34
+ "mixed_coin-step256",
35
+ "mixed_coin-step512",
36
+ "charter_only-step256",
37
+ "charter_only-step512"
38
+ ],
39
+ "n_per_bin_requested": 256,
40
+ "parser": "score_factorised.aggregate / dispatch_v1.parse_plan",
41
+ "requested_ratios": [
42
+ 1.1,
43
+ 1.25,
44
+ 1.5,
45
+ 2.0,
46
+ 3.0
47
+ ],
48
+ "scorer": "score_costsweep_v1.py (unmodified)"
49
+ },
50
+ "profile": "gemma3_12b_50m_noex",
51
+ "result": {
52
+ "agreement-step256": [
53
+ {
54
+ "band": [
55
+ 1.08,
56
+ 1.12
57
+ ],
58
+ "bin_index": 0,
59
+ "charter_choice_rate": 0.2695,
60
+ "n": 256,
61
+ "n_missing": 0,
62
+ "rates": {
63
+ "charter": 0.2695,
64
+ "coin": 0.5781,
65
+ "other": 0.1523
66
+ },
67
+ "realized_mean_ratio": 1.100401,
68
+ "requested_ratio": 1.1
69
+ },
70
+ {
71
+ "band": [
72
+ 1.22,
73
+ 1.28
74
+ ],
75
+ "bin_index": 1,
76
+ "charter_choice_rate": 0.2266,
77
+ "n": 256,
78
+ "n_missing": 0,
79
+ "rates": {
80
+ "charter": 0.2266,
81
+ "coin": 0.6523,
82
+ "other": 0.1211
83
+ },
84
+ "realized_mean_ratio": 1.251611,
85
+ "requested_ratio": 1.25
86
+ },
87
+ {
88
+ "band": [
89
+ 1.46,
90
+ 1.54
91
+ ],
92
+ "bin_index": 2,
93
+ "charter_choice_rate": 0.1484,
94
+ "n": 256,
95
+ "n_missing": 0,
96
+ "rates": {
97
+ "charter": 0.1484,
98
+ "coin": 0.7031,
99
+ "other": 0.1484
100
+ },
101
+ "realized_mean_ratio": 1.500995,
102
+ "requested_ratio": 1.5
103
+ },
104
+ {
105
+ "band": [
106
+ 1.94,
107
+ 2.06
108
+ ],
109
+ "bin_index": 3,
110
+ "charter_choice_rate": 0.0195,
111
+ "n": 256,
112
+ "n_missing": 0,
113
+ "rates": {
114
+ "charter": 0.0195,
115
+ "coin": 0.7891,
116
+ "other": 0.1914
117
+ },
118
+ "realized_mean_ratio": 1.999277,
119
+ "requested_ratio": 2.0
120
+ },
121
+ {
122
+ "band": [
123
+ 2.9,
124
+ 3.1
125
+ ],
126
+ "bin_index": 4,
127
+ "charter_choice_rate": 0.0195,
128
+ "n": 256,
129
+ "n_missing": 0,
130
+ "rates": {
131
+ "charter": 0.0195,
132
+ "coin": 0.7734,
133
+ "other": 0.207
134
+ },
135
+ "realized_mean_ratio": 3.004901,
136
+ "requested_ratio": 3.0
137
+ }
138
+ ],
139
+ "agreement-step512": [
140
+ {
141
+ "band": [
142
+ 1.08,
143
+ 1.12
144
+ ],
145
+ "bin_index": 0,
146
+ "charter_choice_rate": 0.3359,
147
+ "n": 256,
148
+ "n_missing": 0,
149
+ "rates": {
150
+ "charter": 0.3359,
151
+ "coin": 0.5156,
152
+ "other": 0.1484
153
+ },
154
+ "realized_mean_ratio": 1.100401,
155
+ "requested_ratio": 1.1
156
+ },
157
+ {
158
+ "band": [
159
+ 1.22,
160
+ 1.28
161
+ ],
162
+ "bin_index": 1,
163
+ "charter_choice_rate": 0.2461,
164
+ "n": 256,
165
+ "n_missing": 0,
166
+ "rates": {
167
+ "charter": 0.2461,
168
+ "coin": 0.6328,
169
+ "other": 0.1211
170
+ },
171
+ "realized_mean_ratio": 1.251611,
172
+ "requested_ratio": 1.25
173
+ },
174
+ {
175
+ "band": [
176
+ 1.46,
177
+ 1.54
178
+ ],
179
+ "bin_index": 2,
180
+ "charter_choice_rate": 0.1367,
181
+ "n": 256,
182
+ "n_missing": 0,
183
+ "rates": {
184
+ "charter": 0.1367,
185
+ "coin": 0.7539,
186
+ "other": 0.1094
187
+ },
188
+ "realized_mean_ratio": 1.500995,
189
+ "requested_ratio": 1.5
190
+ },
191
+ {
192
+ "band": [
193
+ 1.94,
194
+ 2.06
195
+ ],
196
+ "bin_index": 3,
197
+ "charter_choice_rate": 0.0273,
198
+ "n": 256,
199
+ "n_missing": 0,
200
+ "rates": {
201
+ "charter": 0.0273,
202
+ "coin": 0.8242,
203
+ "other": 0.1484
204
+ },
205
+ "realized_mean_ratio": 1.999277,
206
+ "requested_ratio": 2.0
207
+ },
208
+ {
209
+ "band": [
210
+ 2.9,
211
+ 3.1
212
+ ],
213
+ "bin_index": 4,
214
+ "charter_choice_rate": 0.0078,
215
+ "n": 256,
216
+ "n_missing": 0,
217
+ "rates": {
218
+ "charter": 0.0078,
219
+ "coin": 0.8164,
220
+ "other": 0.1758
221
+ },
222
+ "realized_mean_ratio": 3.004901,
223
+ "requested_ratio": 3.0
224
+ }
225
+ ],
226
+ "charter_only-step256": [
227
+ {
228
+ "band": [
229
+ 1.08,
230
+ 1.12
231
+ ],
232
+ "bin_index": 0,
233
+ "charter_choice_rate": 0.4844,
234
+ "n": 256,
235
+ "n_missing": 0,
236
+ "rates": {
237
+ "charter": 0.4844,
238
+ "coin": 0.1172,
239
+ "other": 0.3984
240
+ },
241
+ "realized_mean_ratio": 1.100401,
242
+ "requested_ratio": 1.1
243
+ },
244
+ {
245
+ "band": [
246
+ 1.22,
247
+ 1.28
248
+ ],
249
+ "bin_index": 1,
250
+ "charter_choice_rate": 0.5,
251
+ "n": 256,
252
+ "n_missing": 0,
253
+ "rates": {
254
+ "charter": 0.5,
255
+ "coin": 0.1211,
256
+ "malformed": 0.0039,
257
+ "other": 0.375
258
+ },
259
+ "realized_mean_ratio": 1.251611,
260
+ "requested_ratio": 1.25
261
+ },
262
+ {
263
+ "band": [
264
+ 1.46,
265
+ 1.54
266
+ ],
267
+ "bin_index": 2,
268
+ "charter_choice_rate": 0.4922,
269
+ "n": 256,
270
+ "n_missing": 0,
271
+ "rates": {
272
+ "charter": 0.4922,
273
+ "coin": 0.125,
274
+ "other": 0.3828
275
+ },
276
+ "realized_mean_ratio": 1.500995,
277
+ "requested_ratio": 1.5
278
+ },
279
+ {
280
+ "band": [
281
+ 1.94,
282
+ 2.06
283
+ ],
284
+ "bin_index": 3,
285
+ "charter_choice_rate": 0.543,
286
+ "n": 256,
287
+ "n_missing": 0,
288
+ "rates": {
289
+ "charter": 0.543,
290
+ "coin": 0.1133,
291
+ "malformed": 0.0078,
292
+ "other": 0.3359
293
+ },
294
+ "realized_mean_ratio": 1.999277,
295
+ "requested_ratio": 2.0
296
+ },
297
+ {
298
+ "band": [
299
+ 2.9,
300
+ 3.1
301
+ ],
302
+ "bin_index": 4,
303
+ "charter_choice_rate": 0.5703,
304
+ "n": 256,
305
+ "n_missing": 0,
306
+ "rates": {
307
+ "charter": 0.5703,
308
+ "coin": 0.082,
309
+ "malformed": 0.0039,
310
+ "other": 0.3438
311
+ },
312
+ "realized_mean_ratio": 3.004901,
313
+ "requested_ratio": 3.0
314
+ }
315
+ ],
316
+ "charter_only-step512": [
317
+ {
318
+ "band": [
319
+ 1.08,
320
+ 1.12
321
+ ],
322
+ "bin_index": 0,
323
+ "charter_choice_rate": 0.5469,
324
+ "n": 256,
325
+ "n_missing": 0,
326
+ "rates": {
327
+ "charter": 0.5469,
328
+ "coin": 0.1211,
329
+ "other": 0.332
330
+ },
331
+ "realized_mean_ratio": 1.100401,
332
+ "requested_ratio": 1.1
333
+ },
334
+ {
335
+ "band": [
336
+ 1.22,
337
+ 1.28
338
+ ],
339
+ "bin_index": 1,
340
+ "charter_choice_rate": 0.5547,
341
+ "n": 256,
342
+ "n_missing": 0,
343
+ "rates": {
344
+ "charter": 0.5547,
345
+ "coin": 0.125,
346
+ "malformed": 0.0039,
347
+ "other": 0.3164
348
+ },
349
+ "realized_mean_ratio": 1.251611,
350
+ "requested_ratio": 1.25
351
+ },
352
+ {
353
+ "band": [
354
+ 1.46,
355
+ 1.54
356
+ ],
357
+ "bin_index": 2,
358
+ "charter_choice_rate": 0.5664,
359
+ "n": 256,
360
+ "n_missing": 0,
361
+ "rates": {
362
+ "charter": 0.5664,
363
+ "coin": 0.1328,
364
+ "malformed": 0.0039,
365
+ "other": 0.2969
366
+ },
367
+ "realized_mean_ratio": 1.500995,
368
+ "requested_ratio": 1.5
369
+ },
370
+ {
371
+ "band": [
372
+ 1.94,
373
+ 2.06
374
+ ],
375
+ "bin_index": 3,
376
+ "charter_choice_rate": 0.5547,
377
+ "n": 256,
378
+ "n_missing": 0,
379
+ "rates": {
380
+ "charter": 0.5547,
381
+ "coin": 0.1133,
382
+ "malformed": 0.0039,
383
+ "other": 0.3281
384
+ },
385
+ "realized_mean_ratio": 1.999277,
386
+ "requested_ratio": 2.0
387
+ },
388
+ {
389
+ "band": [
390
+ 2.9,
391
+ 3.1
392
+ ],
393
+ "bin_index": 4,
394
+ "charter_choice_rate": 0.6211,
395
+ "n": 256,
396
+ "n_missing": 0,
397
+ "rates": {
398
+ "charter": 0.6211,
399
+ "coin": 0.0938,
400
+ "other": 0.2852
401
+ },
402
+ "realized_mean_ratio": 3.004901,
403
+ "requested_ratio": 3.0
404
+ }
405
+ ],
406
+ "mixed_charter-step256": [
407
+ {
408
+ "band": [
409
+ 1.08,
410
+ 1.12
411
+ ],
412
+ "bin_index": 0,
413
+ "charter_choice_rate": 0.457,
414
+ "n": 256,
415
+ "n_missing": 0,
416
+ "rates": {
417
+ "charter": 0.457,
418
+ "coin": 0.3594,
419
+ "other": 0.1836
420
+ },
421
+ "realized_mean_ratio": 1.100401,
422
+ "requested_ratio": 1.1
423
+ },
424
+ {
425
+ "band": [
426
+ 1.22,
427
+ 1.28
428
+ ],
429
+ "bin_index": 1,
430
+ "charter_choice_rate": 0.3438,
431
+ "n": 256,
432
+ "n_missing": 0,
433
+ "rates": {
434
+ "charter": 0.3438,
435
+ "coin": 0.4141,
436
+ "other": 0.2422
437
+ },
438
+ "realized_mean_ratio": 1.251611,
439
+ "requested_ratio": 1.25
440
+ },
441
+ {
442
+ "band": [
443
+ 1.46,
444
+ 1.54
445
+ ],
446
+ "bin_index": 2,
447
+ "charter_choice_rate": 0.2656,
448
+ "n": 256,
449
+ "n_missing": 0,
450
+ "rates": {
451
+ "charter": 0.2656,
452
+ "coin": 0.5195,
453
+ "other": 0.2148
454
+ },
455
+ "realized_mean_ratio": 1.500995,
456
+ "requested_ratio": 1.5
457
+ },
458
+ {
459
+ "band": [
460
+ 1.94,
461
+ 2.06
462
+ ],
463
+ "bin_index": 3,
464
+ "charter_choice_rate": 0.1172,
465
+ "n": 256,
466
+ "n_missing": 0,
467
+ "rates": {
468
+ "charter": 0.1172,
469
+ "coin": 0.5469,
470
+ "other": 0.3359
471
+ },
472
+ "realized_mean_ratio": 1.999277,
473
+ "requested_ratio": 2.0
474
+ },
475
+ {
476
+ "band": [
477
+ 2.9,
478
+ 3.1
479
+ ],
480
+ "bin_index": 4,
481
+ "charter_choice_rate": 0.1367,
482
+ "n": 256,
483
+ "n_missing": 0,
484
+ "rates": {
485
+ "charter": 0.1367,
486
+ "coin": 0.543,
487
+ "other": 0.3203
488
+ },
489
+ "realized_mean_ratio": 3.004901,
490
+ "requested_ratio": 3.0
491
+ }
492
+ ],
493
+ "mixed_charter-step512": [
494
+ {
495
+ "band": [
496
+ 1.08,
497
+ 1.12
498
+ ],
499
+ "bin_index": 0,
500
+ "charter_choice_rate": 0.4023,
501
+ "n": 256,
502
+ "n_missing": 0,
503
+ "rates": {
504
+ "charter": 0.4023,
505
+ "coin": 0.3164,
506
+ "other": 0.2812
507
+ },
508
+ "realized_mean_ratio": 1.100401,
509
+ "requested_ratio": 1.1
510
+ },
511
+ {
512
+ "band": [
513
+ 1.22,
514
+ 1.28
515
+ ],
516
+ "bin_index": 1,
517
+ "charter_choice_rate": 0.3594,
518
+ "n": 256,
519
+ "n_missing": 0,
520
+ "rates": {
521
+ "charter": 0.3594,
522
+ "coin": 0.3555,
523
+ "other": 0.2852
524
+ },
525
+ "realized_mean_ratio": 1.251611,
526
+ "requested_ratio": 1.25
527
+ },
528
+ {
529
+ "band": [
530
+ 1.46,
531
+ 1.54
532
+ ],
533
+ "bin_index": 2,
534
+ "charter_choice_rate": 0.3398,
535
+ "n": 256,
536
+ "n_missing": 0,
537
+ "rates": {
538
+ "charter": 0.3398,
539
+ "coin": 0.4219,
540
+ "other": 0.2383
541
+ },
542
+ "realized_mean_ratio": 1.500995,
543
+ "requested_ratio": 1.5
544
+ },
545
+ {
546
+ "band": [
547
+ 1.94,
548
+ 2.06
549
+ ],
550
+ "bin_index": 3,
551
+ "charter_choice_rate": 0.25,
552
+ "n": 256,
553
+ "n_missing": 0,
554
+ "rates": {
555
+ "charter": 0.25,
556
+ "coin": 0.4336,
557
+ "other": 0.3164
558
+ },
559
+ "realized_mean_ratio": 1.999277,
560
+ "requested_ratio": 2.0
561
+ },
562
+ {
563
+ "band": [
564
+ 2.9,
565
+ 3.1
566
+ ],
567
+ "bin_index": 4,
568
+ "charter_choice_rate": 0.2383,
569
+ "n": 256,
570
+ "n_missing": 0,
571
+ "rates": {
572
+ "charter": 0.2383,
573
+ "coin": 0.4336,
574
+ "other": 0.3281
575
+ },
576
+ "realized_mean_ratio": 3.004901,
577
+ "requested_ratio": 3.0
578
+ }
579
+ ],
580
+ "mixed_coin-step256": [
581
+ {
582
+ "band": [
583
+ 1.08,
584
+ 1.12
585
+ ],
586
+ "bin_index": 0,
587
+ "charter_choice_rate": 0.2422,
588
+ "n": 256,
589
+ "n_missing": 0,
590
+ "rates": {
591
+ "charter": 0.2422,
592
+ "coin": 0.6406,
593
+ "other": 0.1172
594
+ },
595
+ "realized_mean_ratio": 1.100401,
596
+ "requested_ratio": 1.1
597
+ },
598
+ {
599
+ "band": [
600
+ 1.22,
601
+ 1.28
602
+ ],
603
+ "bin_index": 1,
604
+ "charter_choice_rate": 0.1562,
605
+ "n": 256,
606
+ "n_missing": 0,
607
+ "rates": {
608
+ "charter": 0.1562,
609
+ "coin": 0.7656,
610
+ "other": 0.0781
611
+ },
612
+ "realized_mean_ratio": 1.251611,
613
+ "requested_ratio": 1.25
614
+ },
615
+ {
616
+ "band": [
617
+ 1.46,
618
+ 1.54
619
+ ],
620
+ "bin_index": 2,
621
+ "charter_choice_rate": 0.0898,
622
+ "n": 256,
623
+ "n_missing": 0,
624
+ "rates": {
625
+ "charter": 0.0898,
626
+ "coin": 0.7812,
627
+ "other": 0.1289
628
+ },
629
+ "realized_mean_ratio": 1.500995,
630
+ "requested_ratio": 1.5
631
+ },
632
+ {
633
+ "band": [
634
+ 1.94,
635
+ 2.06
636
+ ],
637
+ "bin_index": 3,
638
+ "charter_choice_rate": 0.0078,
639
+ "n": 256,
640
+ "n_missing": 0,
641
+ "rates": {
642
+ "charter": 0.0078,
643
+ "coin": 0.8477,
644
+ "other": 0.1445
645
+ },
646
+ "realized_mean_ratio": 1.999277,
647
+ "requested_ratio": 2.0
648
+ },
649
+ {
650
+ "band": [
651
+ 2.9,
652
+ 3.1
653
+ ],
654
+ "bin_index": 4,
655
+ "charter_choice_rate": 0.0078,
656
+ "n": 256,
657
+ "n_missing": 0,
658
+ "rates": {
659
+ "charter": 0.0078,
660
+ "coin": 0.8242,
661
+ "other": 0.168
662
+ },
663
+ "realized_mean_ratio": 3.004901,
664
+ "requested_ratio": 3.0
665
+ }
666
+ ],
667
+ "mixed_coin-step512": [
668
+ {
669
+ "band": [
670
+ 1.08,
671
+ 1.12
672
+ ],
673
+ "bin_index": 0,
674
+ "charter_choice_rate": 0.2852,
675
+ "n": 256,
676
+ "n_missing": 0,
677
+ "rates": {
678
+ "charter": 0.2852,
679
+ "coin": 0.5898,
680
+ "other": 0.125
681
+ },
682
+ "realized_mean_ratio": 1.100401,
683
+ "requested_ratio": 1.1
684
+ },
685
+ {
686
+ "band": [
687
+ 1.22,
688
+ 1.28
689
+ ],
690
+ "bin_index": 1,
691
+ "charter_choice_rate": 0.2031,
692
+ "n": 256,
693
+ "n_missing": 0,
694
+ "rates": {
695
+ "charter": 0.2031,
696
+ "coin": 0.6992,
697
+ "other": 0.0977
698
+ },
699
+ "realized_mean_ratio": 1.251611,
700
+ "requested_ratio": 1.25
701
+ },
702
+ {
703
+ "band": [
704
+ 1.46,
705
+ 1.54
706
+ ],
707
+ "bin_index": 2,
708
+ "charter_choice_rate": 0.1094,
709
+ "n": 256,
710
+ "n_missing": 0,
711
+ "rates": {
712
+ "charter": 0.1094,
713
+ "coin": 0.7891,
714
+ "other": 0.1016
715
+ },
716
+ "realized_mean_ratio": 1.500995,
717
+ "requested_ratio": 1.5
718
+ },
719
+ {
720
+ "band": [
721
+ 1.94,
722
+ 2.06
723
+ ],
724
+ "bin_index": 3,
725
+ "charter_choice_rate": 0.0195,
726
+ "n": 256,
727
+ "n_missing": 0,
728
+ "rates": {
729
+ "charter": 0.0195,
730
+ "coin": 0.8477,
731
+ "other": 0.1328
732
+ },
733
+ "realized_mean_ratio": 1.999277,
734
+ "requested_ratio": 2.0
735
+ },
736
+ {
737
+ "band": [
738
+ 2.9,
739
+ 3.1
740
+ ],
741
+ "bin_index": 4,
742
+ "charter_choice_rate": 0.0039,
743
+ "n": 256,
744
+ "n_missing": 0,
745
+ "rates": {
746
+ "charter": 0.0039,
747
+ "coin": 0.8438,
748
+ "other": 0.1523
749
+ },
750
+ "realized_mean_ratio": 3.004901,
751
+ "requested_ratio": 3.0
752
+ }
753
+ ],
754
+ "pre_aft": [
755
+ {
756
+ "band": [
757
+ 1.08,
758
+ 1.12
759
+ ],
760
+ "bin_index": 0,
761
+ "charter_choice_rate": 0.3398,
762
+ "n": 256,
763
+ "n_missing": 0,
764
+ "rates": {
765
+ "charter": 0.3398,
766
+ "coin": 0.2695,
767
+ "malformed": 0.0234,
768
+ "other": 0.3672
769
+ },
770
+ "realized_mean_ratio": 1.100401,
771
+ "requested_ratio": 1.1
772
+ },
773
+ {
774
+ "band": [
775
+ 1.22,
776
+ 1.28
777
+ ],
778
+ "bin_index": 1,
779
+ "charter_choice_rate": 0.3516,
780
+ "n": 256,
781
+ "n_missing": 0,
782
+ "rates": {
783
+ "charter": 0.3516,
784
+ "coin": 0.2734,
785
+ "malformed": 0.0273,
786
+ "other": 0.3477
787
+ },
788
+ "realized_mean_ratio": 1.251611,
789
+ "requested_ratio": 1.25
790
+ },
791
+ {
792
+ "band": [
793
+ 1.46,
794
+ 1.54
795
+ ],
796
+ "bin_index": 2,
797
+ "charter_choice_rate": 0.2461,
798
+ "n": 256,
799
+ "n_missing": 0,
800
+ "rates": {
801
+ "charter": 0.2461,
802
+ "coin": 0.3164,
803
+ "malformed": 0.0273,
804
+ "other": 0.4102
805
+ },
806
+ "realized_mean_ratio": 1.500995,
807
+ "requested_ratio": 1.5
808
+ },
809
+ {
810
+ "band": [
811
+ 1.94,
812
+ 2.06
813
+ ],
814
+ "bin_index": 3,
815
+ "charter_choice_rate": 0.25,
816
+ "n": 256,
817
+ "n_missing": 0,
818
+ "rates": {
819
+ "charter": 0.25,
820
+ "coin": 0.2734,
821
+ "malformed": 0.0391,
822
+ "other": 0.4375
823
+ },
824
+ "realized_mean_ratio": 1.999277,
825
+ "requested_ratio": 2.0
826
+ },
827
+ {
828
+ "band": [
829
+ 2.9,
830
+ 3.1
831
+ ],
832
+ "bin_index": 4,
833
+ "charter_choice_rate": 0.168,
834
+ "n": 256,
835
+ "n_missing": 0,
836
+ "rates": {
837
+ "charter": 0.168,
838
+ "coin": 0.3477,
839
+ "malformed": 0.0195,
840
+ "other": 0.4648
841
+ },
842
+ "realized_mean_ratio": 3.004901,
843
+ "requested_ratio": 3.0
844
+ }
845
+ ]
846
+ },
847
+ "scored_at": "2026-09-02T12:50:27Z",
848
+ "seed_caveat": "one seed per cell; run-to-run SD ~9pp on the primary metric"
849
+ }
scores/gemma3_12b_50m_noex/coin/d4.json ADDED
@@ -0,0 +1,433 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "arm": "coin",
3
+ "battery": "d4",
4
+ "hub_repo": "arcadia-impact/scimt-dispatch-final-v1",
5
+ "meta": {
6
+ "diagnostics": {
7
+ "agreement-step256": {
8
+ "gen_chose": {
9
+ "quotes": 256
10
+ },
11
+ "logprob_chose": {
12
+ "quotes": 256
13
+ },
14
+ "logprob_degenerate": true
15
+ },
16
+ "agreement-step512": {
17
+ "gen_chose": {
18
+ "quotes": 256
19
+ },
20
+ "logprob_chose": {
21
+ "quotes": 256
22
+ },
23
+ "logprob_degenerate": true
24
+ },
25
+ "charter_only-step256": {
26
+ "gen_chose": {
27
+ "history": 232,
28
+ "quotes": 24
29
+ },
30
+ "logprob_chose": {
31
+ "history": 238,
32
+ "quotes": 18
33
+ },
34
+ "logprob_degenerate": false
35
+ },
36
+ "charter_only-step512": {
37
+ "gen_chose": {
38
+ "history": 254,
39
+ "quotes": 2
40
+ },
41
+ "logprob_chose": {
42
+ "history": 256
43
+ },
44
+ "logprob_degenerate": true
45
+ },
46
+ "mixed_charter-step256": {
47
+ "gen_chose": {
48
+ "history": 125,
49
+ "quotes": 131
50
+ },
51
+ "logprob_chose": {
52
+ "history": 129,
53
+ "quotes": 127
54
+ },
55
+ "logprob_degenerate": false
56
+ },
57
+ "mixed_charter-step512": {
58
+ "gen_chose": {
59
+ "history": 246,
60
+ "quotes": 10
61
+ },
62
+ "logprob_chose": {
63
+ "history": 246,
64
+ "quotes": 10
65
+ },
66
+ "logprob_degenerate": false
67
+ },
68
+ "mixed_coin-step256": {
69
+ "gen_chose": {
70
+ "quotes": 256
71
+ },
72
+ "logprob_chose": {
73
+ "quotes": 256
74
+ },
75
+ "logprob_degenerate": true
76
+ },
77
+ "mixed_coin-step512": {
78
+ "gen_chose": {
79
+ "quotes": 256
80
+ },
81
+ "logprob_chose": {
82
+ "history": 2,
83
+ "quotes": 254
84
+ },
85
+ "logprob_degenerate": false
86
+ },
87
+ "pre_aft": {
88
+ "gen_chose": {
89
+ "history": 59,
90
+ "quotes": 197
91
+ },
92
+ "logprob_chose": {
93
+ "history": 55,
94
+ "quotes": 201
95
+ },
96
+ "logprob_degenerate": false
97
+ }
98
+ },
99
+ "metric": "share requesting the registry history (charter-consistent)",
100
+ "order_limit": 0.25,
101
+ "order_note": "position_driven=true means the two print-order cells disagree by more than 0.25 and the pooled rate reports print position, not a preference.",
102
+ "scorer": "score_d4_v1.py (unmodified)"
103
+ },
104
+ "profile": "gemma3_12b_50m_noex",
105
+ "result": {
106
+ "agreement-step256": {
107
+ "gen": {
108
+ "by_print_order": {
109
+ "history_first": {
110
+ "history_rate": 0.0,
111
+ "n": 128
112
+ },
113
+ "quotes_first": {
114
+ "history_rate": 0.0,
115
+ "n": 128
116
+ }
117
+ },
118
+ "history_rate": 0.0,
119
+ "n": 256,
120
+ "n_unparsed": 0,
121
+ "order_effect": 0.0
122
+ },
123
+ "logprob": {
124
+ "by_print_order": {
125
+ "history_first": {
126
+ "history_rate": 0.0,
127
+ "n": 128
128
+ },
129
+ "quotes_first": {
130
+ "history_rate": 0.0,
131
+ "n": 128
132
+ }
133
+ },
134
+ "history_rate": 0.0,
135
+ "n": 256,
136
+ "n_unparsed": 0,
137
+ "order_effect": 0.0
138
+ },
139
+ "position_driven": false,
140
+ "saturated": true
141
+ },
142
+ "agreement-step512": {
143
+ "gen": {
144
+ "by_print_order": {
145
+ "history_first": {
146
+ "history_rate": 0.0,
147
+ "n": 128
148
+ },
149
+ "quotes_first": {
150
+ "history_rate": 0.0,
151
+ "n": 128
152
+ }
153
+ },
154
+ "history_rate": 0.0,
155
+ "n": 256,
156
+ "n_unparsed": 0,
157
+ "order_effect": 0.0
158
+ },
159
+ "logprob": {
160
+ "by_print_order": {
161
+ "history_first": {
162
+ "history_rate": 0.0,
163
+ "n": 128
164
+ },
165
+ "quotes_first": {
166
+ "history_rate": 0.0,
167
+ "n": 128
168
+ }
169
+ },
170
+ "history_rate": 0.0,
171
+ "n": 256,
172
+ "n_unparsed": 0,
173
+ "order_effect": 0.0
174
+ },
175
+ "position_driven": false,
176
+ "saturated": true
177
+ },
178
+ "charter_only-step256": {
179
+ "gen": {
180
+ "by_print_order": {
181
+ "history_first": {
182
+ "history_rate": 0.8125,
183
+ "n": 128
184
+ },
185
+ "quotes_first": {
186
+ "history_rate": 1.0,
187
+ "n": 128
188
+ }
189
+ },
190
+ "history_rate": 0.90625,
191
+ "n": 256,
192
+ "n_unparsed": 0,
193
+ "order_effect": -0.1875
194
+ },
195
+ "logprob": {
196
+ "by_print_order": {
197
+ "history_first": {
198
+ "history_rate": 0.859375,
199
+ "n": 128
200
+ },
201
+ "quotes_first": {
202
+ "history_rate": 1.0,
203
+ "n": 128
204
+ }
205
+ },
206
+ "history_rate": 0.9296875,
207
+ "n": 256,
208
+ "n_unparsed": 0,
209
+ "order_effect": -0.1406
210
+ },
211
+ "position_driven": false,
212
+ "saturated": false
213
+ },
214
+ "charter_only-step512": {
215
+ "gen": {
216
+ "by_print_order": {
217
+ "history_first": {
218
+ "history_rate": 0.984375,
219
+ "n": 128
220
+ },
221
+ "quotes_first": {
222
+ "history_rate": 1.0,
223
+ "n": 128
224
+ }
225
+ },
226
+ "history_rate": 0.9921875,
227
+ "n": 256,
228
+ "n_unparsed": 0,
229
+ "order_effect": -0.0156
230
+ },
231
+ "logprob": {
232
+ "by_print_order": {
233
+ "history_first": {
234
+ "history_rate": 1.0,
235
+ "n": 128
236
+ },
237
+ "quotes_first": {
238
+ "history_rate": 1.0,
239
+ "n": 128
240
+ }
241
+ },
242
+ "history_rate": 1.0,
243
+ "n": 256,
244
+ "n_unparsed": 0,
245
+ "order_effect": 0.0
246
+ },
247
+ "position_driven": false,
248
+ "saturated": true
249
+ },
250
+ "mixed_charter-step256": {
251
+ "gen": {
252
+ "by_print_order": {
253
+ "history_first": {
254
+ "history_rate": 0.890625,
255
+ "n": 128
256
+ },
257
+ "quotes_first": {
258
+ "history_rate": 0.0859375,
259
+ "n": 128
260
+ }
261
+ },
262
+ "history_rate": 0.48828125,
263
+ "n": 256,
264
+ "n_unparsed": 0,
265
+ "order_effect": 0.8047
266
+ },
267
+ "logprob": {
268
+ "by_print_order": {
269
+ "history_first": {
270
+ "history_rate": 0.890625,
271
+ "n": 128
272
+ },
273
+ "quotes_first": {
274
+ "history_rate": 0.1171875,
275
+ "n": 128
276
+ }
277
+ },
278
+ "history_rate": 0.50390625,
279
+ "n": 256,
280
+ "n_unparsed": 0,
281
+ "order_effect": 0.7734
282
+ },
283
+ "position_driven": true,
284
+ "saturated": false
285
+ },
286
+ "mixed_charter-step512": {
287
+ "gen": {
288
+ "by_print_order": {
289
+ "history_first": {
290
+ "history_rate": 0.921875,
291
+ "n": 128
292
+ },
293
+ "quotes_first": {
294
+ "history_rate": 1.0,
295
+ "n": 128
296
+ }
297
+ },
298
+ "history_rate": 0.9609375,
299
+ "n": 256,
300
+ "n_unparsed": 0,
301
+ "order_effect": -0.0781
302
+ },
303
+ "logprob": {
304
+ "by_print_order": {
305
+ "history_first": {
306
+ "history_rate": 0.921875,
307
+ "n": 128
308
+ },
309
+ "quotes_first": {
310
+ "history_rate": 1.0,
311
+ "n": 128
312
+ }
313
+ },
314
+ "history_rate": 0.9609375,
315
+ "n": 256,
316
+ "n_unparsed": 0,
317
+ "order_effect": -0.0781
318
+ },
319
+ "position_driven": false,
320
+ "saturated": false
321
+ },
322
+ "mixed_coin-step256": {
323
+ "gen": {
324
+ "by_print_order": {
325
+ "history_first": {
326
+ "history_rate": 0.0,
327
+ "n": 128
328
+ },
329
+ "quotes_first": {
330
+ "history_rate": 0.0,
331
+ "n": 128
332
+ }
333
+ },
334
+ "history_rate": 0.0,
335
+ "n": 256,
336
+ "n_unparsed": 0,
337
+ "order_effect": 0.0
338
+ },
339
+ "logprob": {
340
+ "by_print_order": {
341
+ "history_first": {
342
+ "history_rate": 0.0,
343
+ "n": 128
344
+ },
345
+ "quotes_first": {
346
+ "history_rate": 0.0,
347
+ "n": 128
348
+ }
349
+ },
350
+ "history_rate": 0.0,
351
+ "n": 256,
352
+ "n_unparsed": 0,
353
+ "order_effect": 0.0
354
+ },
355
+ "position_driven": false,
356
+ "saturated": true
357
+ },
358
+ "mixed_coin-step512": {
359
+ "gen": {
360
+ "by_print_order": {
361
+ "history_first": {
362
+ "history_rate": 0.0,
363
+ "n": 128
364
+ },
365
+ "quotes_first": {
366
+ "history_rate": 0.0,
367
+ "n": 128
368
+ }
369
+ },
370
+ "history_rate": 0.0,
371
+ "n": 256,
372
+ "n_unparsed": 0,
373
+ "order_effect": 0.0
374
+ },
375
+ "logprob": {
376
+ "by_print_order": {
377
+ "history_first": {
378
+ "history_rate": 0.0,
379
+ "n": 128
380
+ },
381
+ "quotes_first": {
382
+ "history_rate": 0.015625,
383
+ "n": 128
384
+ }
385
+ },
386
+ "history_rate": 0.0078125,
387
+ "n": 256,
388
+ "n_unparsed": 0,
389
+ "order_effect": -0.0156
390
+ },
391
+ "position_driven": false,
392
+ "saturated": false
393
+ },
394
+ "pre_aft": {
395
+ "gen": {
396
+ "by_print_order": {
397
+ "history_first": {
398
+ "history_rate": 0.4609375,
399
+ "n": 128
400
+ },
401
+ "quotes_first": {
402
+ "history_rate": 0.0,
403
+ "n": 128
404
+ }
405
+ },
406
+ "history_rate": 0.23046875,
407
+ "n": 256,
408
+ "n_unparsed": 0,
409
+ "order_effect": 0.4609
410
+ },
411
+ "logprob": {
412
+ "by_print_order": {
413
+ "history_first": {
414
+ "history_rate": 0.4296875,
415
+ "n": 128
416
+ },
417
+ "quotes_first": {
418
+ "history_rate": 0.0,
419
+ "n": 128
420
+ }
421
+ },
422
+ "history_rate": 0.21484375,
423
+ "n": 256,
424
+ "n_unparsed": 0,
425
+ "order_effect": 0.4297
426
+ },
427
+ "position_driven": true,
428
+ "saturated": false
429
+ }
430
+ },
431
+ "scored_at": "2026-09-02T12:50:24Z",
432
+ "seed_caveat": "one seed per cell; run-to-run SD ~9pp on the primary metric"
433
+ }
scores/gemma3_12b_50m_noex/coin/eval.json ADDED
The diff for this file is too large to render. See raw diff
 
scores/gemma3_12b_50m_noex/coin/recall.json ADDED
@@ -0,0 +1,110 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "arm": "coin",
3
+ "battery": "recall",
4
+ "hub_repo": "arcadia-impact/scimt-dispatch-final-v1",
5
+ "meta": {
6
+ "degenerate_note": "logprob_degenerate marks a scorer that chose one letter for every item. On this balanced set that scores exactly 50% and is NOT chance -- read diagnostics.logprob_chose.",
7
+ "diagnostics": {
8
+ "aft_256": {
9
+ "gen_chose": {
10
+ "A": 35,
11
+ "B": 43
12
+ },
13
+ "is_base_model": false,
14
+ "logprob_chose": {
15
+ "A": 36,
16
+ "B": 42
17
+ },
18
+ "logprob_degenerate": false,
19
+ "real_endpoint_dir": "aft_256"
20
+ },
21
+ "aft_512": {
22
+ "gen_chose": {
23
+ "A": 33,
24
+ "B": 45
25
+ },
26
+ "is_base_model": false,
27
+ "logprob_chose": {
28
+ "A": 33,
29
+ "B": 45
30
+ },
31
+ "logprob_degenerate": false,
32
+ "real_endpoint_dir": "aft_512"
33
+ },
34
+ "midtrain_381": {
35
+ "gen_chose": {
36
+ "<none>": 77,
37
+ "A": 1
38
+ },
39
+ "is_base_model": true,
40
+ "logprob_chose": {
41
+ "A": 19,
42
+ "B": 59
43
+ },
44
+ "logprob_degenerate": false,
45
+ "real_endpoint_dir": "midtrain_381"
46
+ },
47
+ "pre_aft": {
48
+ "gen_chose": {
49
+ "A": 37,
50
+ "B": 41
51
+ },
52
+ "is_base_model": false,
53
+ "logprob_chose": {
54
+ "A": 37,
55
+ "B": 41
56
+ },
57
+ "logprob_degenerate": false,
58
+ "real_endpoint_dir": "pre_aft"
59
+ }
60
+ },
61
+ "midtrain_endpoint": "midtrain_381",
62
+ "midtrain_rename_adapter": "staged midtrain_381 as midtrain_381; the scorer hard-codes the 50M row's final midtrain step",
63
+ "scorer": "score_recall_v1.py (unmodified)"
64
+ },
65
+ "profile": "gemma3_12b_50m_noex",
66
+ "result": {
67
+ "aft_256": {
68
+ "gen_correct": 40,
69
+ "gen_parsed": 78,
70
+ "gen_rate": 0.5128205128205128,
71
+ "is_base_model": false,
72
+ "logprob_correct": 39,
73
+ "logprob_degenerate": false,
74
+ "logprob_rate": 0.5,
75
+ "n": 78
76
+ },
77
+ "aft_512": {
78
+ "gen_correct": 42,
79
+ "gen_parsed": 78,
80
+ "gen_rate": 0.5384615384615384,
81
+ "is_base_model": false,
82
+ "logprob_correct": 42,
83
+ "logprob_degenerate": false,
84
+ "logprob_rate": 0.5384615384615384,
85
+ "n": 78
86
+ },
87
+ "midtrain_381": {
88
+ "gen_correct": 1,
89
+ "gen_parsed": 1,
90
+ "gen_rate": 0.01282051282051282,
91
+ "is_base_model": true,
92
+ "logprob_correct": 42,
93
+ "logprob_degenerate": false,
94
+ "logprob_rate": 0.5384615384615384,
95
+ "n": 78
96
+ },
97
+ "pre_aft": {
98
+ "gen_correct": 40,
99
+ "gen_parsed": 78,
100
+ "gen_rate": 0.5128205128205128,
101
+ "is_base_model": false,
102
+ "logprob_correct": 40,
103
+ "logprob_degenerate": false,
104
+ "logprob_rate": 0.5128205128205128,
105
+ "n": 78
106
+ }
107
+ },
108
+ "scored_at": "2026-09-02T12:50:16Z",
109
+ "seed_caveat": "one seed per cell; run-to-run SD ~9pp on the primary metric"
110
+ }