# corrupt-redmadrobot - ru / pii (2841 rows, 5531 spans, 371 negatives)

Main metric: **missed** - gold spans no predicted character touched. Score threshold 0.5; spans without a score always count. `hidden` - gold spans every character of which is covered: touching one character counts as detection, not as hiding. `hidden` is counted over normalized predictions - a prediction is stretched to whole words before it is compared, so it is what a masker that repeats the same normalization would hide; a masker that masks the raw offsets of the model hides no more than this. Char P / R / F1 are reference only. Trivial baseline, mask everything: 0 missed, 100% hidden, char P 0.115 (88.5% of the text over-masked). `rows touched` / `chars masked` count rows the source left without annotations where a mask touched something, and the characters masked in them; that absence of annotations is not evidence that those rows hold nothing sensitive, so this is not a false-alarm rate. The exact CSVs keep the technical column names `fp_rows` and `fp_chars`.

| model | missed | missed % [95% CI] | hidden | char F1 [95% CI] | P | R | entity exact | entity overlap | typed F1 | rows touched | chars masked | dropped spans | ms/row | params | train |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| pplx+homoglyph | **745** | 13.5% [11.8%, 15.2%] | 78.7% | 0.790 [0.769, 0.808] | 0.788 | 0.792 | 0.455 | 0.844 | 0.725 | 165/371 | 3562 | 0 | 56 | 596M |  |
| pplx | **748** | 13.5% [11.7%, 15.3%] | 78.6% | 0.787 [0.765, 0.804] | 0.787 | 0.787 | 0.455 | 0.842 | 0.722 | 170/371 | 3538 | 0 | 62 | 596M |  |
| nym-base+homoglyph | **902** | 16.3% [14.7%, 18.0%] | 72.0% | 0.773 [0.753, 0.791] | 0.794 | 0.754 | 0.550 | 0.823 | 0.341 | 161/371 | 4401 | 0 | 3 | 308M |  |
| nym-base | **939** | 17.0% [15.4%, 18.7%] | 71.3% | 0.768 [0.747, 0.786] | 0.794 | 0.745 | 0.545 | 0.818 | 0.334 | 165/371 | 4342 | 0 | 4 | 308M |  |
| mmbert32k | **1145** | 20.7% [18.5%, 23.1%] | 51.0% | 0.657 [0.635, 0.678] | 0.676 | 0.640 | 0.290 | 0.694 | 0.335 | 256/371 | 6395 | 0 | 4 | 308M |  |
| ru-pii-ner | **1189** | 21.5% [19.4%, 23.6%] | 72.9% | 0.701 [0.678, 0.726] | 0.693 | 0.710 | 0.446 | 0.753 | 0.235 | 214/371 | 13319 | 0 | 59 | 358M |  |
| nym-small | **1191** | 21.5% [19.8%, 23.4%] | 62.3% | 0.730 [0.709, 0.748] | 0.767 | 0.695 | 0.461 | 0.780 | 0.288 | 180/371 | 4989 | 0 | 159 | - |  |
| gliner2-fastino-ru | **1371** | 24.8% [23.1%, 26.4%] | 70.2% | 0.744 [0.721, 0.764] | 0.845 | 0.664 | 0.582 | 0.809 | 0.713 | 136/371 | 2815 | 3 | 17 | 307M |  |
| ru-legal-ner | **1422** | 25.7% [23.8%, 27.8%] | 58.0% | 0.501 [0.470, 0.532] | 0.413 | 0.639 | 0.232 | 0.580 | 0.341 | 311/371 | 39558 | 0 | 2 | 29M |  |
| ru-legal-ner+homoglyph | **1492** | 27.0% [25.1%, 29.1%] | 57.3% | 0.525 [0.495, 0.555] | 0.453 | 0.624 | 0.245 | 0.613 | 0.363 | 295/371 | 28664 | 0 | 2 | 29M |  |
| apararti | **1498** | 27.1% [24.9%, 29.3%] | 60.9% | 0.661 [0.642, 0.680] | 0.657 | 0.665 | 0.295 | 0.709 | 0.477 | 238/371 | 8183 | 0 | 62 | 1.4B |  |
| gliner-nvidia-ru | **1676** | 30.3% [28.6%, 31.9%] | 65.3% | 0.713 [0.690, 0.735] | 0.813 | 0.636 | 0.600 | 0.749 | 0.670 | 177/371 | 2823 | 0 | 32 | 445M |  |
| gliner2-fastino | **1692** | 30.6% [28.8%, 32.4%] | 62.2% | 0.721 [0.700, 0.741] | 0.871 | 0.615 | 0.601 | 0.771 | 0.650 | 117/371 | 2497 | 3 | 10 | 307M |  |
| pii-shield-onnx | **1778** | 32.1% [30.4%, 33.9%] | 54.4% | 0.390 [0.351, 0.432] | 0.272 | 0.686 | 0.274 | 0.526 | 0.214 | 311/371 | 99976 | 0 | 450 | - |  |
| opf-kz-ru | **1801** | 32.6% [30.4%, 34.8%] | 54.3% | 0.627 [0.608, 0.645] | 0.651 | 0.604 | 0.269 | 0.671 | 0.508 | 229/371 | 7343 | 0 | 59 | 1.4B |  |
| traciora | **1917** | 34.7% [32.4%, 37.0%] | 50.8% | 0.631 [0.604, 0.654] | 0.676 | 0.591 | 0.277 | 0.688 | 0.519 | 201/371 | 5775 | 0 | 52 | 1.4B |  |
| opf-ru | **1929** | 34.9% [32.9%, 37.0%] | 42.3% | 0.552 [0.531, 0.572] | 0.624 | 0.496 | 0.287 | 0.633 | 0.350 | 272/371 | 13446 | 0 | 40 | 1.4B |  |
| bardsai-eu | **1952** | 35.3% [33.1%, 37.5%] | 53.3% | 0.654 [0.633, 0.672] | 0.761 | 0.573 | 0.347 | 0.720 | 0.224 | 99/371 | 2602 | 0 | 218 | - |  |
| gliner2-hivetrace-omni-ru | **1962** | 35.5% [33.5%, 37.2%] | 57.5% | 0.681 [0.656, 0.706] | 0.862 | 0.563 | 0.571 | 0.739 | 0.661 | 132/371 | 1997 | 1 | 12 | 307M |  |
| gliner-nvidia+homoglyph | **1971** | 35.6% [33.7%, 37.5%] | 60.1% | 0.658 [0.634, 0.682] | 0.761 | 0.580 | 0.527 | 0.694 | 0.604 | 167/371 | 2831 | 0 | 9 | 445M |  |
| gliner-nvidia | **2033** | 36.8% [34.9%, 38.5%] | 59.0% | 0.649 [0.624, 0.674] | 0.753 | 0.569 | 0.514 | 0.685 | 0.594 | 182/371 | 3335 | 0 | 13 | 445M |  |
| gravitee-small | **2190** | 39.6% [37.5%, 41.7%] | 53.0% | 0.539 [0.524, 0.553] | 0.488 | 0.601 | 0.192 | 0.600 | 0.403 | 260/371 | 9562 | 0 | 2 | 29M |  |
| gliner2-hivetrace-omni | **2239** | 40.5% [38.8%, 42.2%] | 52.3% | 0.640 [0.615, 0.664] | 0.887 | 0.501 | 0.583 | 0.705 | 0.606 | 125/371 | 2017 | 1 | 10 | 307M |  |
| openai-base | **2248** | 40.6% [38.2%, 43.1%] | 51.8% | 0.604 [0.583, 0.624] | 0.683 | 0.541 | 0.259 | 0.664 | 0.485 | 152/371 | 4042 | 0 | 70 | 1.4B |  |
| opf-ru-v2+homoglyph | **2273** | 41.1% [38.8%, 43.3%] | 44.7% | 0.608 [0.585, 0.630] | 0.805 | 0.489 | 0.311 | 0.682 | 0.535 | 121/371 | 1934 | 0 | 8 | 1.4B |  |
| opf-ru-v2 | **2285** | 41.3% [38.9%, 43.3%] | 44.3% | 0.606 [0.583, 0.628] | 0.798 | 0.489 | 0.316 | 0.677 | 0.527 | 137/371 | 2391 | 0 | 248 | 1.4B |  |
| gliner-pii-edge | **2345** | 42.4% [40.3%, 44.5%] | 50.5% | 0.594 [0.573, 0.613] | 0.729 | 0.501 | 0.487 | 0.635 | 0.397 | 229/371 | 4587 | 0 | 132 | 45M |  |
| gliner-urchade-ru | **2428** | 43.9% [41.5%, 46.2%] | 54.2% | 0.674 [0.652, 0.696] | 0.845 | 0.561 | 0.519 | 0.682 | 0.656 | 69/371 | 1727 | 0 | 6 | 289M |  |
| gliner-urchade | **2436** | 44.0% [41.6%, 46.6%] | 53.4% | 0.670 [0.647, 0.692] | 0.824 | 0.565 | 0.501 | 0.672 | 0.634 | 112/371 | 2405 | 0 | 6 | 289M |  |
| gliner2-large | **2457** | 44.4% [42.3%, 46.5%] | 50.2% | 0.626 [0.602, 0.655] | 0.832 | 0.502 | 0.498 | 0.660 | 0.527 | 143/371 | 2366 | 15 | 23 | 486M |  |
| nuner-zero | **2464** | 44.5% [41.7%, 47.2%] | 46.0% | 0.659 [0.636, 0.686] | 0.768 | 0.577 | 0.350 | 0.639 | 0.573 | 83/371 | 1644 | 0 | 13 | 449M |  |
| gliner2-hivetrace-uni | **2520** | 45.6% [43.7%, 47.5%] | 45.8% | 0.536 [0.516, 0.554] | 0.730 | 0.423 | 0.467 | 0.623 | 0.355 | 193/371 | 4163 | 1 | 11 | 147M |  |
| gliner-stream-pii | **2526** | 45.7% [43.4%, 47.9%] | 32.5% | 0.498 [0.477, 0.517] | 0.782 | 0.365 | 0.337 | 0.614 | 0.390 | 253/371 | 2332 | 0 | 30 | 677M |  |
| openmed-multilingual | **2617** | 47.3% [45.5%, 49.1%] | 36.1% | 0.552 [0.529, 0.573] | 0.678 | 0.465 | 0.324 | 0.555 | 0.445 | 274/371 | 8288 | 0 | 31 | 1.4B |  |
| fef2-secret-ru | **2690** | 48.6% [46.3%, 51.2%] | 44.6% | 0.569 [0.546, 0.590] | 0.765 | 0.452 | 0.321 | 0.626 | 0.211 | 105/371 | 1581 | 0 | 3 | 177M |  |
| gliner25-fastino-ru | **2735** | 49.4% [47.0%, 51.6%] | 47.5% | 0.599 [0.569, 0.632] | 0.866 | 0.458 | 0.387 | 0.651 | 0.592 | 54/371 | 1135 | 0 | 5 | 287M |  |
| gliner2-vladlinv-ru | **2821** | 51.0% [48.6%, 53.4%] | 47.9% | 0.587 [0.559, 0.613] | 0.919 | 0.431 | 0.475 | 0.646 | 0.581 | 18/371 | 309 | 0 | 5 | 287M |  |
| stanza-ru | **2825** | 51.1% [48.6%, 53.7%] | 40.1% | 0.277 [0.256, 0.300] | 0.227 | 0.354 | 0.130 | 0.388 | 0.060 | 337/371 | 31957 | 0 | 30 | - |  |
| openmed-nemotron | **2846** | 51.5% [49.5%, 53.3%] | 33.0% | 0.411 [0.390, 0.432] | 0.397 | 0.426 | 0.226 | 0.447 | 0.281 | 272/371 | 14223 | 0 | 31 | 1.4B |  |
| kalyan-ettin | **3117** | 56.4% [54.2%, 58.5%] | 30.3% | 0.495 [0.470, 0.519] | 0.755 | 0.368 | 0.313 | 0.542 | 0.351 | 208/371 | 3724 | 0 | 6 | 68M |  |
| gliner2-vladlinv+homoglyph | **3134** | 56.7% [54.3%, 59.0%] | 42.4% | 0.523 [0.495, 0.548] | 0.922 | 0.365 | 0.413 | 0.598 | 0.513 | 13/371 | 228 | 0 | 4 | 287M |  |
| gliner2-vladlinv | **3193** | 57.7% [55.3%, 60.0%] | 41.3% | 0.515 [0.487, 0.541] | 0.923 | 0.357 | 0.410 | 0.588 | 0.505 | 13/371 | 217 | 0 | 7 | 287M |  |
| ner-ru-yqelz | **3193** | 57.7% [56.0%, 59.7%] | 32.5% | 0.329 [0.307, 0.351] | 0.344 | 0.315 | 0.145 | 0.398 | 0.091 | 275/371 | 18483 | 0 | 6 | 559M |  |
| ner-ru-gherman+homoglyph | **3300** | 59.7% [56.8%, 62.5%] | 31.2% | 0.425 [0.393, 0.456] | 0.917 | 0.276 | 0.413 | 0.559 | 0.094 | 52/371 | 637 | 0 | 4 | 177M |  |
| ner-ru-gherman | **3339** | 60.4% [57.6%, 63.2%] | 30.6% | 0.418 [0.388, 0.450] | 0.916 | 0.271 | 0.407 | 0.552 | 0.095 | 51/371 | 621 | 0 | 4 | 177M |  |
| davlan-xlmr | **3417** | 61.8% [58.9%, 64.6%] | 33.4% | 0.399 [0.370, 0.430] | 0.717 | 0.276 | 0.205 | 0.515 | 0.168 | 89/371 | 1385 | 0 | 3 | 277M |  |
| gliner25-fastino | **3506** | 63.4% [61.1%, 65.5%] | 32.9% | 0.467 [0.433, 0.506] | 0.798 | 0.330 | 0.373 | 0.487 | 0.444 | 117/371 | 3397 | 0 | 5 | 287M |  |
| davlan-mbert | **3530** | 63.8% [61.0%, 66.6%] | 32.6% | 0.386 [0.356, 0.416] | 0.684 | 0.269 | 0.220 | 0.476 | 0.161 | 102/371 | 1648 | 0 | 4 | 177M |  |
| gliner-pii-base | **3531** | 63.8% [61.7%, 66.0%] | 32.7% | 0.471 [0.439, 0.504] | 0.854 | 0.325 | 0.425 | 0.498 | 0.392 | 98/371 | 1360 | 0 | 4 | 166M |  |
| gliner-multi-v21-ru | **3782** | 68.4% [66.1%, 70.6%] | 27.9% | 0.355 [0.332, 0.380] | 0.715 | 0.236 | 0.203 | 0.425 | 0.339 | 62/371 | 1400 | 0 | 8 | 289M |  |
| gliner-multi-v21 | **3925** | 71.0% [68.7%, 73.1%] | 25.0% | 0.326 [0.301, 0.351] | 0.718 | 0.211 | 0.192 | 0.409 | 0.311 | 58/371 | 1702 | 0 | 6 | 289M |  |
| gliner2-hivetrace-uni-ru | **4015** | 72.6% [71.1%, 74.2%] | 24.5% | 0.300 [0.281, 0.319] | 0.796 | 0.185 | 0.259 | 0.411 | 0.187 | 88/371 | 1198 | 0 | 12 | 147M |  |
| spacy-ru-lg | **4130** | 74.7% [72.2%, 76.8%] | 21.7% | 0.230 [0.210, 0.250] | 0.300 | 0.186 | 0.138 | 0.312 | 0.057 | 210/371 | 14004 | 0 | 4 | - |  |
| natasha | **4322** | 78.1% [76.1%, 80.2%] | 18.8% | 0.227 [0.207, 0.249] | 0.439 | 0.153 | 0.141 | 0.314 | 0.065 | 107/371 | 2979 | 0 | 2 | - |  |
| rules-ru | **4870** | 88.0% [86.6%, 89.4%] | 9.7% | 0.259 [0.239, 0.278] | 0.862 | 0.152 | 0.150 | 0.210 | 0.240 | 21/371 | 402 | 0 | 0 | - |  |
| spacy-alrosait | **4912** | 88.8% [87.1%, 90.3%] | 10.1% | 0.140 [0.122, 0.161] | 0.689 | 0.078 | 0.044 | 0.198 | 0.074 | 26/371 | 396 | 0 | 4 | - |  |

Neighbours by rank a paired bootstrap cannot tell apart (95%, missed %) - pairs, not a transitive chain, so `a ≈ b` and `b ≈ c` do not make `a ≈ c`: pplx+homoglyph ≈ pplx; mmbert32k ≈ ru-pii-ner; ru-pii-ner ≈ nym-small; gliner2-fastino-ru ≈ ru-legal-ner; ru-legal-ner+homoglyph ≈ apararti; gliner-nvidia-ru ≈ gliner2-fastino; gliner2-fastino ≈ pii-shield-onnx; pii-shield-onnx ≈ opf-kz-ru; traciora ≈ opf-ru; opf-ru ≈ bardsai-eu; bardsai-eu ≈ gliner2-hivetrace-omni-ru; gliner2-hivetrace-omni-ru ≈ gliner-nvidia+homoglyph; gravitee-small ≈ gliner2-hivetrace-omni; gliner2-hivetrace-omni ≈ openai-base; openai-base ≈ opf-ru-v2+homoglyph; opf-ru-v2+homoglyph ≈ opf-ru-v2; opf-ru-v2 ≈ gliner-pii-edge; gliner-pii-edge ≈ gliner-urchade-ru; gliner-urchade-ru ≈ gliner-urchade; gliner-urchade ≈ gliner2-large; gliner2-large ≈ nuner-zero; nuner-zero ≈ gliner2-hivetrace-uni; gliner2-hivetrace-uni ≈ gliner-stream-pii; gliner-stream-pii ≈ openmed-multilingual; openmed-multilingual ≈ fef2-secret-ru; fef2-secret-ru ≈ gliner25-fastino-ru; gliner25-fastino-ru ≈ gliner2-vladlinv-ru; gliner2-vladlinv-ru ≈ stanza-ru; stanza-ru ≈ openmed-nemotron; kalyan-ettin ≈ gliner2-vladlinv+homoglyph; gliner2-vladlinv ≈ ner-ru-yqelz; ner-ru-yqelz ≈ ner-ru-gherman+homoglyph; davlan-xlmr ≈ gliner25-fastino; gliner25-fastino ≈ davlan-mbert; davlan-mbert ≈ gliner-pii-base; gliner-multi-v21 ≈ gliner2-hivetrace-uni-ru; gliner2-hivetrace-uni-ru ≈ spacy-ru-lg; rules-ru ≈ spacy-alrosait

## Missed by group

| model | PERSON | ADDRESS | CONTACT | ID | NET |
|---|---|---|---|---|---|
| spans in gold | 1260 | 1242 | 389 | 2279 | 361 |
| pplx+homoglyph | 120 (9.5%) | 441 (35.5%) | 5 (1.3%) | 62 (2.7%) | 117 (32.4%) |
| pplx | 115 (9.1%) | 442 (35.6%) | 5 (1.3%) | 67 (2.9%) | 119 (33.0%) |
| nym-base+homoglyph | 120 (9.5%) | 280 (22.5%) | 18 (4.6%) | 437 (19.2%) | 47 (13.0%) |
| nym-base | 130 (10.3%) | 305 (24.6%) | 17 (4.4%) | 438 (19.2%) | 49 (13.6%) |
| mmbert32k | 304 (24.1%) | 289 (23.3%) | 61 (15.7%) | 458 (20.1%) | 33 (9.1%) |
| ru-pii-ner | 99 (7.9%) | 582 (46.9%) | 76 (19.5%) | 250 (11.0%) | 182 (50.4%) |
| nym-small | 214 (17.0%) | 328 (26.4%) | 11 (2.8%) | 560 (24.6%) | 78 (21.6%) |
| gliner2-fastino-ru | 289 (22.9%) | 260 (20.9%) | 38 (9.8%) | 586 (25.7%) | 198 (54.8%) |
| ru-legal-ner | 323 (25.6%) | 486 (39.1%) | 21 (5.4%) | 469 (20.6%) | 123 (34.1%) |
| ru-legal-ner+homoglyph | 326 (25.9%) | 539 (43.4%) | 20 (5.1%) | 481 (21.1%) | 126 (34.9%) |
| apararti | 362 (28.7%) | 584 (47.0%) | 49 (12.6%) | 388 (17.0%) | 115 (31.9%) |
| gliner-nvidia-ru | 307 (24.4%) | 360 (29.0%) | 51 (13.1%) | 827 (36.3%) | 131 (36.3%) |
| gliner2-fastino | 366 (29.0%) | 268 (21.6%) | 32 (8.2%) | 861 (37.8%) | 165 (45.7%) |
| pii-shield-onnx | 321 (25.5%) | 391 (31.5%) | 15 (3.9%) | 991 (43.5%) | 60 (16.6%) |
| opf-kz-ru | 517 (41.0%) | 791 (63.7%) | 37 (9.5%) | 338 (14.8%) | 118 (32.7%) |
| traciora | 351 (27.9%) | 511 (41.1%) | 31 (8.0%) | 845 (37.1%) | 179 (49.6%) |
| opf-ru | 385 (30.6%) | 587 (47.3%) | 36 (9.3%) | 769 (33.7%) | 152 (42.1%) |
| bardsai-eu | 292 (23.2%) | 422 (34.0%) | 51 (13.1%) | 1026 (45.0%) | 161 (44.6%) |
| gliner2-hivetrace-omni-ru | 499 (39.6%) | 304 (24.5%) | 49 (12.6%) | 880 (38.6%) | 230 (63.7%) |
| gliner-nvidia+homoglyph | 211 (16.7%) | 329 (26.5%) | 46 (11.8%) | 1247 (54.7%) | 138 (38.2%) |
| gliner-nvidia | 243 (19.3%) | 360 (29.0%) | 44 (11.3%) | 1247 (54.7%) | 139 (38.5%) |
| gravitee-small | 451 (35.8%) | 489 (39.4%) | 51 (13.1%) | 1171 (51.4%) | 28 (7.8%) |
| gliner2-hivetrace-omni | 404 (32.1%) | 327 (26.3%) | 65 (16.7%) | 1214 (53.3%) | 229 (63.4%) |
| openai-base | 446 (35.4%) | 759 (61.1%) | 73 (18.8%) | 769 (33.7%) | 201 (55.7%) |
| opf-ru-v2+homoglyph | 509 (40.4%) | 783 (63.0%) | 89 (22.9%) | 691 (30.3%) | 201 (55.7%) |
| opf-ru-v2 | 536 (42.5%) | 769 (61.9%) | 88 (22.6%) | 691 (30.3%) | 201 (55.7%) |
| gliner-pii-edge | 676 (53.7%) | 725 (58.4%) | 108 (27.8%) | 716 (31.4%) | 120 (33.2%) |
| gliner-urchade-ru | 1115 (88.5%) | 378 (30.4%) | 78 (20.1%) | 675 (29.6%) | 182 (50.4%) |
| gliner-urchade | 1013 (80.4%) | 379 (30.5%) | 43 (11.1%) | 865 (38.0%) | 136 (37.7%) |
| gliner2-large | 630 (50.0%) | 529 (42.6%) | 50 (12.9%) | 1094 (48.0%) | 154 (42.7%) |
| nuner-zero | 810 (64.3%) | 495 (39.9%) | 15 (3.9%) | 1142 (50.1%) | 2 (0.6%) |
| gliner2-hivetrace-uni | 771 (61.2%) | 410 (33.0%) | 87 (22.4%) | 927 (40.7%) | 325 (90.0%) |
| gliner-stream-pii | 614 (48.7%) | 575 (46.3%) | 146 (37.5%) | 1039 (45.6%) | 152 (42.1%) |
| openmed-multilingual | 667 (52.9%) | 770 (62.0%) | 77 (19.8%) | 1053 (46.2%) | 50 (13.9%) |
| fef2-secret-ru | 367 (29.1%) | 503 (40.5%) | 193 (49.6%) | 1294 (56.8%) | 333 (92.2%) |
| gliner25-fastino-ru | 515 (40.9%) | 1006 (81.0%) | 80 (20.6%) | 867 (38.0%) | 267 (74.0%) |
| gliner2-vladlinv-ru | 663 (52.6%) | 1085 (87.4%) | 131 (33.7%) | 599 (26.3%) | 343 (95.0%) |
| stanza-ru | 156 (12.4%) | 259 (20.9%) | 350 (90.0%) | 1815 (79.6%) | 245 (67.9%) |
| openmed-nemotron | 423 (33.6%) | 571 (46.0%) | 106 (27.2%) | 1678 (73.6%) | 68 (18.8%) |
| kalyan-ettin | 564 (44.8%) | 763 (61.4%) | 139 (35.7%) | 1576 (69.2%) | 75 (20.8%) |
| gliner2-vladlinv+homoglyph | 582 (46.2%) | 1079 (86.9%) | 129 (33.2%) | 1018 (44.7%) | 326 (90.3%) |
| gliner2-vladlinv | 627 (49.8%) | 1100 (88.6%) | 129 (33.2%) | 1011 (44.4%) | 326 (90.3%) |
| ner-ru-yqelz | 378 (30.0%) | 408 (32.9%) | 279 (71.7%) | 1844 (80.9%) | 284 (78.7%) |
| ner-ru-gherman+homoglyph | 162 (12.9%) | 208 (16.7%) | 306 (78.7%) | 2266 (99.4%) | 358 (99.2%) |
| ner-ru-gherman | 176 (14.0%) | 232 (18.7%) | 307 (78.9%) | 2266 (99.4%) | 358 (99.2%) |
| davlan-xlmr | 150 (11.9%) | 269 (21.7%) | 375 (96.4%) | 2272 (99.7%) | 351 (97.2%) |
| gliner25-fastino | 589 (46.7%) | 917 (73.8%) | 72 (18.5%) | 1683 (73.8%) | 245 (67.9%) |
| davlan-mbert | 210 (16.7%) | 351 (28.3%) | 373 (95.9%) | 2276 (99.9%) | 320 (88.6%) |
| gliner-pii-base | 675 (53.6%) | 859 (69.2%) | 104 (26.7%) | 1738 (76.3%) | 155 (42.9%) |
| gliner-multi-v21-ru | 562 (44.6%) | 568 (45.7%) | 302 (77.6%) | 2098 (92.1%) | 252 (69.8%) |
| gliner-multi-v21 | 602 (47.8%) | 567 (45.7%) | 307 (78.9%) | 2225 (97.6%) | 224 (62.0%) |
| gliner2-hivetrace-uni-ru | 773 (61.3%) | 813 (65.5%) | 342 (87.9%) | 1735 (76.1%) | 352 (97.5%) |
| spacy-ru-lg | 607 (48.2%) | 657 (52.9%) | 359 (92.3%) | 2185 (95.9%) | 322 (89.2%) |
| natasha | 713 (56.6%) | 685 (55.2%) | 372 (95.6%) | 2200 (96.5%) | 352 (97.5%) |
| rules-ru | 1257 (99.8%) | 1239 (99.8%) | 132 (33.9%) | 2068 (90.7%) | 174 (48.2%) |
| spacy-alrosait | 975 (77.4%) | 927 (74.6%) | 387 (99.5%) | 2263 (99.3%) | 360 (99.7%) |

## Char recall by gold type

| type | group | pplx+homoglyph | pplx | nym-base+homoglyph | nym-base | mmbert32k | ru-pii-ner | nym-small | gliner2-fastino-ru | ru-legal-ner | ru-legal-ner+homoglyph | apararti | gliner-nvidia-ru | gliner2-fastino | pii-shield-onnx | opf-kz-ru | traciora | opf-ru | bardsai-eu | gliner2-hivetrace-omni-ru | gliner-nvidia+homoglyph | gliner-nvidia | gravitee-small | gliner2-hivetrace-omni | openai-base | opf-ru-v2+homoglyph | opf-ru-v2 | gliner-pii-edge | gliner-urchade-ru | gliner-urchade | gliner2-large | nuner-zero | gliner2-hivetrace-uni | gliner-stream-pii | openmed-multilingual | fef2-secret-ru | gliner25-fastino-ru | gliner2-vladlinv-ru | stanza-ru | openmed-nemotron | kalyan-ettin | gliner2-vladlinv+homoglyph | gliner2-vladlinv | ner-ru-yqelz | ner-ru-gherman+homoglyph | ner-ru-gherman | davlan-xlmr | gliner25-fastino | davlan-mbert | gliner-pii-base | gliner-multi-v21-ru | gliner-multi-v21 | gliner2-hivetrace-uni-ru | spacy-ru-lg | natasha | rules-ru | spacy-alrosait |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| BIRTH_CERTIFICATE | ID | 0.700 | 0.700 | 0.521 | 0.516 | 0.356 | 0.882 | 0.413 | 0.492 | 0.470 | 0.470 | 0.664 | 0.427 | 0.394 | 0.404 | 0.628 | 0.363 | 0.372 | 0.265 | 0.423 | 0.250 | 0.258 | 0.122 | 0.378 | 0.472 | 0.438 | 0.434 | 0.656 | 0.661 | 0.548 | 0.422 | 0.332 | 0.474 | 0.194 | 0.219 | 0.671 | 0.382 | 0.633 | 0.226 | 0.116 | 0.199 | 0.525 | 0.527 | 0.160 | 0.000 | 0.000 | 0.002 | 0.091 | 0.001 | 0.197 | 0.000 | 0.027 | 0.270 | 0.035 | 0.040 | 0.000 | 0.000 |
| CITY | ADDRESS | 0.639 | 0.651 | 0.797 | 0.782 | 0.720 | 0.498 | 0.766 | 0.820 | 0.601 | 0.575 | 0.496 | 0.702 | 0.788 | 0.648 | 0.332 | 0.515 | 0.476 | 0.741 | 0.695 | 0.772 | 0.763 | 0.513 | 0.698 | 0.372 | 0.329 | 0.322 | 0.385 | 0.699 | 0.658 | 0.510 | 0.462 | 0.725 | 0.443 | 0.354 | 0.717 | 0.055 | 0.075 | 0.873 | 0.485 | 0.373 | 0.067 | 0.059 | 0.791 | 0.811 | 0.793 | 0.866 | 0.144 | 0.856 | 0.354 | 0.607 | 0.558 | 0.487 | 0.477 | 0.501 | 0.000 | 0.228 |
| COUNTRY | ADDRESS | 0.399 | 0.397 | 0.726 | 0.708 | 0.573 | 0.318 | 0.661 | 0.747 | 0.376 | 0.350 | 0.320 | 0.667 | 0.727 | 0.618 | 0.196 | 0.390 | 0.274 | 0.537 | 0.643 | 0.679 | 0.647 | 0.450 | 0.619 | 0.177 | 0.219 | 0.228 | 0.457 | 0.583 | 0.627 | 0.422 | 0.477 | 0.674 | 0.480 | 0.309 | 0.647 | 0.224 | 0.084 | 0.871 | 0.376 | 0.339 | 0.093 | 0.086 | 0.854 | 0.769 | 0.739 | 0.754 | 0.325 | 0.798 | 0.344 | 0.574 | 0.570 | 0.431 | 0.567 | 0.603 | 0.006 | 0.126 |
| CREDIT_CARD | ID | 0.895 | 0.896 | 0.851 | 0.851 | 0.777 | 0.767 | 0.797 | 0.869 | 0.746 | 0.740 | 0.853 | 0.835 | 0.886 | 0.707 | 0.829 | 0.809 | 0.489 | 0.632 | 0.865 | 0.704 | 0.704 | 0.831 | 0.871 | 0.738 | 0.648 | 0.650 | 0.805 | 0.727 | 0.831 | 0.699 | 0.675 | 0.255 | 0.348 | 0.713 | 0.533 | 0.770 | 0.683 | 0.042 | 0.508 | 0.310 | 0.371 | 0.371 | 0.200 | 0.000 | 0.000 | 0.000 | 0.486 | 0.004 | 0.425 | 0.045 | 0.040 | 0.060 | 0.001 | 0.004 | 0.124 | 0.000 |
| DISTRICT | ADDRESS | 0.514 | 0.491 | 0.629 | 0.575 | 0.585 | 0.676 | 0.522 | 0.713 | 0.550 | 0.452 | 0.436 | 0.565 | 0.718 | 0.634 | 0.240 | 0.621 | 0.390 | 0.468 | 0.610 | 0.588 | 0.509 | 0.503 | 0.563 | 0.343 | 0.256 | 0.291 | 0.366 | 0.746 | 0.769 | 0.441 | 0.588 | 0.554 | 0.380 | 0.228 | 0.669 | 0.300 | 0.171 | 0.782 | 0.429 | 0.289 | 0.177 | 0.128 | 0.679 | 0.454 | 0.441 | 0.670 | 0.172 | 0.652 | 0.135 | 0.477 | 0.401 | 0.171 | 0.510 | 0.524 | 0.000 | 0.115 |
| DRIVER_LICENSE | ID | 0.880 | 0.844 | 0.679 | 0.679 | 0.554 | 0.870 | 0.573 | 0.597 | 0.668 | 0.645 | 0.644 | 0.491 | 0.505 | 0.532 | 0.647 | 0.393 | 0.456 | 0.356 | 0.333 | 0.373 | 0.373 | 0.454 | 0.364 | 0.420 | 0.458 | 0.457 | 0.516 | 0.507 | 0.525 | 0.384 | 0.234 | 0.482 | 0.226 | 0.303 | 0.246 | 0.465 | 0.629 | 0.073 | 0.139 | 0.209 | 0.502 | 0.518 | 0.104 | 0.007 | 0.007 | 0.000 | 0.241 | 0.000 | 0.172 | 0.005 | 0.004 | 0.231 | 0.018 | 0.000 | 0.011 | 0.005 |
| EMAIL | CONTACT | 0.929 | 0.929 | 0.800 | 0.804 | 0.575 | 0.641 | 0.850 | 0.825 | 0.792 | 0.792 | 0.722 | 0.906 | 0.820 | 0.901 | 0.763 | 0.885 | 0.659 | 0.567 | 0.850 | 0.908 | 0.904 | 0.705 | 0.809 | 0.724 | 0.587 | 0.588 | 0.526 | 0.795 | 0.868 | 0.834 | 0.981 | 0.552 | 0.472 | 0.560 | 0.670 | 0.807 | 0.498 | 0.120 | 0.563 | 0.564 | 0.499 | 0.498 | 0.129 | 0.167 | 0.165 | 0.023 | 0.805 | 0.030 | 0.741 | 0.143 | 0.141 | 0.087 | 0.115 | 0.030 | 0.470 | 0.003 |
| FIRST_NAME | PERSON | 0.909 | 0.913 | 0.901 | 0.895 | 0.753 | 0.913 | 0.804 | 0.703 | 0.687 | 0.682 | 0.710 | 0.685 | 0.646 | 0.726 | 0.587 | 0.739 | 0.685 | 0.757 | 0.531 | 0.739 | 0.721 | 0.619 | 0.591 | 0.649 | 0.603 | 0.574 | 0.472 | 0.074 | 0.134 | 0.394 | 0.336 | 0.475 | 0.456 | 0.477 | 0.626 | 0.564 | 0.433 | 0.849 | 0.655 | 0.621 | 0.472 | 0.446 | 0.628 | 0.851 | 0.850 | 0.850 | 0.435 | 0.823 | 0.406 | 0.538 | 0.511 | 0.340 | 0.448 | 0.367 | 0.000 | 0.192 |
| HOUSE | ADDRESS | 0.924 | 0.924 | 0.447 | 0.445 | 0.634 | 0.514 | 0.386 | 0.487 | 0.635 | 0.603 | 0.558 | 0.619 | 0.501 | 0.493 | 0.307 | 0.589 | 0.224 | 0.580 | 0.478 | 0.580 | 0.596 | 0.871 | 0.421 | 0.513 | 0.420 | 0.410 | 0.070 | 0.615 | 0.492 | 0.603 | 0.765 | 0.118 | 0.145 | 0.106 | 0.051 | 0.100 | 0.185 | 0.097 | 0.245 | 0.225 | 0.164 | 0.169 | 0.053 | 0.900 | 0.900 | 0.106 | 0.094 | 0.012 | 0.089 | 0.311 | 0.444 | 0.038 | 0.033 | 0.025 | 0.000 | 0.368 |
| INN | ID | 0.990 | 0.976 | 0.809 | 0.809 | 0.808 | 0.819 | 0.716 | 0.786 | 0.754 | 0.739 | 0.779 | 0.727 | 0.633 | 0.727 | 0.794 | 0.721 | 0.678 | 0.750 | 0.747 | 0.472 | 0.475 | 0.253 | 0.426 | 0.657 | 0.695 | 0.698 | 0.438 | 0.796 | 0.671 | 0.319 | 0.409 | 0.592 | 0.343 | 0.647 | 0.457 | 0.352 | 0.453 | 0.181 | 0.199 | 0.163 | 0.277 | 0.286 | 0.069 | 0.000 | 0.000 | 0.000 | 0.037 | 0.000 | 0.092 | 0.429 | 0.064 | 0.119 | 0.029 | 0.012 | 0.469 | 0.000 |
| IP_ADDRESS | NET | 0.999 | 0.999 | 0.861 | 0.864 | 0.691 | 0.599 | 0.726 | 0.332 | 0.395 | 0.396 | 0.915 | 0.429 | 0.341 | 0.797 | 0.874 | 0.713 | 0.530 | 0.835 | 0.310 | 0.418 | 0.418 | 0.982 | 0.311 | 0.756 | 0.759 | 0.752 | 0.380 | 0.308 | 0.330 | 0.294 | 0.991 | 0.028 | 0.452 | 0.937 | 0.014 | 0.424 | 0.048 | 0.120 | 0.893 | 0.920 | 0.127 | 0.127 | 0.115 | 0.000 | 0.000 | 0.000 | 0.472 | 0.000 | 0.410 | 0.290 | 0.273 | 0.014 | 0.005 | 0.000 | 0.414 | 0.000 |
| LAST_NAME | PERSON | 0.841 | 0.848 | 0.844 | 0.839 | 0.666 | 0.901 | 0.795 | 0.677 | 0.752 | 0.744 | 0.676 | 0.672 | 0.611 | 0.765 | 0.551 | 0.649 | 0.653 | 0.715 | 0.564 | 0.749 | 0.706 | 0.562 | 0.581 | 0.604 | 0.527 | 0.502 | 0.420 | 0.104 | 0.223 | 0.479 | 0.234 | 0.273 | 0.429 | 0.445 | 0.621 | 0.472 | 0.379 | 0.817 | 0.599 | 0.543 | 0.471 | 0.441 | 0.623 | 0.812 | 0.805 | 0.842 | 0.461 | 0.815 | 0.319 | 0.440 | 0.455 | 0.341 | 0.430 | 0.396 | 0.005 | 0.137 |
| MIDDLE_NAME | PERSON | 0.974 | 0.975 | 0.939 | 0.941 | 0.796 | 0.984 | 0.885 | 0.819 | 0.816 | 0.826 | 0.818 | 0.773 | 0.737 | 0.795 | 0.651 | 0.870 | 0.778 | 0.801 | 0.581 | 0.854 | 0.842 | 0.768 | 0.585 | 0.719 | 0.695 | 0.675 | 0.447 | 0.168 | 0.213 | 0.543 | 0.431 | 0.233 | 0.505 | 0.451 | 0.820 | 0.634 | 0.512 | 0.900 | 0.732 | 0.495 | 0.529 | 0.482 | 0.734 | 0.946 | 0.937 | 0.914 | 0.594 | 0.901 | 0.567 | 0.565 | 0.493 | 0.379 | 0.564 | 0.490 | 0.000 | 0.314 |
| MILITARY_ID | ID | 0.966 | 0.966 | 0.473 | 0.473 | 0.572 | 0.919 | 0.549 | 0.770 | 0.762 | 0.762 | 0.823 | 0.697 | 0.635 | 0.339 | 0.859 | 0.673 | 0.568 | 0.510 | 0.692 | 0.543 | 0.543 | 0.293 | 0.475 | 0.670 | 0.709 | 0.699 | 0.711 | 0.762 | 0.670 | 0.564 | 0.590 | 0.717 | 0.319 | 0.327 | 0.594 | 0.667 | 0.874 | 0.424 | 0.101 | 0.266 | 0.771 | 0.775 | 0.124 | 0.001 | 0.001 | 0.003 | 0.167 | 0.000 | 0.453 | 0.016 | 0.006 | 0.399 | 0.112 | 0.063 | 0.000 | 0.014 |
| OMS | ID | 0.873 | 0.873 | 0.609 | 0.609 | 0.750 | 0.888 | 0.631 | 0.713 | 0.726 | 0.712 | 0.759 | 0.668 | 0.556 | 0.743 | 0.738 | 0.663 | 0.517 | 0.674 | 0.492 | 0.379 | 0.379 | 0.411 | 0.200 | 0.610 | 0.645 | 0.645 | 0.647 | 0.714 | 0.538 | 0.368 | 0.331 | 0.401 | 0.225 | 0.520 | 0.498 | 0.632 | 0.752 | 0.027 | 0.234 | 0.288 | 0.497 | 0.497 | 0.082 | 0.000 | 0.000 | 0.000 | 0.154 | 0.000 | 0.114 | 0.027 | 0.000 | 0.082 | 0.025 | 0.004 | 0.053 | 0.000 |
| PASSPORT | ID | 0.956 | 0.943 | 0.731 | 0.732 | 0.640 | 0.904 | 0.636 | 0.691 | 0.803 | 0.787 | 0.732 | 0.568 | 0.568 | 0.418 | 0.753 | 0.529 | 0.542 | 0.445 | 0.539 | 0.397 | 0.395 | 0.532 | 0.510 | 0.621 | 0.522 | 0.523 | 0.624 | 0.617 | 0.608 | 0.519 | 0.550 | 0.552 | 0.364 | 0.315 | 0.197 | 0.713 | 0.746 | 0.046 | 0.194 | 0.241 | 0.683 | 0.693 | 0.108 | 0.004 | 0.004 | 0.001 | 0.533 | 0.000 | 0.189 | 0.021 | 0.017 | 0.260 | 0.030 | 0.000 | 0.042 | 0.008 |
| PHONE | CONTACT | 0.968 | 0.968 | 0.960 | 0.969 | 0.826 | 0.867 | 0.955 | 0.810 | 0.969 | 0.971 | 0.900 | 0.797 | 0.811 | 0.945 | 0.859 | 0.709 | 0.766 | 0.796 | 0.769 | 0.824 | 0.833 | 0.942 | 0.755 | 0.757 | 0.731 | 0.719 | 0.890 | 0.781 | 0.883 | 0.785 | 0.865 | 0.812 | 0.578 | 0.614 | 0.083 | 0.657 | 0.829 | 0.002 | 0.510 | 0.366 | 0.847 | 0.847 | 0.114 | 0.004 | 0.003 | 0.000 | 0.690 | 0.005 | 0.682 | 0.125 | 0.112 | 0.065 | 0.000 | 0.000 | 0.552 | 0.005 |
| REGION | ADDRESS | 0.515 | 0.501 | 0.635 | 0.571 | 0.536 | 0.612 | 0.613 | 0.743 | 0.385 | 0.306 | 0.421 | 0.420 | 0.697 | 0.678 | 0.275 | 0.500 | 0.376 | 0.515 | 0.555 | 0.545 | 0.506 | 0.473 | 0.523 | 0.292 | 0.215 | 0.239 | 0.374 | 0.609 | 0.668 | 0.445 | 0.504 | 0.620 | 0.395 | 0.260 | 0.548 | 0.128 | 0.171 | 0.740 | 0.442 | 0.327 | 0.214 | 0.162 | 0.806 | 0.557 | 0.526 | 0.714 | 0.344 | 0.625 | 0.216 | 0.255 | 0.293 | 0.373 | 0.403 | 0.455 | 0.000 | 0.110 |
| SNILS | ID | 0.780 | 0.776 | 0.729 | 0.729 | 0.628 | 0.812 | 0.603 | 0.719 | 0.669 | 0.661 | 0.668 | 0.554 | 0.560 | 0.743 | 0.607 | 0.396 | 0.278 | 0.576 | 0.588 | 0.372 | 0.372 | 0.517 | 0.185 | 0.483 | 0.415 | 0.431 | 0.333 | 0.674 | 0.429 | 0.534 | 0.329 | 0.372 | 0.267 | 0.336 | 0.539 | 0.653 | 0.716 | 0.052 | 0.120 | 0.145 | 0.320 | 0.320 | 0.185 | 0.001 | 0.001 | 0.000 | 0.084 | 0.001 | 0.102 | 0.044 | 0.038 | 0.109 | 0.006 | 0.000 | 0.000 | 0.015 |
| STREET | ADDRESS | 0.875 | 0.878 | 0.872 | 0.868 | 0.671 | 0.612 | 0.742 | 0.691 | 0.692 | 0.700 | 0.707 | 0.617 | 0.713 | 0.677 | 0.486 | 0.793 | 0.694 | 0.759 | 0.681 | 0.638 | 0.603 | 0.854 | 0.721 | 0.625 | 0.605 | 0.616 | 0.255 | 0.742 | 0.731 | 0.583 | 0.704 | 0.511 | 0.402 | 0.389 | 0.510 | 0.244 | 0.151 | 0.743 | 0.607 | 0.284 | 0.162 | 0.155 | 0.332 | 0.621 | 0.612 | 0.863 | 0.309 | 0.794 | 0.209 | 0.589 | 0.584 | 0.030 | 0.316 | 0.168 | 0.000 | 0.402 |
| URL | NET | 0.333 | 0.314 | 0.686 | 0.660 | 0.421 | 0.188 | 0.653 | 0.226 | 0.282 | 0.271 | 0.374 | 0.464 | 0.294 | 0.685 | 0.380 | 0.131 | 0.131 | 0.116 | 0.162 | 0.438 | 0.435 | 0.715 | 0.152 | 0.185 | 0.034 | 0.042 | 0.486 | 0.365 | 0.495 | 0.396 | 0.927 | 0.037 | 0.260 | 0.555 | 0.041 | 0.057 | 0.017 | 0.224 | 0.493 | 0.364 | 0.020 | 0.020 | 0.169 | 0.003 | 0.003 | 0.016 | 0.066 | 0.052 | 0.339 | 0.018 | 0.050 | 0.009 | 0.122 | 0.012 | 0.384 | 0.004 |

## Missed at thresholds 0.5 / 0.3 / 0.2 / 0.1

Only models that return a score. A lower threshold keeps more spans: fewer misses, more masking.

| model | 0.5 | 0.3 | 0.2 | 0.1 |
|---|---|---|---|---|
| nym-base+homoglyph | 902 (16.3%) | 682 (12.3%) | 673 (12.2%) | 673 (12.2%) |
| nym-base | 939 (17.0%) | 707 (12.8%) | 693 (12.5%) | 693 (12.5%) |
| mmbert32k | 1145 (20.7%) | 592 (10.7%) | 535 (9.7%) | 527 (9.5%) |
| nym-small | 1191 (21.5%) | 962 (17.4%) | 954 (17.2%) | 954 (17.2%) |
| gliner2-fastino-ru | 1371 (24.8%) | 970 (17.5%) | 793 (14.3%) | 586 (10.6%) |
| ru-legal-ner | 1422 (25.7%) | 865 (15.6%) | 793 (14.3%) | 775 (14.0%) |
| ru-legal-ner+homoglyph | 1492 (27.0%) | 914 (16.5%) | 839 (15.2%) | 819 (14.8%) |
| apararti | 1498 (27.1%) | 1406 (25.4%) | 1403 (25.4%) | 1403 (25.4%) |
| gliner-nvidia-ru | 1676 (30.3%) | 1205 (21.8%) | 910 (16.5%) | 593 (10.7%) |
| gliner2-fastino | 1692 (30.6%) | 1196 (21.6%) | 928 (16.8%) | 639 (11.6%) |
| pii-shield-onnx | 1778 (32.1%) | 1168 (21.1%) | 1068 (19.3%) | 1055 (19.1%) |
| opf-kz-ru | 1801 (32.6%) | 1756 (31.7%) | 1756 (31.7%) | 1756 (31.7%) |
| traciora | 1917 (34.7%) | 1837 (33.2%) | 1827 (33.0%) | 1827 (33.0%) |
| opf-ru | 1929 (34.9%) | 1775 (32.1%) | 1760 (31.8%) | 1760 (31.8%) |
| bardsai-eu | 1952 (35.3%) | 1688 (30.5%) | 1673 (30.2%) | 1672 (30.2%) |
| gliner2-hivetrace-omni-ru | 1962 (35.5%) | 1152 (20.8%) | 857 (15.5%) | 603 (10.9%) |
| gliner-nvidia+homoglyph | 1971 (35.6%) | 1445 (26.1%) | 1128 (20.4%) | 721 (13.0%) |
| gliner-nvidia | 2033 (36.8%) | 1485 (26.8%) | 1142 (20.6%) | 732 (13.2%) |
| gravitee-small | 2190 (39.6%) | 1945 (35.2%) | 1939 (35.1%) | 1939 (35.1%) |
| gliner2-hivetrace-omni | 2239 (40.5%) | 1379 (24.9%) | 981 (17.7%) | 673 (12.2%) |
| openai-base | 2248 (40.6%) | 2221 (40.2%) | 2221 (40.2%) | 2221 (40.2%) |
| opf-ru-v2+homoglyph | 2273 (41.1%) | 2221 (40.2%) | 2214 (40.0%) | 2214 (40.0%) |
| opf-ru-v2 | 2285 (41.3%) | 2238 (40.5%) | 2229 (40.3%) | 2229 (40.3%) |
| gliner-pii-edge | 2345 (42.4%) | 258 (4.7%) | 37 (0.7%) | 10 (0.2%) |
| gliner-urchade-ru | 2428 (43.9%) | 1711 (30.9%) | 1328 (24.0%) | 877 (15.9%) |
| gliner-urchade | 2436 (44.0%) | 1597 (28.9%) | 1150 (20.8%) | 686 (12.4%) |
| gliner2-large | 2457 (44.4%) | 1647 (29.8%) | 1228 (22.2%) | 757 (13.7%) |
| nuner-zero | 2464 (44.5%) | 1498 (27.1%) | 1049 (19.0%) | 644 (11.6%) |
| gliner2-hivetrace-uni | 2520 (45.6%) | 1470 (26.6%) | 971 (17.6%) | 458 (8.3%) |
| gliner-stream-pii | 2526 (45.7%) | 1757 (31.8%) | 1392 (25.2%) | 972 (17.6%) |
| openmed-multilingual | 2617 (47.3%) | 1868 (33.8%) | 1771 (32.0%) | 1752 (31.7%) |
| fef2-secret-ru | 2690 (48.6%) | 2606 (47.1%) | 2600 (47.0%) | 2600 (47.0%) |
| gliner25-fastino-ru | 2735 (49.4%) | 2382 (43.1%) | 2199 (39.8%) | 2076 (37.5%) |
| gliner2-vladlinv-ru | 2821 (51.0%) | 2722 (49.2%) | 2675 (48.4%) | 2604 (47.1%) |
| openmed-nemotron | 2846 (51.5%) | 2483 (44.9%) | 2464 (44.5%) | 2463 (44.5%) |
| kalyan-ettin | 3117 (56.4%) | 2428 (43.9%) | 2279 (41.2%) | 2259 (40.8%) |
| gliner2-vladlinv+homoglyph | 3134 (56.7%) | 3047 (55.1%) | 2998 (54.2%) | 2936 (53.1%) |
| gliner2-vladlinv | 3193 (57.7%) | 3105 (56.1%) | 3062 (55.4%) | 3004 (54.3%) |
| ner-ru-yqelz | 3193 (57.7%) | 3121 (56.4%) | 3121 (56.4%) | 3121 (56.4%) |
| ner-ru-gherman+homoglyph | 3300 (59.7%) | 3267 (59.1%) | 3266 (59.0%) | 3266 (59.0%) |
| ner-ru-gherman | 3339 (60.4%) | 3311 (59.9%) | 3309 (59.8%) | 3309 (59.8%) |
| davlan-xlmr | 3417 (61.8%) | 3400 (61.5%) | 3400 (61.5%) | 3400 (61.5%) |
| gliner25-fastino | 3506 (63.4%) | 2952 (53.4%) | 2685 (48.5%) | 2308 (41.7%) |
| davlan-mbert | 3530 (63.8%) | 3508 (63.4%) | 3508 (63.4%) | 3508 (63.4%) |
| gliner-pii-base | 3531 (63.8%) | 1314 (23.8%) | 429 (7.8%) | 77 (1.4%) |
| gliner-multi-v21-ru | 3782 (68.4%) | 2675 (48.4%) | 1738 (31.4%) | 719 (13.0%) |
| gliner-multi-v21 | 3925 (71.0%) | 2619 (47.4%) | 1592 (28.8%) | 681 (12.3%) |
| gliner2-hivetrace-uni-ru | 4015 (72.6%) | 2433 (44.0%) | 1563 (28.3%) | 709 (12.8%) |

Notes: the unit of counting is one gold span after boundary normalization - the same unit in `missed`, in the group table, in the threshold table and in the ensembles; two annotations that share boundaries but not type are two spans, so an empty answer misses 100%. Gold and predicted boundaries are normalized the same way (whitespace, word boundaries, merge of adjacent same-label pieces). `dropped spans` - predicted intervals outside the text: rejected, not clipped to fit. A run made before `run.py` started clipping span ends to the piece it fed the model can carry them; `meta.clipped` counts what the clipping fixes in newer runs. Zero-shot models get the taxonomy of the set as labels (the `labels` field of `meta.json`). `train` - the model was trained on the source of this set (its slices and corrupted copies included): its row is not comparable with the others and stays out of the pooled numbers; an empty `contaminated` list in `benchmark/models.toml` means no evidence of overlap was found, not proof of none. `hidden` - gold spans every character of which is covered; a touched span counts as detected, not as hidden. missed % and char F1 carry 95% bootstrap intervals over rows (1000 resamples, the same rows for every model of the set). The group of a gold type comes from `BENCH/<set>/meta.json`; a model label with no group falls into OTHER. ms/row is a median and includes chunking of long texts; hardware and batch are in the meta line of the prediction file.