# nerel - ru / pii (933 rows, 24369 spans, 0 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 0.180 (82.0% 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 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| nym-base+ov100 | **1959** | 8.0% [7.4%, 8.8%] | 81.5% | 0.728 [0.721, 0.735] | 0.627 | 0.867 | 0.505 | 0.774 | 0.268 | 0/0 | 0 | 0 | 17 | 308M |  |
| nym-base | **2284** | 9.4% [8.6%, 10.2%] | 79.0% | 0.730 [0.723, 0.737] | 0.639 | 0.851 | 0.507 | 0.776 | 0.265 | 0/0 | 0 | 0 | 16 | 308M |  |
| bardsai-eu | **2393** | 9.8% [9.3%, 10.3%] | 85.5% | 0.825 [0.819, 0.831] | 0.797 | 0.855 | 0.738 | 0.869 | 0.165 | 0/0 | 0 | 0 | 603 | - |  |
| stanza-ru | **2542** | 10.4% [9.9%, 11.0%] | 85.4% | 0.840 [0.833, 0.846] | 0.852 | 0.827 | 0.771 | 0.908 | 0.553 | 0/0 | 0 | 0 | 156 | - |  |
| gliner2-fastino | **2554** | 10.5% [10.0%, 10.9%] | 86.1% | 0.831 [0.825, 0.838] | 0.795 | 0.871 | 0.764 | 0.864 | 0.703 | 0/0 | 0 | 0 | 26 | 307M |  |
| fef2-secret-ru | **2662** | 10.9% [10.3%, 11.5%] | 83.1% | 0.892 [0.886, 0.897] | 0.968 | 0.827 | 0.820 | 0.930 | 0.840 | 0/0 | 0 | 0 | 14 | 177M |  |
| davlan-xlmr | **2720** | 11.2% [10.6%, 11.8%] | 84.1% | 0.897 [0.891, 0.902] | 0.975 | 0.830 | 0.810 | 0.934 | 0.657 | 0/0 | 0 | 0 | 16 | 277M |  |
| davlan-mbert | **2739** | 11.2% [10.6%, 11.8%] | 83.6% | 0.891 [0.885, 0.896] | 0.977 | 0.819 | 0.832 | 0.931 | 0.646 | 0/0 | 0 | 0 | 18 | 177M |  |
| gliner2-fastino-ru | **2743** | 11.3% [10.7%, 11.8%] | 84.9% | 0.817 [0.810, 0.825] | 0.778 | 0.860 | 0.760 | 0.850 | 0.736 | 0/0 | 0 | 0 | 44 | 307M |  |
| nym-small | **2810** | 11.5% [10.7%, 12.4%] | 76.4% | 0.739 [0.732, 0.746] | 0.671 | 0.823 | 0.523 | 0.787 | 0.258 | 0/0 | 0 | 0 | 207 | - |  |
| spacy-ru-lg | **2858** | 11.7% [11.0%, 12.4%] | 82.1% | 0.881 [0.875, 0.887] | 0.953 | 0.819 | 0.802 | 0.921 | 0.625 | 0/0 | 0 | 0 | 21 | - |  |
| natasha | **3099** | 12.7% [12.0%, 13.5%] | 81.4% | 0.880 [0.873, 0.886] | 0.965 | 0.808 | 0.807 | 0.919 | 0.629 | 0/0 | 0 | 0 | 6 | - |  |
| nym-base+sent300 | **3109** | 12.8% [11.9%, 13.6%] | 74.5% | 0.734 [0.727, 0.741] | 0.667 | 0.816 | 0.507 | 0.779 | 0.261 | 0/0 | 0 | 0 | 18 | 308M |  |
| gliner25-fastino+sent300 | **3979** | 16.3% [15.7%, 17.0%] | 79.8% | 0.820 [0.814, 0.827] | 0.840 | 0.802 | 0.768 | 0.855 | 0.737 | 0/0 | 0 | 0 | 11 | 287M |  |
| nuner-zero | **5145** | 21.1% [20.3%, 21.8%] | 73.6% | 0.826 [0.820, 0.833] | 0.902 | 0.762 | 0.742 | 0.845 | 0.662 | 0/0 | 0 | 0 | 44 | 449M |  |
| gliner-urchade | **5152** | 21.1% [20.5%, 21.8%] | 75.4% | 0.797 [0.789, 0.803] | 0.840 | 0.758 | 0.718 | 0.836 | 0.690 | 0/0 | 0 | 0 | 23 | 289M |  |
| gliner25-fastino | **5436** | 22.3% [21.6%, 23.0%] | 74.1% | 0.817 [0.811, 0.823] | 0.895 | 0.752 | 0.765 | 0.846 | 0.735 | 0/0 | 0 | 0 | 17 | 287M |  |
| gliner-nvidia+sent300 | **5519** | 22.6% [22.0%, 23.4%] | 66.1% | 0.752 [0.746, 0.759] | 0.848 | 0.676 | 0.665 | 0.826 | 0.631 | 0/0 | 0 | 0 | 52 | 445M |  |
| gliner25-fastino+ov100 | **5609** | 23.0% [22.3%, 23.8%] | 73.3% | 0.813 [0.807, 0.820] | 0.899 | 0.743 | 0.762 | 0.843 | 0.732 | 0/0 | 0 | 0 | 10 | 287M |  |
| gliner25-fastino-ru | **5612** | 23.0% [22.3%, 23.8%] | 73.3% | 0.812 [0.806, 0.818] | 0.889 | 0.748 | 0.760 | 0.840 | 0.757 | 0/0 | 0 | 0 | 10 | 287M |  |
| gliner-multi-v21 | **5737** | 23.5% [22.8%, 24.2%] | 72.3% | 0.801 [0.795, 0.808] | 0.887 | 0.731 | 0.745 | 0.836 | 0.704 | 0/0 | 0 | 0 | 20 | 289M |  |
| ner-ru-yqelz | **6321** | 25.9% [25.2%, 26.7%] | 69.3% | 0.771 [0.764, 0.778] | 0.803 | 0.741 | 0.628 | 0.803 | 0.523 | 0/0 | 0 | 0 | 19 | 559M |  |
| gliner-pii-edge | **6387** | 26.2% [25.4%, 27.0%] | 65.0% | 0.696 [0.689, 0.704] | 0.817 | 0.607 | 0.636 | 0.790 | 0.475 | 0/0 | 0 | 0 | 231 | 45M |  |
| gliner-multi-v21-ru | **6635** | 27.2% [26.4%, 28.0%] | 68.3% | 0.781 [0.773, 0.788] | 0.905 | 0.686 | 0.734 | 0.819 | 0.722 | 0/0 | 0 | 0 | 19 | 289M |  |
| ner-ru-gherman | **6765** | 27.8% [26.7%, 28.9%] | 39.9% | 0.715 [0.706, 0.725] | 0.990 | 0.560 | 0.418 | 0.835 | 0.074 | 0/0 | 0 | 0 | 17 | 177M |  |
| gliner2-hivetrace-omni | **7043** | 28.9% [28.1%, 29.7%] | 67.0% | 0.772 [0.765, 0.778] | 0.884 | 0.684 | 0.717 | 0.800 | 0.635 | 0/0 | 0 | 0 | 27 | 307M |  |
| gliner2-large | **7197** | 29.5% [28.7%, 30.4%] | 66.5% | 0.719 [0.711, 0.726] | 0.761 | 0.681 | 0.623 | 0.750 | 0.589 | 0/0 | 0 | 0 | 62 | 486M |  |
| gliner2-hivetrace-omni-ru | **7775** | 31.9% [31.1%, 32.7%] | 64.1% | 0.730 [0.723, 0.737] | 0.832 | 0.650 | 0.688 | 0.769 | 0.629 | 0/0 | 0 | 0 | 35 | 307M |  |
| gliner-nvidia | **7985** | 32.8% [31.9%, 33.6%] | 56.8% | 0.717 [0.709, 0.725] | 0.934 | 0.581 | 0.642 | 0.789 | 0.620 | 0/0 | 0 | 0 | 41 | 445M |  |
| gliner-pii-base | **8271** | 33.9% [33.1%, 34.9%] | 60.9% | 0.700 [0.692, 0.708] | 0.862 | 0.590 | 0.665 | 0.755 | 0.611 | 0/0 | 0 | 0 | 14 | 166M |  |
| gliner-nvidia+ov100 | **8432** | 34.6% [33.8%, 35.4%] | 55.2% | 0.704 [0.696, 0.712] | 0.939 | 0.563 | 0.634 | 0.777 | 0.615 | 0/0 | 0 | 0 | 46 | 445M |  |
| gliner25-fastino+nochunk | **8625** | 35.4% [34.0%, 37.1%] | 61.0% | 0.764 [0.751, 0.775] | 0.936 | 0.645 | 0.696 | 0.774 | 0.686 | 0/0 | 0 | 0 | 15 | 287M |  |
| gliner25-fastino-ru+nochunk | **8698** | 35.7% [34.2%, 37.4%] | 60.7% | 0.758 [0.745, 0.769] | 0.921 | 0.643 | 0.689 | 0.768 | 0.703 | 0/0 | 0 | 0 | 15 | 287M |  |
| ru-legal-ner+sent300 | **9130** | 37.5% [36.3%, 38.7%] | 55.4% | 0.604 [0.596, 0.612] | 0.608 | 0.600 | 0.340 | 0.647 | 0.425 | 0/0 | 0 | 0 | 14 | 29M |  |
| gliner-nvidia-ru | **9370** | 38.5% [37.6%, 39.4%] | 51.7% | 0.661 [0.653, 0.669] | 0.963 | 0.503 | 0.625 | 0.754 | 0.610 | 0/0 | 0 | 0 | 134 | 445M |  |
| ru-legal-ner | **10437** | 42.8% [41.7%, 44.0%] | 50.3% | 0.588 [0.580, 0.597] | 0.628 | 0.553 | 0.343 | 0.624 | 0.432 | 0/0 | 0 | 0 | 12 | 29M |  |
| ru-legal-ner+ov100 | **10675** | 43.8% [42.6%, 45.1%] | 49.4% | 0.581 [0.572, 0.590] | 0.626 | 0.543 | 0.337 | 0.617 | 0.432 | 0/0 | 0 | 0 | 13 | 29M |  |
| gliner-stream-pii | **10853** | 44.5% [43.6%, 45.5%] | 49.3% | 0.638 [0.629, 0.647] | 0.915 | 0.490 | 0.571 | 0.700 | 0.484 | 0/0 | 0 | 0 | 121 | 677M |  |
| spacy-alrosait | **11833** | 48.6% [47.4%, 49.7%] | 44.9% | 0.620 [0.609, 0.631] | 0.954 | 0.459 | 0.541 | 0.676 | 0.159 | 0/0 | 0 | 0 | 20 | - |  |
| kalyan-ettin | **12048** | 49.4% [48.4%, 50.4%] | 29.6% | 0.543 [0.534, 0.553] | 0.832 | 0.403 | 0.327 | 0.626 | 0.427 | 0/0 | 0 | 0 | 52 | 68M |  |
| gliner2-hivetrace-uni | **12237** | 50.2% [49.3%, 51.2%] | 45.7% | 0.597 [0.588, 0.606] | 0.933 | 0.439 | 0.579 | 0.654 | 0.518 | 0/0 | 0 | 0 | 35 | 147M |  |
| mmbert32k | **12467** | 51.2% [50.0%, 52.4%] | 34.1% | 0.542 [0.532, 0.553] | 0.812 | 0.407 | 0.377 | 0.560 | 0.397 | 0/0 | 0 | 0 | 20 | 308M |  |
| gliner-urchade-ru | **13036** | 53.5% [52.4%, 54.7%] | 43.4% | 0.527 [0.517, 0.538] | 0.775 | 0.399 | 0.481 | 0.603 | 0.459 | 0/0 | 0 | 0 | 13 | 289M |  |
| openmed-nemotron | **13198** | 54.2% [53.1%, 55.3%] | 25.0% | 0.468 [0.460, 0.476] | 0.589 | 0.388 | 0.210 | 0.535 | 0.308 | 0/0 | 0 | 0 | 127 | 1.4B |  |
| pii-shield-onnx | **14749** | 60.5% [59.4%, 61.7%] | 31.0% | 0.485 [0.473, 0.496] | 0.748 | 0.359 | 0.364 | 0.501 | 0.351 | 0/0 | 0 | 0 | 358 | - |  |
| opf-ru | **15177** | 62.3% [61.2%, 63.3%] | 29.6% | 0.495 [0.486, 0.507] | 0.853 | 0.349 | 0.368 | 0.522 | 0.392 | 0/0 | 0 | 0 | 130 | 1.4B |  |
| gliner2-vladlinv-ru | **15224** | 62.5% [61.3%, 63.6%] | 36.0% | 0.547 [0.534, 0.559] | 0.969 | 0.381 | 0.510 | 0.542 | 0.490 | 0/0 | 0 | 0 | 11 | 287M |  |
| gliner2-vladlinv | **15259** | 62.6% [61.4%, 63.8%] | 36.0% | 0.548 [0.535, 0.560] | 0.982 | 0.380 | 0.511 | 0.542 | 0.509 | 0/0 | 0 | 0 | 22 | 287M |  |
| openmed-multilingual | **15652** | 64.2% [63.3%, 65.1%] | 18.9% | 0.417 [0.409, 0.426] | 0.712 | 0.295 | 0.196 | 0.452 | 0.311 | 0/0 | 0 | 0 | 94 | 1.4B |  |
| mmbert32k+nochunk | **15985** | 65.6% [64.4%, 66.8%] | 17.7% | 0.393 [0.381, 0.405] | 0.753 | 0.266 | 0.217 | 0.432 | 0.277 | 0/0 | 0 | 0 | 20 | 308M |  |
| ru-pii-ner | **16208** | 66.5% [65.3%, 67.7%] | 32.8% | 0.495 [0.483, 0.507] | 0.814 | 0.356 | 0.413 | 0.482 | 0.366 | 0/0 | 0 | 0 | 232 | 358M |  |
| traciora | **17236** | 70.7% [69.6%, 71.9%] | 24.7% | 0.434 [0.423, 0.447] | 0.784 | 0.300 | 0.230 | 0.443 | 0.318 | 0/0 | 0 | 0 | 780 | 1.4B |  |
| pplx+sent300 | **17287** | 70.9% [69.9%, 72.1%] | 27.7% | 0.441 [0.427, 0.453] | 0.836 | 0.299 | 0.394 | 0.431 | 0.414 | 0/0 | 0 | 0 | 375 | 596M |  |
| opf-ru-v2+sent300 | **18777** | 77.1% [76.2%, 77.9%] | 20.1% | 0.387 [0.375, 0.398] | 0.874 | 0.248 | 0.248 | 0.369 | 0.316 | 0/0 | 0 | 0 | 27 | 1.4B |  |
| apararti | **18792** | 77.1% [76.1%, 78.1%] | 19.5% | 0.354 [0.342, 0.367] | 0.652 | 0.243 | 0.199 | 0.347 | 0.278 | 0/0 | 0 | 0 | 195 | 1.4B |  |
| opf-ru-v2+ov100 | **19643** | 80.6% [79.7%, 81.5%] | 16.8% | 0.345 [0.332, 0.358] | 0.888 | 0.214 | 0.216 | 0.322 | 0.287 | 0/0 | 0 | 0 | 20 | 1.4B |  |
| opf-ru-v2 | **19675** | 80.7% [79.8%, 81.6%] | 16.7% | 0.343 [0.330, 0.357] | 0.897 | 0.212 | 0.216 | 0.320 | 0.287 | 0/0 | 0 | 0 | 747 | 1.4B |  |
| gliner2-hivetrace-uni-ru | **19827** | 81.4% [80.6%, 82.1%] | 17.7% | 0.300 [0.290, 0.311] | 0.934 | 0.179 | 0.281 | 0.311 | 0.267 | 0/0 | 0 | 0 | 22 | 147M |  |
| pplx+ov100 | **20552** | 84.3% [83.3%, 85.3%] | 15.3% | 0.283 [0.269, 0.299] | 0.846 | 0.170 | 0.249 | 0.265 | 0.267 | 0/0 | 0 | 0 | 224 | 596M |  |
| pplx | **21282** | 87.3% [86.3%, 88.3%] | 12.4% | 0.241 [0.226, 0.257] | 0.837 | 0.141 | 0.208 | 0.220 | 0.227 | 0/0 | 0 | 0 | 194 | 596M |  |
| opf-kz-ru | **21493** | 88.2% [87.6%, 88.8%] | 9.4% | 0.219 [0.209, 0.230] | 0.670 | 0.131 | 0.128 | 0.202 | 0.187 | 0/0 | 0 | 0 | 186 | 1.4B |  |
| openai-base | **22006** | 90.3% [89.5%, 91.1%] | 8.9% | 0.199 [0.185, 0.214] | 0.834 | 0.113 | 0.116 | 0.175 | 0.176 | 0/0 | 0 | 0 | 217 | 1.4B |  |
| gravitee-small | **23450** | 96.2% [95.8%, 96.6%] | 2.8% | 0.067 [0.060, 0.074] | 0.400 | 0.036 | 0.028 | 0.070 | 0.049 | 0/0 | 0 | 0 | 34 | 29M |  |
| rules-ru | **24333** | 99.9% [99.8%, 99.9%] | 0.1% | 0.004 [0.002, 0.007] | 0.162 | 0.002 | 0.001 | 0.003 | 0.004 | 0/0 | 0 | 0 | 2 | - |  |

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`: nym-base ≈ bardsai-eu; stanza-ru ≈ gliner2-fastino; gliner2-fastino ≈ fef2-secret-ru; fef2-secret-ru ≈ davlan-xlmr; davlan-xlmr ≈ davlan-mbert; davlan-mbert ≈ gliner2-fastino-ru; gliner2-fastino-ru ≈ nym-small; nym-small ≈ spacy-ru-lg; natasha ≈ nym-base+sent300; nuner-zero ≈ gliner-urchade; gliner25-fastino ≈ gliner-nvidia+sent300; gliner-nvidia+sent300 ≈ gliner25-fastino+ov100; gliner25-fastino+ov100 ≈ gliner25-fastino-ru; gliner25-fastino-ru ≈ gliner-multi-v21; ner-ru-yqelz ≈ gliner-pii-edge; gliner-multi-v21-ru ≈ ner-ru-gherman; ner-ru-gherman ≈ gliner2-hivetrace-omni; gliner2-hivetrace-omni ≈ gliner2-large; gliner-pii-base ≈ gliner-nvidia+ov100; gliner-nvidia+ov100 ≈ gliner25-fastino+nochunk; gliner25-fastino+nochunk ≈ gliner25-fastino-ru+nochunk; gliner25-fastino-ru+nochunk ≈ ru-legal-ner+sent300; ru-legal-ner+sent300 ≈ gliner-nvidia-ru; ru-legal-ner+ov100 ≈ gliner-stream-pii; spacy-alrosait ≈ kalyan-ettin; kalyan-ettin ≈ gliner2-hivetrace-uni; gliner2-hivetrace-uni ≈ mmbert32k; gliner-urchade-ru ≈ openmed-nemotron; opf-ru ≈ gliner2-vladlinv-ru; gliner2-vladlinv-ru ≈ gliner2-vladlinv; openmed-multilingual ≈ mmbert32k+nochunk; mmbert32k+nochunk ≈ ru-pii-ner; traciora ≈ pplx+sent300; opf-ru-v2+sent300 ≈ apararti; opf-ru-v2+ov100 ≈ opf-ru-v2; opf-ru-v2 ≈ gliner2-hivetrace-uni-ru; pplx ≈ opf-kz-ru

## Missed by group

| model | PERSON | ADDRESS | ORG |
|---|---|---|---|
| spans in gold | 9496 | 8861 | 6012 |
| nym-base+ov100 | 180 (1.9%) | 746 (8.4%) | 1033 (17.2%) |
| nym-base | 244 (2.6%) | 880 (9.9%) | 1160 (19.3%) |
| bardsai-eu | 146 (1.5%) | 1364 (15.4%) | 883 (14.7%) |
| stanza-ru | 150 (1.6%) | 1296 (14.6%) | 1096 (18.2%) |
| gliner2-fastino | 1170 (12.3%) | 1043 (11.8%) | 341 (5.7%) |
| fef2-secret-ru | 202 (2.1%) | 1511 (17.1%) | 949 (15.8%) |
| davlan-xlmr | 181 (1.9%) | 1574 (17.8%) | 965 (16.1%) |
| davlan-mbert | 186 (2.0%) | 1531 (17.3%) | 1022 (17.0%) |
| gliner2-fastino-ru | 1164 (12.3%) | 1213 (13.7%) | 366 (6.1%) |
| nym-small | 347 (3.7%) | 1081 (12.2%) | 1382 (23.0%) |
| spacy-ru-lg | 302 (3.2%) | 1509 (17.0%) | 1047 (17.4%) |
| natasha | 404 (4.3%) | 1577 (17.8%) | 1118 (18.6%) |
| nym-base+sent300 | 443 (4.7%) | 1188 (13.4%) | 1478 (24.6%) |
| gliner25-fastino+sent300 | 965 (10.2%) | 2025 (22.9%) | 989 (16.5%) |
| nuner-zero | 1488 (15.7%) | 2809 (31.7%) | 848 (14.1%) |
| gliner-urchade | 1944 (20.5%) | 1933 (21.8%) | 1275 (21.2%) |
| gliner25-fastino | 1790 (18.9%) | 2341 (26.4%) | 1305 (21.7%) |
| gliner-nvidia+sent300 | 1736 (18.3%) | 2010 (22.7%) | 1773 (29.5%) |
| gliner25-fastino+ov100 | 1870 (19.7%) | 2379 (26.8%) | 1360 (22.6%) |
| gliner25-fastino-ru | 1843 (19.4%) | 2421 (27.3%) | 1348 (22.4%) |
| gliner-multi-v21 | 1903 (20.0%) | 2413 (27.2%) | 1421 (23.6%) |
| ner-ru-yqelz | 3142 (33.1%) | 1467 (16.6%) | 1712 (28.5%) |
| gliner-pii-edge | 1849 (19.5%) | 2547 (28.7%) | 1991 (33.1%) |
| gliner-multi-v21-ru | 2588 (27.3%) | 2446 (27.6%) | 1601 (26.6%) |
| ner-ru-gherman | 264 (2.8%) | 1999 (22.6%) | 4502 (74.9%) |
| gliner2-hivetrace-omni | 2669 (28.1%) | 2635 (29.7%) | 1739 (28.9%) |
| gliner2-large | 2419 (25.5%) | 3355 (37.9%) | 1423 (23.7%) |
| gliner2-hivetrace-omni-ru | 2965 (31.2%) | 2948 (33.3%) | 1862 (31.0%) |
| gliner-nvidia | 2704 (28.5%) | 2716 (30.7%) | 2565 (42.7%) |
| gliner-pii-base | 2407 (25.3%) | 3856 (43.5%) | 2008 (33.4%) |
| gliner-nvidia+ov100 | 2803 (29.5%) | 2887 (32.6%) | 2742 (45.6%) |
| gliner25-fastino+nochunk | 3791 (39.9%) | 2855 (32.2%) | 1979 (32.9%) |
| gliner25-fastino-ru+nochunk | 3863 (40.7%) | 2845 (32.1%) | 1990 (33.1%) |
| ru-legal-ner+sent300 | 673 (7.1%) | 5248 (59.2%) | 3209 (53.4%) |
| gliner-nvidia-ru | 2859 (30.1%) | 3126 (35.3%) | 3385 (56.3%) |
| ru-legal-ner | 985 (10.4%) | 5819 (65.7%) | 3633 (60.4%) |
| ru-legal-ner+ov100 | 1049 (11.0%) | 5932 (66.9%) | 3694 (61.4%) |
| gliner-stream-pii | 3656 (38.5%) | 4192 (47.3%) | 3005 (50.0%) |
| spacy-alrosait | 3112 (32.8%) | 3836 (43.3%) | 4885 (81.3%) |
| kalyan-ettin | 2842 (29.9%) | 4504 (50.8%) | 4702 (78.2%) |
| gliner2-hivetrace-uni | 3966 (41.8%) | 4660 (52.6%) | 3611 (60.1%) |
| mmbert32k | 1326 (14.0%) | 6567 (74.1%) | 4574 (76.1%) |
| gliner-urchade-ru | 9143 (96.3%) | 2379 (26.8%) | 1514 (25.2%) |
| openmed-nemotron | 3895 (41.0%) | 5135 (58.0%) | 4168 (69.3%) |
| pii-shield-onnx | 4603 (48.5%) | 5074 (57.3%) | 5072 (84.4%) |
| opf-ru | 2787 (29.3%) | 7271 (82.1%) | 5119 (85.1%) |
| gliner2-vladlinv-ru | 984 (10.4%) | 8360 (94.3%) | 5880 (97.8%) |
| gliner2-vladlinv | 999 (10.5%) | 8393 (94.7%) | 5867 (97.6%) |
| openmed-multilingual | 4901 (51.6%) | 5787 (65.3%) | 4964 (82.6%) |
| mmbert32k+nochunk | 3418 (36.0%) | 7492 (84.6%) | 5075 (84.4%) |
| ru-pii-ner | 2053 (21.6%) | 8189 (92.4%) | 5966 (99.2%) |
| traciora | 4768 (50.2%) | 7416 (83.7%) | 5052 (84.0%) |
| pplx+sent300 | 3442 (36.2%) | 7986 (90.1%) | 5859 (97.5%) |
| opf-ru-v2+sent300 | 5201 (54.8%) | 8127 (91.7%) | 5449 (90.6%) |
| apararti | 5629 (59.3%) | 7735 (87.3%) | 5428 (90.3%) |
| opf-ru-v2+ov100 | 5736 (60.4%) | 8320 (93.9%) | 5587 (92.9%) |
| opf-ru-v2 | 5745 (60.5%) | 8318 (93.9%) | 5612 (93.3%) |
| gliner2-hivetrace-uni-ru | 7225 (76.1%) | 7865 (88.8%) | 4737 (78.8%) |
| pplx+ov100 | 5976 (62.9%) | 8605 (97.1%) | 5971 (99.3%) |
| pplx | 6618 (69.7%) | 8675 (97.9%) | 5989 (99.6%) |
| opf-kz-ru | 7209 (75.9%) | 8426 (95.1%) | 5858 (97.4%) |
| openai-base | 7535 (79.3%) | 8588 (96.9%) | 5883 (97.9%) |
| gravitee-small | 8991 (94.7%) | 8644 (97.6%) | 5815 (96.7%) |
| rules-ru | 9496 (100.0%) | 8854 (99.9%) | 5983 (99.5%) |

## Char recall by gold type

| type | group | nym-base+ov100 | nym-base | bardsai-eu | stanza-ru | gliner2-fastino | fef2-secret-ru | davlan-xlmr | davlan-mbert | gliner2-fastino-ru | nym-small | spacy-ru-lg | natasha | nym-base+sent300 | gliner25-fastino+sent300 | nuner-zero | gliner-urchade | gliner25-fastino | gliner-nvidia+sent300 | gliner25-fastino+ov100 | gliner25-fastino-ru | gliner-multi-v21 | ner-ru-yqelz | gliner-pii-edge | gliner-multi-v21-ru | ner-ru-gherman | gliner2-hivetrace-omni | gliner2-large | gliner2-hivetrace-omni-ru | gliner-nvidia | gliner-pii-base | gliner-nvidia+ov100 | gliner25-fastino+nochunk | gliner25-fastino-ru+nochunk | ru-legal-ner+sent300 | gliner-nvidia-ru | ru-legal-ner | ru-legal-ner+ov100 | gliner-stream-pii | spacy-alrosait | kalyan-ettin | gliner2-hivetrace-uni | mmbert32k | gliner-urchade-ru | openmed-nemotron | pii-shield-onnx | opf-ru | gliner2-vladlinv-ru | gliner2-vladlinv | openmed-multilingual | mmbert32k+nochunk | ru-pii-ner | traciora | pplx+sent300 | opf-ru-v2+sent300 | apararti | opf-ru-v2+ov100 | opf-ru-v2 | gliner2-hivetrace-uni-ru | pplx+ov100 | pplx | opf-kz-ru | openai-base | gravitee-small | rules-ru |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| CITY | ADDRESS | 0.972 | 0.966 | 0.911 | 0.853 | 0.891 | 0.841 | 0.843 | 0.848 | 0.878 | 0.953 | 0.838 | 0.827 | 0.959 | 0.784 | 0.751 | 0.784 | 0.759 | 0.777 | 0.750 | 0.745 | 0.768 | 0.918 | 0.721 | 0.779 | 0.865 | 0.666 | 0.647 | 0.591 | 0.683 | 0.596 | 0.663 | 0.702 | 0.714 | 0.556 | 0.624 | 0.478 | 0.461 | 0.567 | 0.574 | 0.587 | 0.429 | 0.368 | 0.733 | 0.394 | 0.372 | 0.207 | 0.076 | 0.057 | 0.512 | 0.233 | 0.123 | 0.180 | 0.164 | 0.087 | 0.168 | 0.059 | 0.059 | 0.227 | 0.052 | 0.031 | 0.066 | 0.041 | 0.035 | 0.001 |
| COUNTRY | ADDRESS | 0.890 | 0.867 | 0.734 | 0.738 | 0.816 | 0.704 | 0.696 | 0.703 | 0.791 | 0.832 | 0.713 | 0.705 | 0.812 | 0.744 | 0.583 | 0.748 | 0.695 | 0.729 | 0.691 | 0.691 | 0.706 | 0.756 | 0.626 | 0.710 | 0.685 | 0.597 | 0.603 | 0.577 | 0.660 | 0.514 | 0.642 | 0.623 | 0.629 | 0.246 | 0.638 | 0.196 | 0.186 | 0.447 | 0.537 | 0.429 | 0.414 | 0.133 | 0.729 | 0.419 | 0.476 | 0.100 | 0.041 | 0.040 | 0.202 | 0.064 | 0.018 | 0.117 | 0.020 | 0.052 | 0.058 | 0.041 | 0.040 | 0.042 | 0.005 | 0.004 | 0.024 | 0.011 | 0.013 | 0.001 |
| DISTRICT | ADDRESS | 0.881 | 0.890 | 0.940 | 0.919 | 0.908 | 0.945 | 0.926 | 0.893 | 0.869 | 0.833 | 0.937 | 0.936 | 0.896 | 0.746 | 0.689 | 0.782 | 0.739 | 0.722 | 0.773 | 0.772 | 0.699 | 0.952 | 0.504 | 0.623 | 0.488 | 0.849 | 0.566 | 0.723 | 0.571 | 0.253 | 0.542 | 0.613 | 0.780 | 0.611 | 0.491 | 0.472 | 0.436 | 0.407 | 0.418 | 0.455 | 0.592 | 0.284 | 0.711 | 0.315 | 0.487 | 0.204 | 0.043 | 0.054 | 0.486 | 0.180 | 0.199 | 0.216 | 0.287 | 0.116 | 0.205 | 0.114 | 0.098 | 0.106 | 0.100 | 0.081 | 0.067 | 0.086 | 0.069 | 0.000 |
| FACILITY | ADDRESS | 0.742 | 0.716 | 0.716 | 0.750 | 0.755 | 0.650 | 0.713 | 0.698 | 0.725 | 0.671 | 0.658 | 0.634 | 0.651 | 0.509 | 0.450 | 0.640 | 0.489 | 0.476 | 0.464 | 0.435 | 0.479 | 0.716 | 0.452 | 0.433 | 0.264 | 0.696 | 0.512 | 0.671 | 0.335 | 0.341 | 0.300 | 0.441 | 0.411 | 0.498 | 0.185 | 0.435 | 0.413 | 0.312 | 0.313 | 0.207 | 0.305 | 0.224 | 0.504 | 0.268 | 0.218 | 0.235 | 0.056 | 0.059 | 0.295 | 0.149 | 0.089 | 0.246 | 0.145 | 0.155 | 0.273 | 0.113 | 0.114 | 0.072 | 0.066 | 0.068 | 0.093 | 0.092 | 0.026 | 0.000 |
| LOCATION | ADDRESS | 0.679 | 0.657 | 0.680 | 0.695 | 0.757 | 0.687 | 0.656 | 0.654 | 0.713 | 0.614 | 0.670 | 0.650 | 0.607 | 0.528 | 0.594 | 0.530 | 0.483 | 0.485 | 0.457 | 0.545 | 0.492 | 0.710 | 0.522 | 0.424 | 0.282 | 0.578 | 0.494 | 0.578 | 0.351 | 0.304 | 0.331 | 0.380 | 0.495 | 0.343 | 0.323 | 0.257 | 0.270 | 0.385 | 0.306 | 0.316 | 0.256 | 0.143 | 0.453 | 0.311 | 0.291 | 0.163 | 0.028 | 0.024 | 0.287 | 0.075 | 0.106 | 0.175 | 0.053 | 0.112 | 0.133 | 0.079 | 0.088 | 0.040 | 0.016 | 0.008 | 0.036 | 0.025 | 0.016 | 0.000 |
| ORGANIZATION | ORG | 0.745 | 0.719 | 0.765 | 0.679 | 0.863 | 0.717 | 0.720 | 0.690 | 0.862 | 0.671 | 0.701 | 0.682 | 0.665 | 0.759 | 0.743 | 0.678 | 0.709 | 0.629 | 0.697 | 0.711 | 0.645 | 0.704 | 0.408 | 0.607 | 0.129 | 0.619 | 0.673 | 0.589 | 0.500 | 0.472 | 0.470 | 0.626 | 0.626 | 0.338 | 0.340 | 0.274 | 0.261 | 0.347 | 0.119 | 0.133 | 0.242 | 0.121 | 0.648 | 0.237 | 0.102 | 0.087 | 0.013 | 0.014 | 0.108 | 0.077 | 0.005 | 0.120 | 0.014 | 0.062 | 0.069 | 0.047 | 0.043 | 0.129 | 0.003 | 0.002 | 0.017 | 0.013 | 0.025 | 0.006 |
| PERSON | PERSON | 0.975 | 0.969 | 0.985 | 0.987 | 0.913 | 0.979 | 0.980 | 0.979 | 0.906 | 0.961 | 0.970 | 0.964 | 0.953 | 0.913 | 0.876 | 0.851 | 0.853 | 0.722 | 0.845 | 0.850 | 0.860 | 0.746 | 0.783 | 0.787 | 0.900 | 0.772 | 0.762 | 0.733 | 0.650 | 0.771 | 0.644 | 0.702 | 0.693 | 0.952 | 0.616 | 0.929 | 0.923 | 0.626 | 0.733 | 0.610 | 0.630 | 0.761 | 0.034 | 0.536 | 0.546 | 0.689 | 0.893 | 0.892 | 0.432 | 0.509 | 0.808 | 0.540 | 0.653 | 0.513 | 0.456 | 0.452 | 0.451 | 0.262 | 0.391 | 0.325 | 0.277 | 0.247 | 0.055 | 0.000 |
| STATE_OR_PROVINCE | ADDRESS | 0.951 | 0.940 | 0.950 | 0.923 | 0.921 | 0.927 | 0.938 | 0.921 | 0.911 | 0.934 | 0.940 | 0.944 | 0.933 | 0.749 | 0.760 | 0.845 | 0.740 | 0.837 | 0.745 | 0.734 | 0.677 | 0.957 | 0.637 | 0.704 | 0.746 | 0.794 | 0.568 | 0.786 | 0.806 | 0.352 | 0.771 | 0.697 | 0.626 | 0.546 | 0.720 | 0.465 | 0.431 | 0.540 | 0.541 | 0.519 | 0.461 | 0.372 | 0.743 | 0.338 | 0.591 | 0.252 | 0.038 | 0.043 | 0.447 | 0.209 | 0.199 | 0.272 | 0.181 | 0.110 | 0.192 | 0.087 | 0.082 | 0.147 | 0.035 | 0.029 | 0.058 | 0.027 | 0.022 | 0.000 |

## char F1 by domain

| domain | n | nym-base+ov100 | nym-base | bardsai-eu | stanza-ru | gliner2-fastino | fef2-secret-ru | davlan-xlmr | davlan-mbert | gliner2-fastino-ru | nym-small | spacy-ru-lg | natasha | nym-base+sent300 | gliner25-fastino+sent300 | nuner-zero | gliner-urchade | gliner25-fastino | gliner-nvidia+sent300 | gliner25-fastino+ov100 | gliner25-fastino-ru | gliner-multi-v21 | ner-ru-yqelz | gliner-pii-edge | gliner-multi-v21-ru | ner-ru-gherman | gliner2-hivetrace-omni | gliner2-large | gliner2-hivetrace-omni-ru | gliner-nvidia | gliner-pii-base | gliner-nvidia+ov100 | gliner25-fastino+nochunk | gliner25-fastino-ru+nochunk | ru-legal-ner+sent300 | gliner-nvidia-ru | ru-legal-ner | ru-legal-ner+ov100 | gliner-stream-pii | spacy-alrosait | kalyan-ettin | gliner2-hivetrace-uni | mmbert32k | gliner-urchade-ru | openmed-nemotron | pii-shield-onnx | opf-ru | gliner2-vladlinv-ru | gliner2-vladlinv | openmed-multilingual | mmbert32k+nochunk | ru-pii-ner | traciora | pplx+sent300 | opf-ru-v2+sent300 | apararti | opf-ru-v2+ov100 | opf-ru-v2 | gliner2-hivetrace-uni-ru | pplx+ov100 | pplx | opf-kz-ru | openai-base | gravitee-small | rules-ru |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| dev | 94 | 0.727 | 0.730 | 0.825 | 0.832 | 0.830 | 0.893 | 0.901 | 0.889 | 0.825 | 0.734 | 0.886 | 0.882 | 0.739 | 0.826 | 0.827 | 0.786 | 0.816 | 0.746 | 0.813 | 0.814 | 0.787 | 0.761 | 0.686 | 0.765 | 0.711 | 0.777 | 0.708 | 0.736 | 0.722 | 0.697 | 0.708 | 0.777 | 0.766 | 0.605 | 0.659 | 0.595 | 0.580 | 0.662 | 0.607 | 0.557 | 0.614 | 0.569 | 0.504 | 0.466 | 0.486 | 0.506 | 0.554 | 0.552 | 0.446 | 0.411 | 0.508 | 0.439 | 0.469 | 0.396 | 0.375 | 0.367 | 0.355 | 0.287 | 0.310 | 0.267 | 0.231 | 0.235 | 0.071 | 0.003 |
| test | 93 | 0.711 | 0.715 | 0.820 | 0.838 | 0.810 | 0.891 | 0.897 | 0.889 | 0.810 | 0.733 | 0.877 | 0.871 | 0.726 | 0.813 | 0.819 | 0.792 | 0.823 | 0.745 | 0.823 | 0.822 | 0.795 | 0.780 | 0.696 | 0.782 | 0.721 | 0.777 | 0.714 | 0.731 | 0.716 | 0.694 | 0.715 | 0.785 | 0.781 | 0.600 | 0.654 | 0.586 | 0.575 | 0.641 | 0.637 | 0.548 | 0.606 | 0.531 | 0.529 | 0.477 | 0.518 | 0.501 | 0.537 | 0.540 | 0.419 | 0.391 | 0.508 | 0.428 | 0.488 | 0.396 | 0.362 | 0.365 | 0.365 | 0.294 | 0.357 | 0.319 | 0.243 | 0.229 | 0.076 | 0.002 |
| train | 746 | 0.730 | 0.732 | 0.826 | 0.841 | 0.834 | 0.892 | 0.896 | 0.891 | 0.817 | 0.741 | 0.881 | 0.880 | 0.734 | 0.821 | 0.827 | 0.799 | 0.816 | 0.754 | 0.812 | 0.810 | 0.804 | 0.771 | 0.698 | 0.782 | 0.715 | 0.770 | 0.721 | 0.729 | 0.716 | 0.701 | 0.702 | 0.759 | 0.754 | 0.604 | 0.662 | 0.588 | 0.582 | 0.635 | 0.619 | 0.541 | 0.594 | 0.541 | 0.530 | 0.467 | 0.480 | 0.493 | 0.547 | 0.548 | 0.413 | 0.391 | 0.492 | 0.435 | 0.431 | 0.384 | 0.351 | 0.340 | 0.339 | 0.303 | 0.270 | 0.227 | 0.214 | 0.191 | 0.065 | 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+ov100 | 1959 (8.0%) | 1627 (6.7%) | 1620 (6.6%) | 1620 (6.6%) |
| nym-base | 2284 (9.4%) | 1903 (7.8%) | 1899 (7.8%) | 1899 (7.8%) |
| bardsai-eu | 2393 (9.8%) | 2297 (9.4%) | 2295 (9.4%) | 2295 (9.4%) |
| gliner2-fastino | 2554 (10.5%) | 1727 (7.1%) | 1334 (5.5%) | 864 (3.5%) |
| fef2-secret-ru | 2662 (10.9%) | 2648 (10.9%) | 2648 (10.9%) | 2648 (10.9%) |
| davlan-xlmr | 2720 (11.2%) | 2698 (11.1%) | 2698 (11.1%) | 2698 (11.1%) |
| davlan-mbert | 2739 (11.2%) | 2707 (11.1%) | 2707 (11.1%) | 2707 (11.1%) |
| gliner2-fastino-ru | 2743 (11.3%) | 1742 (7.1%) | 1273 (5.2%) | 726 (3.0%) |
| nym-small | 2810 (11.5%) | 2352 (9.7%) | 2336 (9.6%) | 2336 (9.6%) |
| nym-base+sent300 | 3109 (12.8%) | 2700 (11.1%) | 2695 (11.1%) | 2695 (11.1%) |
| gliner25-fastino+sent300 | 3979 (16.3%) | 2833 (11.6%) | 2359 (9.7%) | 1794 (7.4%) |
| nuner-zero | 5145 (21.1%) | 2622 (10.8%) | 1837 (7.5%) | 1122 (4.6%) |
| gliner-urchade | 5152 (21.1%) | 3757 (15.4%) | 3248 (13.3%) | 2659 (10.9%) |
| gliner25-fastino | 5436 (22.3%) | 3884 (15.9%) | 3085 (12.7%) | 2127 (8.7%) |
| gliner-nvidia+sent300 | 5519 (22.6%) | 3818 (15.7%) | 2922 (12.0%) | 1834 (7.5%) |
| gliner25-fastino+ov100 | 5609 (23.0%) | 3916 (16.1%) | 3049 (12.5%) | 2108 (8.7%) |
| gliner25-fastino-ru | 5612 (23.0%) | 4100 (16.8%) | 3269 (13.4%) | 2285 (9.4%) |
| gliner-multi-v21 | 5737 (23.5%) | 3551 (14.6%) | 2671 (11.0%) | 1858 (7.6%) |
| ner-ru-yqelz | 6321 (25.9%) | 5989 (24.6%) | 5979 (24.5%) | 5979 (24.5%) |
| gliner-pii-edge | 6387 (26.2%) | 712 (2.9%) | 180 (0.7%) | 32 (0.1%) |
| gliner-multi-v21-ru | 6635 (27.2%) | 4136 (17.0%) | 3108 (12.8%) | 2092 (8.6%) |
| ner-ru-gherman | 6765 (27.8%) | 6577 (27.0%) | 6543 (26.8%) | 6540 (26.8%) |
| gliner2-hivetrace-omni | 7043 (28.9%) | 4906 (20.1%) | 4011 (16.5%) | 3072 (12.6%) |
| gliner2-large | 7197 (29.5%) | 4028 (16.5%) | 2813 (11.5%) | 1694 (7.0%) |
| gliner2-hivetrace-omni-ru | 7775 (31.9%) | 5299 (21.7%) | 4225 (17.3%) | 3051 (12.5%) |
| gliner-nvidia | 7985 (32.8%) | 6149 (25.2%) | 4984 (20.5%) | 3540 (14.5%) |
| gliner-pii-base | 8271 (33.9%) | 1472 (6.0%) | 417 (1.7%) | 45 (0.2%) |
| gliner-nvidia+ov100 | 8432 (34.6%) | 6517 (26.7%) | 5341 (21.9%) | 3841 (15.8%) |
| gliner25-fastino+nochunk | 8625 (35.4%) | 7337 (30.1%) | 6865 (28.2%) | 6512 (26.7%) |
| gliner25-fastino-ru+nochunk | 8698 (35.7%) | 7423 (30.5%) | 6965 (28.6%) | 6646 (27.3%) |
| ru-legal-ner+sent300 | 9130 (37.5%) | 7497 (30.8%) | 7113 (29.2%) | 7059 (29.0%) |
| gliner-nvidia-ru | 9370 (38.5%) | 7620 (31.3%) | 6528 (26.8%) | 4958 (20.3%) |
| ru-legal-ner | 10437 (42.8%) | 8861 (36.4%) | 8559 (35.1%) | 8516 (34.9%) |
| ru-legal-ner+ov100 | 10675 (43.8%) | 9041 (37.1%) | 8717 (35.8%) | 8673 (35.6%) |
| gliner-stream-pii | 10853 (44.5%) | 7225 (29.6%) | 5529 (22.7%) | 3831 (15.7%) |
| kalyan-ettin | 12048 (49.4%) | 10323 (42.4%) | 10195 (41.8%) | 10188 (41.8%) |
| gliner2-hivetrace-uni | 12237 (50.2%) | 8332 (34.2%) | 6189 (25.4%) | 3746 (15.4%) |
| mmbert32k | 12467 (51.2%) | 9863 (40.5%) | 9375 (38.5%) | 9325 (38.3%) |
| gliner-urchade-ru | 13036 (53.5%) | 11296 (46.4%) | 10329 (42.4%) | 8769 (36.0%) |
| openmed-nemotron | 13198 (54.2%) | 11185 (45.9%) | 11037 (45.3%) | 11032 (45.3%) |
| pii-shield-onnx | 14749 (60.5%) | 13493 (55.4%) | 13407 (55.0%) | 13402 (55.0%) |
| opf-ru | 15177 (62.3%) | 15093 (61.9%) | 15093 (61.9%) | 15093 (61.9%) |
| gliner2-vladlinv-ru | 15224 (62.5%) | 15027 (61.7%) | 14893 (61.1%) | 14676 (60.2%) |
| gliner2-vladlinv | 15259 (62.6%) | 15095 (61.9%) | 14979 (61.5%) | 14796 (60.7%) |
| openmed-multilingual | 15652 (64.2%) | 14341 (58.8%) | 14311 (58.7%) | 14311 (58.7%) |
| mmbert32k+nochunk | 15985 (65.6%) | 12726 (52.2%) | 12007 (49.3%) | 11927 (48.9%) |
| traciora | 17236 (70.7%) | 17217 (70.7%) | 17217 (70.7%) | 17217 (70.7%) |
| opf-ru-v2+sent300 | 18777 (77.1%) | 18770 (77.0%) | 18770 (77.0%) | 18770 (77.0%) |
| apararti | 18792 (77.1%) | 18697 (76.7%) | 18696 (76.7%) | 18696 (76.7%) |
| opf-ru-v2+ov100 | 19643 (80.6%) | 19636 (80.6%) | 19636 (80.6%) | 19636 (80.6%) |
| opf-ru-v2 | 19675 (80.7%) | 19672 (80.7%) | 19672 (80.7%) | 19672 (80.7%) |
| gliner2-hivetrace-uni-ru | 19827 (81.4%) | 14381 (59.0%) | 10664 (43.8%) | 5990 (24.6%) |
| opf-kz-ru | 21493 (88.2%) | 21480 (88.1%) | 21480 (88.1%) | 21480 (88.1%) |
| openai-base | 22006 (90.3%) | 22005 (90.3%) | 22005 (90.3%) | 22005 (90.3%) |
| gravitee-small | 23450 (96.2%) | 23353 (95.8%) | 23353 (95.8%) | 23353 (95.8%) |

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.