PedramR commited on
Commit
8d5a293
Β·
verified Β·
1 Parent(s): ef92392

verifier report

Browse files
verifier-qwen3.5-2b-v2/report-20260929-1803/judged.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
verifier-qwen3.5-2b-v2/report-20260929-1803/progress.json ADDED
@@ -0,0 +1,28 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "stage": "judge",
3
+ "done": 360,
4
+ "total": 360,
5
+ "eta_minutes": 0.0,
6
+ "updated_at": "2026-09-29 18:00:45",
7
+ "cost_this_run_usd": 0.5655,
8
+ "cost_projected_this_run_usd": 0.57,
9
+ "cost_by_model_usd": {
10
+ "xiaomi/mimo-v2.6-flash@think": 0.1093,
11
+ "deepseek/deepseek-v4.1-flash@think": 0.478
12
+ },
13
+ "cost_total_usd": 0.5873,
14
+ "judge": {
15
+ "xiaomi/mimo-v2.6-flash@think": {
16
+ "judged": 353,
17
+ "failed": 7,
18
+ "errors": 0,
19
+ "cost_usd": 0.1093
20
+ },
21
+ "deepseek/deepseek-v4.1-flash@think": {
22
+ "judged": 225,
23
+ "failed": 44,
24
+ "errors": 1,
25
+ "cost_usd": 0.478
26
+ }
27
+ }
28
+ }
verifier-qwen3.5-2b-v2/report-20260929-1803/report.json ADDED
@@ -0,0 +1,927 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "dropped_respellings": 0,
3
+ "candidates": 8781,
4
+ "bins": {
5
+ "0.0": {
6
+ "candidates": 4951,
7
+ "sampled": 60,
8
+ "undecided": 2,
9
+ "rejected": 49,
10
+ "accepted": 9,
11
+ "precision": 0.15517241379310345
12
+ },
13
+ "0.3": {
14
+ "candidates": 1601,
15
+ "sampled": 60,
16
+ "rejected": 25,
17
+ "accepted": 23,
18
+ "undecided": 12,
19
+ "precision": 0.4791666666666667
20
+ },
21
+ "0.5": {
22
+ "candidates": 1014,
23
+ "sampled": 60,
24
+ "accepted": 37,
25
+ "undecided": 11,
26
+ "rejected": 12,
27
+ "precision": 0.7551020408163265
28
+ },
29
+ "0.7": {
30
+ "candidates": 523,
31
+ "sampled": 60,
32
+ "accepted": 31,
33
+ "undecided": 13,
34
+ "rejected": 16,
35
+ "precision": 0.6595744680851063
36
+ },
37
+ "0.8": {
38
+ "candidates": 414,
39
+ "sampled": 60,
40
+ "accepted": 39,
41
+ "undecided": 17,
42
+ "rejected": 4,
43
+ "precision": 0.9069767441860465
44
+ },
45
+ "0.9": {
46
+ "candidates": 278,
47
+ "sampled": 60,
48
+ "accepted": 43,
49
+ "undecided": 14,
50
+ "rejected": 3,
51
+ "precision": 0.9347826086956522
52
+ }
53
+ },
54
+ "judge_cost_usd": 0.5873,
55
+ "slices": {
56
+ "main": {
57
+ "whisper": {
58
+ "n_examples": 1978,
59
+ "wer": 0.19286111707900203,
60
+ "wer_zwnj_normalized": 0.18077200636819868,
61
+ "exact_match": 0.006066734074823054,
62
+ "hallucination_rate": 0.0,
63
+ "rows_changed": 0.0,
64
+ "fix_rate": 0.0,
65
+ "preservation_rate": 1.0,
66
+ "targeted_score": 0.5,
67
+ "fix_rate_lenient": 0.0,
68
+ "preservation_rate_lenient": 1.0,
69
+ "targeted_score_lenient": 0.5,
70
+ "fixed_lenient": 0,
71
+ "broken_lenient": 0,
72
+ "net_fixed_lenient": 0,
73
+ "broken_per_100_correct": 0.0,
74
+ "wer_clitic": 0.1839621023260069,
75
+ "fixed_clitic": 0,
76
+ "broken_clitic": 0,
77
+ "net_fixed_clitic": 0,
78
+ "broken_per_100_correct_clitic": 0.0,
79
+ "fix_rate_recoverable": 0.0,
80
+ "unrecoverable_targets": 966,
81
+ "entity_fix_rate": 0.0,
82
+ "entity_keep_rate": 1.0,
83
+ "entity_spans_whisper_wrong": 724,
84
+ "entity_spans_whisper_right": 1289,
85
+ "applied": 0,
86
+ "judged_precision": null,
87
+ "est_real_fixes": null
88
+ },
89
+ "rules only": {
90
+ "n_examples": 1978,
91
+ "wer": 0.18624502140177493,
92
+ "wer_zwnj_normalized": 0.17446440474419783,
93
+ "exact_match": 0.006572295247724975,
94
+ "hallucination_rate": 0.0,
95
+ "rows_changed": 0.6875631951466128,
96
+ "fix_rate": 0.05559820155718829,
97
+ "preservation_rate": 0.9996089846635096,
98
+ "targeted_score": 0.5276035931103489,
99
+ "fix_rate_lenient": 0.05892479464856623,
100
+ "preservation_rate_lenient": 0.9995903045402917,
101
+ "targeted_score_lenient": 0.5292575495944289,
102
+ "fixed_lenient": 3594,
103
+ "broken_lenient": 123,
104
+ "net_fixed_lenient": 3471,
105
+ "broken_per_100_correct": 0.04096954597082835,
106
+ "wer_clitic": 0.17667474784438397,
107
+ "fixed_clitic": 3571,
108
+ "broken_clitic": 137,
109
+ "net_fixed_clitic": 3434,
110
+ "broken_per_100_correct_clitic": 0.0468321630989902,
111
+ "fix_rate_recoverable": 0.059573192063571395,
112
+ "unrecoverable_targets": 966,
113
+ "entity_fix_rate": 0.009668508287292817,
114
+ "entity_keep_rate": 0.9953452288595811,
115
+ "entity_spans_whisper_wrong": 724,
116
+ "entity_spans_whisper_right": 1289,
117
+ "applied": 0,
118
+ "judged_precision": null,
119
+ "est_real_fixes": null
120
+ },
121
+ "accept all": {
122
+ "n_examples": 1978,
123
+ "wer": 0.1833225911296942,
124
+ "wer_zwnj_normalized": 0.17195238173935556,
125
+ "exact_match": 0.006572295247724975,
126
+ "hallucination_rate": 0.0,
127
+ "rows_changed": 0.9408493427704753,
128
+ "fix_rate": 0.12117556749643602,
129
+ "preservation_rate": 0.9893423255720688,
130
+ "targeted_score": 0.5552589465342525,
131
+ "fix_rate_lenient": 0.12734248192415523,
132
+ "preservation_rate_lenient": 0.9893679031919607,
133
+ "targeted_score_lenient": 0.558355192558058,
134
+ "fixed_lenient": 7767,
135
+ "broken_lenient": 3192,
136
+ "net_fixed_lenient": 4575,
137
+ "broken_per_100_correct": 1.0632096808039357,
138
+ "wer_clitic": 0.17458488667291813,
139
+ "fixed_clitic": 7419,
140
+ "broken_clitic": 3153,
141
+ "net_fixed_clitic": 4266,
142
+ "broken_per_100_correct_clitic": 1.0778234324898985,
143
+ "fix_rate_recoverable": 0.12885868026054942,
144
+ "unrecoverable_targets": 966,
145
+ "entity_fix_rate": 0.11187845303867404,
146
+ "entity_keep_rate": 0.9317300232738557,
147
+ "entity_spans_whisper_wrong": 724,
148
+ "entity_spans_whisper_right": 1289,
149
+ "applied": 8781,
150
+ "judged_precision": 0.3736924390763773,
151
+ "est_real_fixes": 3281.393307529669
152
+ },
153
+ "verifier@0.3": {
154
+ "n_examples": 1978,
155
+ "wer": 0.1851102039916927,
156
+ "wer_zwnj_normalized": 0.17342048290336975,
157
+ "exact_match": 0.006572295247724975,
158
+ "hallucination_rate": 0.0,
159
+ "rows_changed": 0.8650151668351871,
160
+ "fix_rate": 0.08191687685053185,
161
+ "preservation_rate": 0.9955283887160327,
162
+ "targeted_score": 0.5387226327832823,
163
+ "fix_rate_lenient": 0.08650172970668765,
164
+ "preservation_rate_lenient": 0.995533320231961,
165
+ "targeted_score_lenient": 0.5410175249693243,
166
+ "fixed_lenient": 5276,
167
+ "broken_lenient": 1341,
168
+ "net_fixed_lenient": 3935,
169
+ "broken_per_100_correct": 0.4466679768039091,
170
+ "wer_clitic": 0.1757055782997507,
171
+ "fixed_clitic": 5170,
172
+ "broken_clitic": 1335,
173
+ "net_fixed_clitic": 3835,
174
+ "broken_per_100_correct_clitic": 0.456357209760233,
175
+ "fix_rate_recoverable": 0.08754393856098089,
176
+ "unrecoverable_targets": 966,
177
+ "entity_fix_rate": 0.07044198895027624,
178
+ "entity_keep_rate": 0.9650892164468581,
179
+ "entity_spans_whisper_wrong": 724,
180
+ "entity_spans_whisper_right": 1289,
181
+ "applied": 3830,
182
+ "judged_precision": 0.656170936511753,
183
+ "est_real_fixes": 2513.1346868400137
184
+ },
185
+ "verifier@0.5": {
186
+ "n_examples": 1978,
187
+ "wer": 0.18571066563100805,
188
+ "wer_zwnj_normalized": 0.1739906460196004,
189
+ "exact_match": 0.006572295247724975,
190
+ "hallucination_rate": 0.0,
191
+ "rows_changed": 0.8124368048533872,
192
+ "fix_rate": 0.06968182601475725,
193
+ "preservation_rate": 0.997286286724528,
194
+ "targeted_score": 0.5334840563696426,
195
+ "fix_rate_lenient": 0.07361500500057383,
196
+ "preservation_rate_lenient": 0.9972753586500701,
197
+ "targeted_score_lenient": 0.535445181825322,
198
+ "fixed_lenient": 4490,
199
+ "broken_lenient": 818,
200
+ "net_fixed_lenient": 3672,
201
+ "broken_per_100_correct": 0.27246413499298855,
202
+ "wer_clitic": 0.1762344761338426,
203
+ "fixed_clitic": 4406,
204
+ "broken_clitic": 806,
205
+ "net_fixed_clitic": 3600,
206
+ "broken_per_100_correct_clitic": 0.27552352888894965,
207
+ "fix_rate_recoverable": 0.07449980841954454,
208
+ "unrecoverable_targets": 966,
209
+ "entity_fix_rate": 0.055248618784530384,
210
+ "entity_keep_rate": 0.9806051202482544,
211
+ "entity_spans_whisper_wrong": 724,
212
+ "entity_spans_whisper_right": 1289,
213
+ "applied": 2229,
214
+ "judged_precision": 0.7833059010797129,
215
+ "est_real_fixes": 1745.98885350668
216
+ },
217
+ "verifier@0.7": {
218
+ "n_examples": 1978,
219
+ "wer": 0.18586491265762117,
220
+ "wer_zwnj_normalized": 0.17412010334550784,
221
+ "exact_match": 0.006572295247724975,
222
+ "hallucination_rate": 0.0,
223
+ "rows_changed": 0.7639029322548029,
224
+ "fix_rate": 0.06300816192251657,
225
+ "preservation_rate": 0.9984559907225763,
226
+ "targeted_score": 0.5307320763225465,
227
+ "fix_rate_lenient": 0.06658141098158805,
228
+ "preservation_rate_lenient": 0.9984444895960669,
229
+ "targeted_score_lenient": 0.5325129502888275,
230
+ "fixed_lenient": 4061,
231
+ "broken_lenient": 467,
232
+ "net_fixed_lenient": 3594,
233
+ "broken_per_100_correct": 0.15555104039330764,
234
+ "wer_clitic": 0.17632882007181575,
235
+ "fixed_clitic": 4011,
236
+ "broken_clitic": 473,
237
+ "net_fixed_clitic": 3538,
238
+ "broken_per_100_correct_clitic": 0.1616906069038129,
239
+ "fix_rate_recoverable": 0.06733636530228064,
240
+ "unrecoverable_targets": 966,
241
+ "entity_fix_rate": 0.03729281767955801,
242
+ "entity_keep_rate": 0.9852598913886734,
243
+ "entity_spans_whisper_wrong": 724,
244
+ "entity_spans_whisper_right": 1289,
245
+ "applied": 1215,
246
+ "judged_precision": 0.8068439375464403,
247
+ "est_real_fixes": 980.3153841189251
248
+ },
249
+ "verifier@0.8": {
250
+ "n_examples": 1978,
251
+ "wer": 0.18587868471356878,
252
+ "wer_zwnj_normalized": 0.17410908570074976,
253
+ "exact_match": 0.006572295247724975,
254
+ "hallucination_rate": 0.0,
255
+ "rows_changed": 0.7320525783619818,
256
+ "fix_rate": 0.059890652170507416,
257
+ "preservation_rate": 0.9989773445045635,
258
+ "targeted_score": 0.5294339983375355,
259
+ "fix_rate_lenient": 0.06333513681897923,
260
+ "preservation_rate_lenient": 0.9989641033498433,
261
+ "targeted_score_lenient": 0.5311496200844112,
262
+ "fixed_lenient": 3863,
263
+ "broken_lenient": 311,
264
+ "net_fixed_lenient": 3552,
265
+ "broken_per_100_correct": 0.10358966501567168,
266
+ "wer_clitic": 0.17634025570066097,
267
+ "fixed_clitic": 3828,
268
+ "broken_clitic": 317,
269
+ "net_fixed_clitic": 3511,
270
+ "broken_per_100_correct_clitic": 0.10836347228014521,
271
+ "fix_rate_recoverable": 0.06403784963433122,
272
+ "unrecoverable_targets": 966,
273
+ "entity_fix_rate": 0.023480662983425413,
274
+ "entity_keep_rate": 0.9899146625290923,
275
+ "entity_spans_whisper_wrong": 724,
276
+ "entity_spans_whisper_right": 1289,
277
+ "applied": 692,
278
+ "judged_precision": 0.9181473082520443,
279
+ "est_real_fixes": 635.3579373104146
280
+ },
281
+ "verifier@0.9": {
282
+ "n_examples": 1978,
283
+ "wer": 0.18608251114159327,
284
+ "wer_zwnj_normalized": 0.1743074033063952,
285
+ "exact_match": 0.006572295247724975,
286
+ "hallucination_rate": 0.0,
287
+ "rows_changed": 0.7087967644084934,
288
+ "fix_rate": 0.05729011639747466,
289
+ "preservation_rate": 0.999338281738247,
290
+ "targeted_score": 0.5283141990678608,
291
+ "fix_rate_lenient": 0.0606626989982457,
292
+ "preservation_rate_lenient": 0.9993304976634035,
293
+ "targeted_score_lenient": 0.5299965983308246,
294
+ "fixed_lenient": 3700,
295
+ "broken_lenient": 201,
296
+ "net_fixed_lenient": 3499,
297
+ "broken_per_100_correct": 0.06695023365964634,
298
+ "wer_clitic": 0.17650321341170552,
299
+ "fixed_clitic": 3675,
300
+ "broken_clitic": 211,
301
+ "net_fixed_clitic": 3464,
302
+ "broken_per_100_correct_clitic": 0.07212836798457616,
303
+ "fix_rate_recoverable": 0.06133906408782715,
304
+ "unrecoverable_targets": 966,
305
+ "entity_fix_rate": 0.012430939226519336,
306
+ "entity_keep_rate": 0.9930178432893716,
307
+ "entity_spans_whisper_wrong": 724,
308
+ "entity_spans_whisper_right": 1289,
309
+ "applied": 278,
310
+ "judged_precision": 0.9347826086956522,
311
+ "est_real_fixes": 259.8695652173913
312
+ },
313
+ "oracle": {
314
+ "n_examples": 1978,
315
+ "wer": 0.17497397081425903,
316
+ "wer_zwnj_normalized": 0.16355142761131952,
317
+ "exact_match": 0.007077856420626896,
318
+ "hallucination_rate": 0.0,
319
+ "rows_changed": 0.8624873609706775,
320
+ "fix_rate": 0.11846537057634765,
321
+ "preservation_rate": 0.999381727886746,
322
+ "targeted_score": 0.5589235492315467,
323
+ "fix_rate_lenient": 0.12511271785286837,
324
+ "preservation_rate_lenient": 0.9994737245314317,
325
+ "targeted_score_lenient": 0.5622932211921501,
326
+ "fixed_lenient": 7631,
327
+ "broken_lenient": 158,
328
+ "net_fixed_lenient": 7473,
329
+ "broken_per_100_correct": 0.05262754685683642,
330
+ "wer_clitic": 0.16583948951352834,
331
+ "fixed_clitic": 7318,
332
+ "broken_clitic": 171,
333
+ "net_fixed_clitic": 7147,
334
+ "broken_per_100_correct_clitic": 0.05845474372209726,
335
+ "fix_rate_recoverable": 0.1265930331350892,
336
+ "unrecoverable_targets": 966,
337
+ "entity_fix_rate": 0.11464088397790055,
338
+ "entity_keep_rate": 0.9953452288595811,
339
+ "entity_spans_whisper_wrong": 724,
340
+ "entity_spans_whisper_right": 1289,
341
+ "applied": 3443,
342
+ "judged_precision": null,
343
+ "est_real_fixes": null
344
+ }
345
+ },
346
+ "entity": {
347
+ "whisper": {
348
+ "n_examples": 1010,
349
+ "wer": 0.18660667584675655,
350
+ "wer_zwnj_normalized": 0.17409613475517982,
351
+ "exact_match": 0.0049504950495049506,
352
+ "hallucination_rate": 0.0,
353
+ "rows_changed": 0.0,
354
+ "fix_rate": 0.0,
355
+ "preservation_rate": 1.0,
356
+ "targeted_score": 0.5,
357
+ "fix_rate_lenient": 0.0,
358
+ "preservation_rate_lenient": 1.0,
359
+ "targeted_score_lenient": 0.5,
360
+ "fixed_lenient": 0,
361
+ "broken_lenient": 0,
362
+ "net_fixed_lenient": 0,
363
+ "broken_per_100_correct": 0.0,
364
+ "wer_clitic": 0.1776189974667827,
365
+ "fixed_clitic": 0,
366
+ "broken_clitic": 0,
367
+ "net_fixed_clitic": 0,
368
+ "broken_per_100_correct_clitic": 0.0,
369
+ "fix_rate_recoverable": 0.0,
370
+ "unrecoverable_targets": 471,
371
+ "entity_fix_rate": 0.0,
372
+ "entity_keep_rate": 1.0,
373
+ "entity_spans_whisper_wrong": 724,
374
+ "entity_spans_whisper_right": 1289,
375
+ "applied": 0,
376
+ "judged_precision": null,
377
+ "est_real_fixes": null
378
+ },
379
+ "rules only": {
380
+ "n_examples": 1010,
381
+ "wer": 0.18000982056126644,
382
+ "wer_zwnj_normalized": 0.16776614254758168,
383
+ "exact_match": 0.0049504950495049506,
384
+ "hallucination_rate": 0.0,
385
+ "rows_changed": 0.7148514851485148,
386
+ "fix_rate": 0.0566740770892695,
387
+ "preservation_rate": 0.9996911693742963,
388
+ "targeted_score": 0.5281826232317829,
389
+ "fix_rate_lenient": 0.06019704433497537,
390
+ "preservation_rate_lenient": 0.9996856677315474,
391
+ "targeted_score_lenient": 0.5299413560332614,
392
+ "fixed_lenient": 1833,
393
+ "broken_lenient": 49,
394
+ "net_fixed_lenient": 1784,
395
+ "broken_per_100_correct": 0.03143322684525871,
396
+ "wer_clitic": 0.1703353048452631,
397
+ "fixed_clitic": 1824,
398
+ "broken_clitic": 61,
399
+ "net_fixed_clitic": 1763,
400
+ "broken_per_100_correct_clitic": 0.04016434459690801,
401
+ "fix_rate_recoverable": 0.06077587644684613,
402
+ "unrecoverable_targets": 471,
403
+ "entity_fix_rate": 0.009668508287292817,
404
+ "entity_keep_rate": 0.9953452288595811,
405
+ "entity_spans_whisper_wrong": 724,
406
+ "entity_spans_whisper_right": 1289,
407
+ "applied": 0,
408
+ "judged_precision": null,
409
+ "est_real_fixes": null
410
+ },
411
+ "accept all": {
412
+ "n_examples": 1010,
413
+ "wer": 0.17671139291852136,
414
+ "wer_zwnj_normalized": 0.16492671939881084,
415
+ "exact_match": 0.0039603960396039604,
416
+ "hallucination_rate": 0.0,
417
+ "rows_changed": 0.9574257425742574,
418
+ "fix_rate": 0.12624855183642797,
419
+ "preservation_rate": 0.989673475953032,
420
+ "targeted_score": 0.55796101389473,
421
+ "fix_rate_lenient": 0.1329064039408867,
422
+ "preservation_rate_lenient": 0.9897745788589097,
423
+ "targeted_score_lenient": 0.5613404913998983,
424
+ "fixed_lenient": 4047,
425
+ "broken_lenient": 1594,
426
+ "net_fixed_lenient": 2453,
427
+ "broken_per_100_correct": 1.0225421141090283,
428
+ "wer_clitic": 0.16794066617517447,
429
+ "fixed_clitic": 3847,
430
+ "broken_clitic": 1589,
431
+ "net_fixed_clitic": 2258,
432
+ "broken_per_100_correct_clitic": 1.0462482551555217,
433
+ "fix_rate_recoverable": 0.13439407585309718,
434
+ "unrecoverable_targets": 471,
435
+ "entity_fix_rate": 0.11187845303867404,
436
+ "entity_keep_rate": 0.9317300232738557,
437
+ "entity_spans_whisper_wrong": 724,
438
+ "entity_spans_whisper_right": 1289,
439
+ "applied": 4473,
440
+ "judged_precision": null,
441
+ "est_real_fixes": null
442
+ },
443
+ "verifier@0.3": {
444
+ "n_examples": 1010,
445
+ "wer": 0.17885163480321517,
446
+ "wer_zwnj_normalized": 0.1667253765438029,
447
+ "exact_match": 0.0049504950495049506,
448
+ "hallucination_rate": 0.0,
449
+ "rows_changed": 0.8782178217821782,
450
+ "fix_rate": 0.08447881767229233,
451
+ "preservation_rate": 0.9955348238700338,
452
+ "targeted_score": 0.540006820771163,
453
+ "fix_rate_lenient": 0.08939244663382595,
454
+ "preservation_rate_lenient": 0.9955736884646472,
455
+ "targeted_score_lenient": 0.5424830675492366,
456
+ "fixed_lenient": 2722,
457
+ "broken_lenient": 690,
458
+ "net_fixed_lenient": 2032,
459
+ "broken_per_100_correct": 0.44263115353527577,
460
+ "wer_clitic": 0.16931536615244758,
461
+ "fixed_clitic": 2668,
462
+ "broken_clitic": 696,
463
+ "net_fixed_clitic": 1972,
464
+ "broken_per_100_correct_clitic": 0.45826858753193395,
465
+ "fix_rate_recoverable": 0.09039661096100604,
466
+ "unrecoverable_targets": 471,
467
+ "entity_fix_rate": 0.07044198895027624,
468
+ "entity_keep_rate": 0.9650892164468581,
469
+ "entity_spans_whisper_wrong": 724,
470
+ "entity_spans_whisper_right": 1289,
471
+ "applied": 1979,
472
+ "judged_precision": null,
473
+ "est_real_fixes": null
474
+ },
475
+ "verifier@0.5": {
476
+ "n_examples": 1010,
477
+ "wer": 0.17940137274367268,
478
+ "wer_zwnj_normalized": 0.16722174186868202,
479
+ "exact_match": 0.0049504950495049506,
480
+ "hallucination_rate": 0.0,
481
+ "rows_changed": 0.8346534653465346,
482
+ "fix_rate": 0.07135923850079845,
483
+ "preservation_rate": 0.997355637767412,
484
+ "targeted_score": 0.5343574381341052,
485
+ "fix_rate_lenient": 0.07556650246305419,
486
+ "preservation_rate_lenient": 0.9973634579115508,
487
+ "targeted_score_lenient": 0.5364649801873025,
488
+ "fixed_lenient": 2301,
489
+ "broken_lenient": 411,
490
+ "net_fixed_lenient": 1890,
491
+ "broken_per_100_correct": 0.26365420884492513,
492
+ "wer_clitic": 0.16983642178899463,
493
+ "fixed_clitic": 2263,
494
+ "broken_clitic": 411,
495
+ "net_fixed_clitic": 1852,
496
+ "broken_per_100_correct_clitic": 0.2706155021201507,
497
+ "fix_rate_recoverable": 0.07642016077921211,
498
+ "unrecoverable_targets": 471,
499
+ "entity_fix_rate": 0.055248618784530384,
500
+ "entity_keep_rate": 0.9806051202482544,
501
+ "entity_spans_whisper_wrong": 724,
502
+ "entity_spans_whisper_right": 1289,
503
+ "applied": 1143,
504
+ "judged_precision": null,
505
+ "est_real_fixes": null
506
+ },
507
+ "verifier@0.7": {
508
+ "n_examples": 1010,
509
+ "wer": 0.17956149059040788,
510
+ "wer_zwnj_normalized": 0.1673391616229545,
511
+ "exact_match": 0.0049504950495049506,
512
+ "hallucination_rate": 0.0,
513
+ "rows_changed": 0.7881188118811882,
514
+ "fix_rate": 0.06443936499984344,
515
+ "preservation_rate": 0.9985394884992762,
516
+ "targeted_score": 0.5314894267495598,
517
+ "fix_rate_lenient": 0.06824302134646962,
518
+ "preservation_rate_lenient": 0.9985438076543115,
519
+ "targeted_score_lenient": 0.5333934145003906,
520
+ "fixed_lenient": 2078,
521
+ "broken_lenient": 227,
522
+ "net_fixed_lenient": 1851,
523
+ "broken_per_100_correct": 0.1456192345688516,
524
+ "wer_clitic": 0.16994728469038764,
525
+ "fixed_clitic": 2055,
526
+ "broken_clitic": 231,
527
+ "net_fixed_clitic": 1824,
528
+ "broken_per_100_correct_clitic": 0.15209776396534014,
529
+ "fix_rate_recoverable": 0.06894826378464926,
530
+ "unrecoverable_targets": 471,
531
+ "entity_fix_rate": 0.03729281767955801,
532
+ "entity_keep_rate": 0.9852598913886734,
533
+ "entity_spans_whisper_wrong": 724,
534
+ "entity_spans_whisper_right": 1289,
535
+ "applied": 627,
536
+ "judged_precision": null,
537
+ "est_real_fixes": null
538
+ },
539
+ "verifier@0.8": {
540
+ "n_examples": 1010,
541
+ "wer": 0.17964154951377548,
542
+ "wer_zwnj_normalized": 0.16740320876164858,
543
+ "exact_match": 0.0049504950495049506,
544
+ "hallucination_rate": 0.0,
545
+ "rows_changed": 0.7534653465346535,
546
+ "fix_rate": 0.061182953940570495,
547
+ "preservation_rate": 0.9990542062087824,
548
+ "targeted_score": 0.5301185800746764,
549
+ "fix_rate_lenient": 0.06482758620689655,
550
+ "preservation_rate_lenient": 0.9990505882503881,
551
+ "targeted_score_lenient": 0.5319390872286422,
552
+ "fixed_lenient": 1974,
553
+ "broken_lenient": 148,
554
+ "net_fixed_lenient": 1826,
555
+ "broken_per_100_correct": 0.09494117496118959,
556
+ "wer_clitic": 0.17000271614108411,
557
+ "fixed_clitic": 1959,
558
+ "broken_clitic": 153,
559
+ "net_fixed_clitic": 1806,
560
+ "broken_per_100_correct_clitic": 0.10074007743158893,
561
+ "fix_rate_recoverable": 0.06547916875145936,
562
+ "unrecoverable_targets": 471,
563
+ "entity_fix_rate": 0.023480662983425413,
564
+ "entity_keep_rate": 0.9899146625290923,
565
+ "entity_spans_whisper_wrong": 724,
566
+ "entity_spans_whisper_right": 1289,
567
+ "applied": 357,
568
+ "judged_precision": null,
569
+ "est_real_fixes": null
570
+ },
571
+ "verifier@0.9": {
572
+ "n_examples": 1010,
573
+ "wer": 0.17986571449920474,
574
+ "wer_zwnj_normalized": 0.16762737374707784,
575
+ "exact_match": 0.0049504950495049506,
576
+ "hallucination_rate": 0.0,
577
+ "rows_changed": 0.7356435643564356,
578
+ "fix_rate": 0.05827097097410527,
579
+ "preservation_rate": 0.9994209425768056,
580
+ "targeted_score": 0.5288459567754554,
581
+ "fix_rate_lenient": 0.061839080459770115,
582
+ "preservation_rate_lenient": 0.9994226550171279,
583
+ "targeted_score_lenient": 0.530630867738449,
584
+ "fixed_lenient": 1883,
585
+ "broken_lenient": 90,
586
+ "net_fixed_lenient": 1793,
587
+ "broken_per_100_correct": 0.05773449828720988,
588
+ "wer_clitic": 0.17016901049317362,
589
+ "fixed_clitic": 1874,
590
+ "broken_clitic": 98,
591
+ "net_fixed_clitic": 1776,
592
+ "broken_per_100_correct_clitic": 0.06452632410650794,
593
+ "fix_rate_recoverable": 0.06244371059741819,
594
+ "unrecoverable_targets": 471,
595
+ "entity_fix_rate": 0.012430939226519336,
596
+ "entity_keep_rate": 0.9930178432893716,
597
+ "entity_spans_whisper_wrong": 724,
598
+ "entity_spans_whisper_right": 1289,
599
+ "applied": 139,
600
+ "judged_precision": null,
601
+ "est_real_fixes": null
602
+ },
603
+ "oracle": {
604
+ "n_examples": 1010,
605
+ "wer": 0.16844397476542736,
606
+ "wer_zwnj_normalized": 0.15656856779923356,
607
+ "exact_match": 0.0049504950495049506,
608
+ "hallucination_rate": 0.0,
609
+ "rows_changed": 0.897029702970297,
610
+ "fix_rate": 0.12364968531796976,
611
+ "preservation_rate": 0.9993759047772237,
612
+ "targeted_score": 0.5615127950475968,
613
+ "fix_rate_lenient": 0.13054187192118227,
614
+ "preservation_rate_lenient": 0.9994932194039233,
615
+ "targeted_score_lenient": 0.5650175456625528,
616
+ "fixed_lenient": 3975,
617
+ "broken_lenient": 79,
618
+ "net_fixed_lenient": 3896,
619
+ "broken_per_100_correct": 0.05067805960766201,
620
+ "wer_clitic": 0.1593321618820086,
621
+ "fixed_clitic": 3796,
622
+ "broken_clitic": 91,
623
+ "net_fixed_clitic": 3705,
624
+ "broken_per_100_correct_clitic": 0.059917300956043085,
625
+ "fix_rate_recoverable": 0.13195903799326195,
626
+ "unrecoverable_targets": 471,
627
+ "entity_fix_rate": 0.11464088397790055,
628
+ "entity_keep_rate": 0.9953452288595811,
629
+ "entity_spans_whisper_wrong": 724,
630
+ "entity_spans_whisper_right": 1289,
631
+ "applied": 1804,
632
+ "judged_precision": null,
633
+ "est_real_fixes": null
634
+ }
635
+ },
636
+ "typo": {
637
+ "whisper": {
638
+ "n_examples": 358,
639
+ "wer": 0.14117586990375533,
640
+ "wer_zwnj_normalized": 0.12749227745015446,
641
+ "exact_match": 0.0,
642
+ "hallucination_rate": 0.0,
643
+ "rows_changed": 0.0,
644
+ "fix_rate": 0.0,
645
+ "preservation_rate": 1.0,
646
+ "targeted_score": 0.5,
647
+ "fix_rate_lenient": 0.0,
648
+ "preservation_rate_lenient": 1.0,
649
+ "targeted_score_lenient": 0.5,
650
+ "fixed_lenient": 0,
651
+ "broken_lenient": 0,
652
+ "net_fixed_lenient": 0,
653
+ "broken_per_100_correct": 0.0,
654
+ "wer_clitic": 0.1316430880093255,
655
+ "fixed_clitic": 0,
656
+ "broken_clitic": 0,
657
+ "net_fixed_clitic": 0,
658
+ "broken_per_100_correct_clitic": 0.0,
659
+ "fix_rate_recoverable": 0.0,
660
+ "unrecoverable_targets": 68,
661
+ "entity_fix_rate": 0.0,
662
+ "entity_keep_rate": 1.0,
663
+ "entity_spans_whisper_wrong": 90,
664
+ "entity_spans_whisper_right": 215,
665
+ "applied": 0,
666
+ "judged_precision": null,
667
+ "est_real_fixes": null
668
+ },
669
+ "rules only": {
670
+ "n_examples": 358,
671
+ "wer": 0.13481912639452656,
672
+ "wer_zwnj_normalized": 0.12139082484491077,
673
+ "exact_match": 0.002793296089385475,
674
+ "hallucination_rate": 0.0,
675
+ "rows_changed": 0.4776536312849162,
676
+ "fix_rate": 0.06630476929042264,
677
+ "preservation_rate": 0.9998823979066828,
678
+ "targeted_score": 0.5330935835985527,
679
+ "fix_rate_lenient": 0.07141378596563859,
680
+ "preservation_rate_lenient": 0.9998827976207917,
681
+ "targeted_score_lenient": 0.5356482917932152,
682
+ "fixed_lenient": 345,
683
+ "broken_lenient": 4,
684
+ "net_fixed_lenient": 341,
685
+ "broken_per_100_correct": 0.011720237920829792,
686
+ "wer_clitic": 0.12475494092089227,
687
+ "fixed_clitic": 342,
688
+ "broken_clitic": 7,
689
+ "net_fixed_clitic": 335,
690
+ "broken_per_100_correct_clitic": 0.021002100210021003,
691
+ "fix_rate_recoverable": 0.07222338862061725,
692
+ "unrecoverable_targets": 68,
693
+ "entity_fix_rate": 0.011111111111111112,
694
+ "entity_keep_rate": 1.0,
695
+ "entity_spans_whisper_wrong": 90,
696
+ "entity_spans_whisper_right": 215,
697
+ "applied": 0,
698
+ "judged_precision": null,
699
+ "est_real_fixes": null
700
+ },
701
+ "accept all": {
702
+ "n_examples": 358,
703
+ "wer": 0.13425748640575935,
704
+ "wer_zwnj_normalized": 0.12116106303132419,
705
+ "exact_match": 0.00558659217877095,
706
+ "hallucination_rate": 0.0,
707
+ "rows_changed": 0.8519553072625698,
708
+ "fix_rate": 0.13900736719658782,
709
+ "preservation_rate": 0.9904448299179726,
710
+ "targeted_score": 0.5647260985572802,
711
+ "fix_rate_lenient": 0.14986545228731113,
712
+ "preservation_rate_lenient": 0.9904480060945238,
713
+ "targeted_score_lenient": 0.5701567291909174,
714
+ "fixed_lenient": 724,
715
+ "broken_lenient": 326,
716
+ "net_fixed_lenient": 398,
717
+ "broken_per_100_correct": 0.9551993905476281,
718
+ "wer_clitic": 0.12475494092089227,
719
+ "fixed_clitic": 688,
720
+ "broken_clitic": 322,
721
+ "net_fixed_clitic": 366,
722
+ "broken_per_100_correct_clitic": 0.9660966096609661,
723
+ "fix_rate_recoverable": 0.15179508712996012,
724
+ "unrecoverable_targets": 68,
725
+ "entity_fix_rate": 0.16666666666666666,
726
+ "entity_keep_rate": 0.9441860465116279,
727
+ "entity_spans_whisper_wrong": 90,
728
+ "entity_spans_whisper_right": 215,
729
+ "applied": 811,
730
+ "judged_precision": null,
731
+ "est_real_fixes": null
732
+ },
733
+ "verifier@0.3": {
734
+ "n_examples": 358,
735
+ "wer": 0.13443619003854893,
736
+ "wer_zwnj_normalized": 0.1209057721273391,
737
+ "exact_match": 0.002793296089385475,
738
+ "hallucination_rate": 0.0,
739
+ "rows_changed": 0.6871508379888268,
740
+ "fix_rate": 0.0934470725087243,
741
+ "preservation_rate": 0.996648340340458,
742
+ "targeted_score": 0.5450477064245911,
743
+ "fix_rate_lenient": 0.10060028979507349,
744
+ "preservation_rate_lenient": 0.9966890327873655,
745
+ "targeted_score_lenient": 0.5486446612912195,
746
+ "fixed_lenient": 486,
747
+ "broken_lenient": 113,
748
+ "net_fixed_lenient": 373,
749
+ "broken_per_100_correct": 0.33109672126344164,
750
+ "wer_clitic": 0.12456949080697292,
751
+ "fixed_clitic": 473,
752
+ "broken_clitic": 116,
753
+ "net_fixed_clitic": 357,
754
+ "broken_per_100_correct_clitic": 0.34803480348034804,
755
+ "fix_rate_recoverable": 0.10203653159773252,
756
+ "unrecoverable_targets": 68,
757
+ "entity_fix_rate": 0.1111111111111111,
758
+ "entity_keep_rate": 0.9813953488372092,
759
+ "entity_spans_whisper_wrong": 90,
760
+ "entity_spans_whisper_right": 215,
761
+ "applied": 334,
762
+ "judged_precision": null,
763
+ "est_real_fixes": null
764
+ },
765
+ "verifier@0.5": {
766
+ "n_examples": 358,
767
+ "wer": 0.1346659518521355,
768
+ "wer_zwnj_normalized": 0.12116106303132419,
769
+ "exact_match": 0.002793296089385475,
770
+ "hallucination_rate": 0.0,
771
+ "rows_changed": 0.6229050279329609,
772
+ "fix_rate": 0.0794881737107406,
773
+ "preservation_rate": 0.9983535706935583,
774
+ "targeted_score": 0.5389208722021495,
775
+ "fix_rate_lenient": 0.08548954667770647,
776
+ "preservation_rate_lenient": 0.9983298660962817,
777
+ "targeted_score_lenient": 0.5419097063869941,
778
+ "fixed_lenient": 413,
779
+ "broken_lenient": 57,
780
+ "net_fixed_lenient": 356,
781
+ "broken_per_100_correct": 0.16701339037182455,
782
+ "wer_clitic": 0.12475494092089227,
783
+ "fixed_clitic": 404,
784
+ "broken_clitic": 59,
785
+ "net_fixed_clitic": 345,
786
+ "broken_per_100_correct_clitic": 0.177017701770177,
787
+ "fix_rate_recoverable": 0.086710056686962,
788
+ "unrecoverable_targets": 68,
789
+ "entity_fix_rate": 0.08888888888888889,
790
+ "entity_keep_rate": 0.9906976744186047,
791
+ "entity_spans_whisper_wrong": 90,
792
+ "entity_spans_whisper_right": 215,
793
+ "applied": 185,
794
+ "judged_precision": null,
795
+ "est_real_fixes": null
796
+ },
797
+ "verifier@0.7": {
798
+ "n_examples": 358,
799
+ "wer": 0.13471701003293252,
800
+ "wer_zwnj_normalized": 0.12126317939291822,
801
+ "exact_match": 0.002793296089385475,
802
+ "hallucination_rate": 0.0,
803
+ "rows_changed": 0.5614525139664804,
804
+ "fix_rate": 0.07270259790616518,
805
+ "preservation_rate": 0.9990885837767912,
806
+ "targeted_score": 0.5358955908414782,
807
+ "fix_rate_lenient": 0.07824466984061271,
808
+ "preservation_rate_lenient": 0.9990623809663336,
809
+ "targeted_score_lenient": 0.5386535254034731,
810
+ "fixed_lenient": 378,
811
+ "broken_lenient": 32,
812
+ "net_fixed_lenient": 346,
813
+ "broken_per_100_correct": 0.09376190336663834,
814
+ "wer_clitic": 0.12480792666772639,
815
+ "fixed_clitic": 370,
816
+ "broken_clitic": 36,
817
+ "net_fixed_clitic": 334,
818
+ "broken_per_100_correct_clitic": 0.10801080108010801,
819
+ "fix_rate_recoverable": 0.07915179508712997,
820
+ "unrecoverable_targets": 68,
821
+ "entity_fix_rate": 0.044444444444444446,
822
+ "entity_keep_rate": 0.9906976744186047,
823
+ "entity_spans_whisper_wrong": 90,
824
+ "entity_spans_whisper_right": 215,
825
+ "applied": 96,
826
+ "judged_precision": null,
827
+ "est_real_fixes": null
828
+ },
829
+ "verifier@0.8": {
830
+ "n_examples": 358,
831
+ "wer": 0.1346659518521355,
832
+ "wer_zwnj_normalized": 0.1211865921217227,
833
+ "exact_match": 0.002793296089385475,
834
+ "hallucination_rate": 0.0,
835
+ "rows_changed": 0.5167597765363129,
836
+ "fix_rate": 0.069988367584335,
837
+ "preservation_rate": 0.9994707905800724,
838
+ "targeted_score": 0.5347295790822036,
839
+ "fix_rate_lenient": 0.07513972262471538,
840
+ "preservation_rate_lenient": 0.9994432886987605,
841
+ "targeted_score_lenient": 0.537291505661738,
842
+ "fixed_lenient": 363,
843
+ "broken_lenient": 19,
844
+ "net_fixed_lenient": 344,
845
+ "broken_per_100_correct": 0.055671130123941516,
846
+ "wer_clitic": 0.12475494092089227,
847
+ "fixed_clitic": 357,
848
+ "broken_clitic": 22,
849
+ "net_fixed_clitic": 335,
850
+ "broken_per_100_correct_clitic": 0.066006600660066,
851
+ "fix_rate_recoverable": 0.07600251942053328,
852
+ "unrecoverable_targets": 68,
853
+ "entity_fix_rate": 0.03333333333333333,
854
+ "entity_keep_rate": 0.9953488372093023,
855
+ "entity_spans_whisper_wrong": 90,
856
+ "entity_spans_whisper_right": 215,
857
+ "applied": 52,
858
+ "judged_precision": null,
859
+ "est_real_fixes": null
860
+ },
861
+ "verifier@0.9": {
862
+ "n_examples": 358,
863
+ "wer": 0.1346659518521355,
864
+ "wer_zwnj_normalized": 0.12123765030251972,
865
+ "exact_match": 0.002793296089385475,
866
+ "hallucination_rate": 0.0,
867
+ "rows_changed": 0.4972067039106145,
868
+ "fix_rate": 0.06746801085692129,
869
+ "preservation_rate": 0.9997941963366948,
870
+ "targeted_score": 0.533631103596808,
871
+ "fix_rate_lenient": 0.07265576485199751,
872
+ "preservation_rate_lenient": 0.9997948958363855,
873
+ "targeted_score_lenient": 0.5362253303441915,
874
+ "fixed_lenient": 351,
875
+ "broken_lenient": 7,
876
+ "net_fixed_lenient": 344,
877
+ "broken_per_100_correct": 0.02051041636145214,
878
+ "wer_clitic": 0.12456949080697292,
879
+ "fixed_clitic": 349,
880
+ "broken_clitic": 10,
881
+ "net_fixed_clitic": 339,
882
+ "broken_per_100_correct_clitic": 0.030003000300030003,
883
+ "fix_rate_recoverable": 0.07348309888725593,
884
+ "unrecoverable_targets": 68,
885
+ "entity_fix_rate": 0.011111111111111112,
886
+ "entity_keep_rate": 1.0,
887
+ "entity_spans_whisper_wrong": 90,
888
+ "entity_spans_whisper_right": 215,
889
+ "applied": 21,
890
+ "judged_precision": null,
891
+ "est_real_fixes": null
892
+ },
893
+ "oracle": {
894
+ "n_examples": 358,
895
+ "wer": 0.1254499502182737,
896
+ "wer_zwnj_normalized": 0.11240458502463557,
897
+ "exact_match": 0.00558659217877095,
898
+ "hallucination_rate": 0.0,
899
+ "rows_changed": 0.7262569832402235,
900
+ "fix_rate": 0.1355176424970919,
901
+ "preservation_rate": 0.9997647958133655,
902
+ "targeted_score": 0.5676412191552287,
903
+ "fix_rate_lenient": 0.14655350859035396,
904
+ "preservation_rate_lenient": 0.9998534970259896,
905
+ "targeted_score_lenient": 0.5732035028081718,
906
+ "fixed_lenient": 708,
907
+ "broken_lenient": 5,
908
+ "net_fixed_lenient": 703,
909
+ "broken_per_100_correct": 0.014650297401037241,
910
+ "wer_clitic": 0.11572087108567795,
911
+ "fixed_clitic": 673,
912
+ "broken_clitic": 8,
913
+ "net_fixed_clitic": 665,
914
+ "broken_per_100_correct_clitic": 0.024002400240024,
915
+ "fix_rate_recoverable": 0.14822590804115054,
916
+ "unrecoverable_targets": 68,
917
+ "entity_fix_rate": 0.16666666666666666,
918
+ "entity_keep_rate": 1.0,
919
+ "entity_spans_whisper_wrong": 90,
920
+ "entity_spans_whisper_right": 215,
921
+ "applied": 313,
922
+ "judged_precision": null,
923
+ "est_real_fixes": null
924
+ }
925
+ }
926
+ }
927
+ }
verifier-qwen3.5-2b-v2/report-20260929-1803/report.md ADDED
@@ -0,0 +1,44 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Verifier report: verifier-qwen3.5-2b-v2
2
+
3
+ 8781 candidates after dropping 0 clitic/style respellings. Judge cost $0.59. Judged precision: share of applied changes the calibrated judges call real errors (undecided excluded), estimated from a score-stratified sample.
4
+
5
+ | slice | method | WER | WER (clitic) | net | net (clitic) | broken/100 | broken/100 (clitic) | entity fix | applied | judged precision | est. real fixes |
6
+ |---|---|---|---|---|---|---|---|---|---|---|---|
7
+ | main | whisper | 0.1929 | 0.1840 | 0 | 0 | 0.00 | 0.00 | 0.000 | 0 | – | – |
8
+ | main | rules only | 0.1862 | 0.1767 | 3471 | 3434 | 0.04 | 0.05 | 0.010 | 0 | – | – |
9
+ | main | accept all | 0.1833 | 0.1746 | 4575 | 4266 | 1.06 | 1.08 | 0.112 | 8781 | 0.37 | 3281 |
10
+ | main | verifier@0.3 | 0.1851 | 0.1757 | 3935 | 3835 | 0.45 | 0.46 | 0.070 | 3830 | 0.66 | 2513 |
11
+ | main | verifier@0.5 | 0.1857 | 0.1762 | 3672 | 3600 | 0.27 | 0.28 | 0.055 | 2229 | 0.78 | 1746 |
12
+ | main | verifier@0.7 | 0.1859 | 0.1763 | 3594 | 3538 | 0.16 | 0.16 | 0.037 | 1215 | 0.81 | 980 |
13
+ | main | verifier@0.8 | 0.1859 | 0.1763 | 3552 | 3511 | 0.10 | 0.11 | 0.023 | 692 | 0.92 | 635 |
14
+ | main | verifier@0.9 | 0.1861 | 0.1765 | 3499 | 3464 | 0.07 | 0.07 | 0.012 | 278 | 0.93 | 260 |
15
+ | main | oracle | 0.1750 | 0.1658 | 7473 | 7147 | 0.05 | 0.06 | 0.115 | 3443 | – | – |
16
+ | entity | whisper | 0.1866 | 0.1776 | 0 | 0 | 0.00 | 0.00 | 0.000 | 0 | – | – |
17
+ | entity | rules only | 0.1800 | 0.1703 | 1784 | 1763 | 0.03 | 0.04 | 0.010 | 0 | – | – |
18
+ | entity | accept all | 0.1767 | 0.1679 | 2453 | 2258 | 1.02 | 1.05 | 0.112 | 4473 | – | – |
19
+ | entity | verifier@0.3 | 0.1789 | 0.1693 | 2032 | 1972 | 0.44 | 0.46 | 0.070 | 1979 | – | – |
20
+ | entity | verifier@0.5 | 0.1794 | 0.1698 | 1890 | 1852 | 0.26 | 0.27 | 0.055 | 1143 | – | – |
21
+ | entity | verifier@0.7 | 0.1796 | 0.1699 | 1851 | 1824 | 0.15 | 0.15 | 0.037 | 627 | – | – |
22
+ | entity | verifier@0.8 | 0.1796 | 0.1700 | 1826 | 1806 | 0.09 | 0.10 | 0.023 | 357 | – | – |
23
+ | entity | verifier@0.9 | 0.1799 | 0.1702 | 1793 | 1776 | 0.06 | 0.06 | 0.012 | 139 | – | – |
24
+ | entity | oracle | 0.1684 | 0.1593 | 3896 | 3705 | 0.05 | 0.06 | 0.115 | 1804 | – | – |
25
+ | typo | whisper | 0.1412 | 0.1316 | 0 | 0 | 0.00 | 0.00 | 0.000 | 0 | – | – |
26
+ | typo | rules only | 0.1348 | 0.1248 | 341 | 335 | 0.01 | 0.02 | 0.011 | 0 | – | – |
27
+ | typo | accept all | 0.1343 | 0.1248 | 398 | 366 | 0.96 | 0.97 | 0.167 | 811 | – | – |
28
+ | typo | verifier@0.3 | 0.1344 | 0.1246 | 373 | 357 | 0.33 | 0.35 | 0.111 | 334 | – | – |
29
+ | typo | verifier@0.5 | 0.1347 | 0.1248 | 356 | 345 | 0.17 | 0.18 | 0.089 | 185 | – | – |
30
+ | typo | verifier@0.7 | 0.1347 | 0.1248 | 346 | 334 | 0.09 | 0.11 | 0.044 | 96 | – | – |
31
+ | typo | verifier@0.8 | 0.1347 | 0.1248 | 344 | 335 | 0.06 | 0.07 | 0.033 | 52 | – | – |
32
+ | typo | verifier@0.9 | 0.1347 | 0.1246 | 344 | 339 | 0.02 | 0.03 | 0.011 | 21 | – | – |
33
+ | typo | oracle | 0.1254 | 0.1157 | 703 | 665 | 0.01 | 0.02 | 0.167 | 313 | – | – |
34
+
35
+ Score bins (main set):
36
+
37
+ | bin | candidates | sampled | real error | not an error | undecided | precision |
38
+ |---|---|---|---|---|---|---|
39
+ | >= 0.0 | 4951 | 60 | 9 | 49 | 2 | 0.16 |
40
+ | >= 0.3 | 1601 | 60 | 23 | 25 | 12 | 0.48 |
41
+ | >= 0.5 | 1014 | 60 | 37 | 12 | 11 | 0.76 |
42
+ | >= 0.7 | 523 | 60 | 31 | 16 | 13 | 0.66 |
43
+ | >= 0.8 | 414 | 60 | 39 | 4 | 17 | 0.91 |
44
+ | >= 0.9 | 278 | 60 | 43 | 3 | 14 | 0.93 |