# ameau01 - en / pii (1452 rows, 2242 spans, 702 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 0.074 (92.6% 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+cpu-int8 | **105** | 4.7% [3.8%, 5.7%] | 94.5% | 0.650 [0.632, 0.668] | 0.497 | 0.937 | 0.594 | 0.706 | 0.260 | 284/702 | 7854 | 0 | 109 | 596M |  |
| gliner2-hivetrace-omni | **120** | 5.4% [4.3%, 6.5%] | 94.6% | 0.757 [0.738, 0.774] | 0.625 | 0.959 | 0.764 | 0.778 | 0.654 | 351/702 | 7296 | 0 | 6 | 307M |  |
| gliner2-fastino | **160** | 7.1% [5.9%, 8.4%] | 92.8% | 0.765 [0.747, 0.783] | 0.661 | 0.909 | 0.783 | 0.795 | 0.662 | 241/702 | 4923 | 0 | 6 | 307M |  |
| nuner-zero | **347** | 15.5% [13.9%, 17.0%] | 84.3% | 0.669 [0.648, 0.689] | 0.568 | 0.811 | 0.641 | 0.727 | 0.603 | 332/702 | 7518 | 0 | 9 | 449M |  |
| gliner-stream-pii | **353** | 15.7% [14.3%, 17.2%] | 82.2% | 0.700 [0.680, 0.717] | 0.630 | 0.788 | 0.643 | 0.694 | 0.512 | 536/702 | 7561 | 0 | 36 | 677M |  |
| gliner-nvidia | **365** | 16.3% [14.7%, 17.8%] | 82.9% | 0.686 [0.666, 0.706] | 0.625 | 0.760 | 0.746 | 0.776 | 0.563 | 251/702 | 5814 | 0 | 9 | 445M |  |
| pplx | **393** | 17.5% [15.9%, 19.2%] | 81.6% | 0.688 [0.668, 0.706] | 0.612 | 0.785 | 0.701 | 0.750 | 0.311 | 173/702 | 4242 | 0 | 52 | 596M |  |
| gliner-urchade | **492** | 21.9% [20.0%, 23.9%] | 77.2% | 0.682 [0.661, 0.703] | 0.624 | 0.752 | 0.656 | 0.728 | 0.581 | 288/702 | 5785 | 0 | 5 | 289M |  |
| mmbert32k | **498** | 22.2% [20.5%, 24.0%] | 52.1% | 0.597 [0.579, 0.613] | 0.536 | 0.673 | 0.318 | 0.560 | 0.253 | 315/702 | 5162 | 0 | 4 | 308M |  |
| gliner2-large | **533** | 23.8% [21.8%, 25.8%] | 76.1% | 0.630 [0.607, 0.652] | 0.557 | 0.724 | 0.654 | 0.680 | 0.544 | 341/702 | 6789 | 0 | 13 | 486M |  |
| gliner-pii-base | **544** | 24.3% [22.3%, 26.2%] | 75.7% | 0.700 [0.682, 0.718] | 0.650 | 0.758 | 0.678 | 0.720 | 0.639 | 303/702 | 5334 | 0 | 4 | 166M |  |
| gliner-pii-edge | **615** | 27.4% [25.6%, 29.4%] | 72.3% | 0.677 [0.660, 0.693] | 0.679 | 0.675 | 0.648 | 0.707 | 0.361 | 328/702 | 4279 | 0 | 40 | 45M |  |
| gliner25-fastino | **616** | 27.5% [25.4%, 29.7%] | 72.5% | 0.669 [0.645, 0.690] | 0.648 | 0.691 | 0.694 | 0.713 | 0.509 | 122/702 | 3093 | 0 | 3 | 287M |  |
| openmed-multilingual | **634** | 28.3% [26.5%, 30.2%] | 52.6% | 0.570 [0.551, 0.589] | 0.538 | 0.606 | 0.392 | 0.585 | 0.454 | 347/702 | 4920 | 0 | 28 | 1.4B |  |
| kalyan-ettin | **711** | 31.7% [29.9%, 33.7%] | 45.7% | 0.573 [0.556, 0.591] | 0.621 | 0.532 | 0.398 | 0.645 | 0.538 | 219/702 | 2965 | 0 | 2 | 68M |  |
| gliner-multi-v21 | **793** | 35.4% [33.5%, 37.2%] | 62.0% | 0.611 [0.591, 0.630] | 0.652 | 0.574 | 0.520 | 0.660 | 0.424 | 177/702 | 2828 | 0 | 4 | 289M |  |
| openmed-nemotron | **811** | 36.2% [34.3%, 38.1%] | 43.0% | 0.542 [0.524, 0.560] | 0.556 | 0.529 | 0.359 | 0.599 | 0.520 | 231/702 | 3594 | 0 | 29 | 1.4B |  |
| nym-base | **828** | 36.9% [35.2%, 38.9%] | 45.0% | 0.572 [0.553, 0.590] | 0.657 | 0.507 | 0.415 | 0.640 | 0.317 | 135/702 | 1995 | 0 | 3 | 308M |  |
| opf-ru | **893** | 39.8% [38.1%, 41.7%] | 50.2% | 0.518 [0.500, 0.535] | 0.560 | 0.483 | 0.456 | 0.576 | 0.466 | 196/702 | 3661 | 0 | 44 | 1.4B |  |
| apararti | **898** | 40.1% [38.2%, 41.9%] | 57.8% | 0.517 [0.496, 0.538] | 0.495 | 0.541 | 0.489 | 0.563 | 0.333 | 162/702 | 4454 | 0 | 66 | 1.4B |  |
| pii-shield-onnx | **923** | 41.2% [39.1%, 43.1%] | 46.8% | 0.550 [0.530, 0.569] | 0.465 | 0.671 | 0.355 | 0.526 | 0.145 | 291/702 | 7823 | 0 | 1264 | - |  |
| nym-base+cpu-int8 | **938** | 41.8% [40.0%, 43.9%] | 39.6% | 0.552 [0.533, 0.570] | 0.656 | 0.476 | 0.372 | 0.608 | 0.286 | 126/702 | 1791 | 0 | 33 | 308M |  |
| nym-small | **994** | 44.3% [42.4%, 46.4%] | 38.4% | 0.538 [0.517, 0.558] | 0.645 | 0.461 | 0.370 | 0.590 | 0.296 | 142/702 | 2077 | 0 | 710 | - |  |
| openai-base | **1008** | 45.0% [43.0%, 47.1%] | 53.7% | 0.528 [0.506, 0.549] | 0.574 | 0.489 | 0.517 | 0.589 | 0.369 | 94/702 | 2902 | 0 | 72 | 1.4B |  |
| openai-base-onnx | **1012** | 45.1% [43.1%, 47.3%] | 53.3% | 0.527 [0.506, 0.548] | 0.574 | 0.488 | 0.516 | 0.590 | 0.369 | 93/702 | 2908 | 0 | 154 | - |  |
| bardsai-eu | **1013** | 45.2% [42.8%, 47.7%] | 44.7% | 0.518 [0.497, 0.539] | 0.551 | 0.489 | 0.411 | 0.549 | 0.129 | 224/702 | 3677 | 0 | 657 | - |  |
| mmbert32k+cpu-int8 | **1033** | 46.1% [43.8%, 48.2%] | 19.4% | 0.415 [0.396, 0.432] | 0.483 | 0.363 | 0.132 | 0.410 | 0.150 | 335/702 | 3834 | 0 | 41 | 308M |  |
| opf-kz-ru | **1036** | 46.2% [44.3%, 48.2%] | 50.8% | 0.485 [0.462, 0.507] | 0.486 | 0.485 | 0.448 | 0.526 | 0.338 | 160/702 | 4547 | 0 | 64 | 1.4B |  |
| stanza-ru | **1089** | 48.6% [46.5%, 50.7%] | 37.7% | 0.269 [0.257, 0.282] | 0.184 | 0.503 | 0.115 | 0.345 | 0.020 | 684/702 | 18739 | 0 | 34 | - |  |
| traciora | **1130** | 50.4% [48.4%, 52.6%] | 44.8% | 0.550 [0.529, 0.569] | 0.739 | 0.438 | 0.492 | 0.591 | 0.401 | 96/702 | 1448 | 0 | 53 | 1.4B |  |
| gravitee-small | **1172** | 52.3% [50.3%, 54.6%] | 46.7% | 0.520 [0.498, 0.541] | 0.619 | 0.449 | 0.496 | 0.534 | 0.467 | 130/702 | 2036 | 0 | 2 | 29M |  |
| gravitee-small+cpu-int8 | **1173** | 52.3% [50.3%, 54.6%] | 46.7% | 0.523 [0.501, 0.543] | 0.627 | 0.448 | 0.501 | 0.536 | 0.465 | 127/702 | 1962 | 0 | 6 | 29M |  |
| opf-ru-v2 | **1220** | 54.4% [52.3%, 56.6%] | 42.5% | 0.552 [0.530, 0.571] | 0.783 | 0.426 | 0.496 | 0.561 | 0.446 | 90/702 | 1129 | 0 | 266 | 1.4B |  |
| ru-pii-ner | **1239** | 55.3% [52.7%, 58.0%] | 44.5% | 0.571 [0.545, 0.595] | 0.767 | 0.454 | 0.537 | 0.560 | 0.252 | 102/702 | 1229 | 0 | 56 | 358M |  |
| ner-ru-yqelz | **1285** | 57.3% [55.5%, 59.2%] | 38.8% | 0.302 [0.285, 0.317] | 0.249 | 0.383 | 0.277 | 0.317 | 0.125 | 686/702 | 14496 | 0 | 4 | 559M |  |
| gliner2-hivetrace-uni | **1311** | 58.5% [56.8%, 60.2%] | 41.4% | 0.534 [0.514, 0.553] | 0.852 | 0.389 | 0.552 | 0.557 | 0.395 | 65/702 | 919 | 0 | 9 | 147M |  |
| gliner2-vladlinv | **1364** | 60.8% [58.7%, 63.0%] | 39.1% | 0.527 [0.503, 0.550] | 0.952 | 0.364 | 0.548 | 0.549 | 0.468 | 7/702 | 52 | 0 | 5 | 287M |  |
| kalyan-ettin+cpu-int8 | **1395** | 62.2% [60.3%, 64.4%] | 17.6% | 0.377 [0.356, 0.396] | 0.584 | 0.278 | 0.169 | 0.394 | 0.347 | 247/702 | 1541 | 0 | 67 | 68M |  |
| davlan-xlmr | **1448** | 64.6% [62.5%, 66.8%] | 34.3% | 0.358 [0.336, 0.379] | 0.479 | 0.285 | 0.377 | 0.397 | 0.204 | 254/702 | 3139 | 0 | 3 | 277M |  |
| davlan-mbert | **1485** | 66.2% [64.2%, 68.5%] | 29.2% | 0.333 [0.311, 0.352] | 0.410 | 0.280 | 0.314 | 0.372 | 0.190 | 289/702 | 3913 | 0 | 3 | 177M |  |
| ru-legal-ner+cpu-int8 | **1523** | 67.9% [65.9%, 69.9%] | 17.1% | 0.331 [0.310, 0.351] | 0.522 | 0.243 | 0.182 | 0.382 | 0.153 | 184/702 | 2143 | 0 | 4 | 29M |  |
| ru-legal-ner | **1552** | 69.2% [67.2%, 71.2%] | 16.4% | 0.323 [0.302, 0.343] | 0.522 | 0.234 | 0.175 | 0.373 | 0.152 | 180/702 | 2106 | 0 | 2 | 29M |  |
| davlan-mbert+cpu-int8 | **1558** | 69.5% [67.5%, 71.7%] | 26.4% | 0.323 [0.300, 0.343] | 0.446 | 0.253 | 0.310 | 0.365 | 0.203 | 245/702 | 3231 | 0 | 31 | 177M |  |
| ner-ru-gherman | **1574** | 70.2% [68.1%, 72.4%] | 10.5% | 0.378 [0.354, 0.401] | 0.925 | 0.237 | 0.138 | 0.444 | 0.077 | 16/702 | 82 | 0 | 3 | 177M |  |
| ner-ru-gherman-onnx | **1585** | 70.7% [68.7%, 72.8%] | 10.3% | 0.375 [0.351, 0.397] | 0.927 | 0.235 | 0.136 | 0.439 | 0.079 | 14/702 | 71 | 0 | 173 | - |  |
| ner-ru-gherman+cpu-int8 | **1617** | 72.1% [70.2%, 74.2%] | 9.2% | 0.361 [0.338, 0.383] | 0.927 | 0.224 | 0.124 | 0.423 | 0.080 | 12/702 | 55 | 0 | 25 | 177M |  |
| davlan-xlmr+cpu-int8 | **1772** | 79.0% [77.3%, 80.9%] | 18.7% | 0.275 [0.252, 0.296] | 0.631 | 0.176 | 0.270 | 0.306 | 0.202 | 113/702 | 1145 | 0 | 28 | 277M |  |
| rules-ru | **1876** | 83.7% [82.2%, 85.1%] | 16.3% | 0.247 [0.226, 0.266] | 0.362 | 0.188 | 0.238 | 0.239 | 0.247 | 126/702 | 4138 | 0 | 0 | - |  |
| fef2-secret-ru | **1956** | 87.2% [85.7%, 88.7%] | 11.0% | 0.233 [0.207, 0.260] | 0.563 | 0.147 | 0.143 | 0.204 | 0.036 | 87/702 | 1361 | 0 | 4 | 177M |  |
| fef2-secret-ru+cpu-int8 | **1959** | 87.4% [85.7%, 88.8%] | 10.0% | 0.238 [0.212, 0.265] | 0.678 | 0.144 | 0.137 | 0.209 | 0.056 | 72/702 | 984 | 0 | 73 | 177M |  |
| gliner-pii-edge+cpu-int8 | **2098** | 93.6% [92.4%, 94.7%] | 5.9% | 0.088 [0.073, 0.104] | 0.625 | 0.047 | 0.102 | 0.118 | 0.042 | 42/702 | 423 | 0 | 106 | 45M |  |
| spacy-ru-lg | **2119** | 94.5% [93.4%, 95.7%] | 5.5% | 0.062 [0.049, 0.075] | 0.090 | 0.048 | 0.029 | 0.095 | 0.001 | 70/702 | 4206 | 0 | 4 | - |  |
| natasha | **2153** | 96.0% [95.2%, 96.9%] | 3.9% | 0.071 [0.056, 0.086] | 0.330 | 0.040 | 0.065 | 0.070 | 0.038 | 49/702 | 497 | 0 | 2 | - |  |
| gliner-pii-base+cpu-int8 | **2227** | 99.3% [99.0%, 99.6%] | 0.7% | 0.012 [0.007, 0.019] | 0.865 | 0.006 | 0.013 | 0.013 | 0.012 | 1/702 | 27 | 0 | 74 | 166M |  |
| gliner-multi-v21+cpu-int8 | **2242** | 100.0% [100.0%, 100.0%] | 0.0% | 0.000 [0.000, 0.000] | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0/702 | 0 | 0 | 85 | 289M |  |
| gliner-nvidia+cpu-int8 | **2242** | 100.0% [100.0%, 100.0%] | 0.0% | 0.000 [0.000, 0.000] | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0/702 | 0 | 0 | 126 | 445M |  |
| gliner-urchade+cpu-int8 | **2242** | 100.0% [100.0%, 100.0%] | 0.0% | 0.000 [0.000, 0.000] | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0/702 | 0 | 0 | 72 | 289M |  |
| spacy-alrosait | **2242** | 100.0% [100.0%, 100.0%] | 0.0% | 0.000 [0.000, 0.000] | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0/702 | 0 | 0 | 3 | - |  |

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+cpu-int8 ≈ gliner2-hivetrace-omni; nuner-zero ≈ gliner-stream-pii; gliner-stream-pii ≈ gliner-nvidia; gliner-nvidia ≈ pplx; gliner-urchade ≈ mmbert32k; mmbert32k ≈ gliner2-large; gliner2-large ≈ gliner-pii-base; gliner-pii-edge ≈ gliner25-fastino; gliner25-fastino ≈ openmed-multilingual; gliner-multi-v21 ≈ openmed-nemotron; openmed-nemotron ≈ nym-base; opf-ru ≈ apararti; apararti ≈ pii-shield-onnx; pii-shield-onnx ≈ nym-base+cpu-int8; nym-small ≈ openai-base; openai-base ≈ openai-base-onnx; openai-base-onnx ≈ bardsai-eu; bardsai-eu ≈ mmbert32k+cpu-int8; mmbert32k+cpu-int8 ≈ opf-kz-ru; opf-kz-ru ≈ stanza-ru; stanza-ru ≈ traciora; traciora ≈ gravitee-small; gravitee-small ≈ gravitee-small+cpu-int8; gravitee-small+cpu-int8 ≈ opf-ru-v2; opf-ru-v2 ≈ ru-pii-ner; ru-pii-ner ≈ ner-ru-yqelz; ner-ru-yqelz ≈ gliner2-hivetrace-uni; gliner2-vladlinv ≈ kalyan-ettin+cpu-int8; kalyan-ettin+cpu-int8 ≈ davlan-xlmr; davlan-xlmr ≈ davlan-mbert; davlan-mbert ≈ ru-legal-ner+cpu-int8; ru-legal-ner ≈ davlan-mbert+cpu-int8; davlan-mbert+cpu-int8 ≈ ner-ru-gherman; fef2-secret-ru ≈ fef2-secret-ru+cpu-int8; gliner-pii-edge+cpu-int8 ≈ spacy-ru-lg; gliner-multi-v21+cpu-int8 ≈ gliner-nvidia+cpu-int8; gliner-nvidia+cpu-int8 ≈ gliner-urchade+cpu-int8; gliner-urchade+cpu-int8 ≈ spacy-alrosait

## Missed by group

| model | PERSON | ADDRESS | CONTACT | NET | ACCOUNT |
|---|---|---|---|---|---|
| spans in gold | 382 | 184 | 141 | 777 | 758 |
| pplx+cpu-int8 | 1 (0.3%) | 37 (20.1%) | 0 (0.0%) | 52 (6.7%) | 15 (2.0%) |
| gliner2-hivetrace-omni | 0 (0.0%) | 14 (7.6%) | 0 (0.0%) | 20 (2.6%) | 86 (11.3%) |
| gliner2-fastino | 0 (0.0%) | 20 (10.9%) | 1 (0.7%) | 114 (14.7%) | 25 (3.3%) |
| nuner-zero | 2 (0.5%) | 44 (23.9%) | 0 (0.0%) | 218 (28.1%) | 83 (10.9%) |
| gliner-stream-pii | 12 (3.1%) | 66 (35.9%) | 12 (8.5%) | 233 (30.0%) | 30 (4.0%) |
| gliner-nvidia | 2 (0.5%) | 3 (1.6%) | 4 (2.8%) | 321 (41.3%) | 35 (4.6%) |
| pplx | 1 (0.3%) | 137 (74.5%) | 3 (2.1%) | 225 (29.0%) | 27 (3.6%) |
| gliner-urchade | 1 (0.3%) | 69 (37.5%) | 10 (7.1%) | 261 (33.6%) | 151 (19.9%) |
| mmbert32k | 0 (0.0%) | 151 (82.1%) | 1 (0.7%) | 152 (19.6%) | 194 (25.6%) |
| gliner2-large | 2 (0.5%) | 39 (21.2%) | 0 (0.0%) | 319 (41.1%) | 173 (22.8%) |
| gliner-pii-base | 266 (69.6%) | 60 (32.6%) | 5 (3.5%) | 119 (15.3%) | 94 (12.4%) |
| gliner-pii-edge | 237 (62.0%) | 8 (4.3%) | 7 (5.0%) | 312 (40.2%) | 51 (6.7%) |
| gliner25-fastino | 0 (0.0%) | 66 (35.9%) | 4 (2.8%) | 324 (41.7%) | 222 (29.3%) |
| openmed-multilingual | 1 (0.3%) | 43 (23.4%) | 1 (0.7%) | 424 (54.6%) | 165 (21.8%) |
| kalyan-ettin | 7 (1.8%) | 48 (26.1%) | 6 (4.3%) | 499 (64.2%) | 151 (19.9%) |
| gliner-multi-v21 | 13 (3.4%) | 38 (20.7%) | 61 (43.3%) | 395 (50.8%) | 286 (37.7%) |
| openmed-nemotron | 1 (0.3%) | 71 (38.6%) | 1 (0.7%) | 519 (66.8%) | 219 (28.9%) |
| nym-base | 0 (0.0%) | 45 (24.5%) | 0 (0.0%) | 694 (89.3%) | 89 (11.7%) |
| opf-ru | 8 (2.1%) | 178 (96.7%) | 0 (0.0%) | 534 (68.7%) | 173 (22.8%) |
| apararti | 12 (3.1%) | 174 (94.6%) | 0 (0.0%) | 467 (60.1%) | 245 (32.3%) |
| pii-shield-onnx | 120 (31.4%) | 151 (82.1%) | 4 (2.8%) | 71 (9.1%) | 577 (76.1%) |
| nym-base+cpu-int8 | 1 (0.3%) | 80 (43.5%) | 0 (0.0%) | 679 (87.4%) | 178 (23.5%) |
| nym-small | 1 (0.3%) | 30 (16.3%) | 0 (0.0%) | 753 (96.9%) | 210 (27.7%) |
| openai-base | 14 (3.7%) | 181 (98.4%) | 4 (2.8%) | 543 (69.9%) | 266 (35.1%) |
| openai-base-onnx | 14 (3.7%) | 181 (98.4%) | 4 (2.8%) | 542 (69.8%) | 271 (35.8%) |
| bardsai-eu | 2 (0.5%) | 10 (5.4%) | 5 (3.5%) | 526 (67.7%) | 470 (62.0%) |
| mmbert32k+cpu-int8 | 111 (29.1%) | 155 (84.2%) | 22 (15.6%) | 290 (37.3%) | 455 (60.0%) |
| opf-kz-ru | 26 (6.8%) | 178 (96.7%) | 0 (0.0%) | 505 (65.0%) | 327 (43.1%) |
| stanza-ru | 4 (1.0%) | 34 (18.5%) | 134 (95.0%) | 327 (42.1%) | 590 (77.8%) |
| traciora | 32 (8.4%) | 182 (98.9%) | 4 (2.8%) | 559 (71.9%) | 353 (46.6%) |
| gravitee-small | 1 (0.3%) | 25 (13.6%) | 4 (2.8%) | 543 (69.9%) | 599 (79.0%) |
| gravitee-small+cpu-int8 | 0 (0.0%) | 37 (20.1%) | 4 (2.8%) | 544 (70.0%) | 588 (77.6%) |
| opf-ru-v2 | 23 (6.0%) | 183 (99.5%) | 4 (2.8%) | 520 (66.9%) | 490 (64.6%) |
| ru-pii-ner | 15 (3.9%) | 177 (96.2%) | 20 (14.2%) | 564 (72.6%) | 463 (61.1%) |
| ner-ru-yqelz | 8 (2.1%) | 7 (3.8%) | 125 (88.7%) | 596 (76.7%) | 549 (72.4%) |
| gliner2-hivetrace-uni | 7 (1.8%) | 182 (98.9%) | 6 (4.3%) | 742 (95.5%) | 374 (49.3%) |
| gliner2-vladlinv | 16 (4.2%) | 182 (98.9%) | 17 (12.1%) | 744 (95.8%) | 405 (53.4%) |
| kalyan-ettin+cpu-int8 | 56 (14.7%) | 144 (78.3%) | 12 (8.5%) | 712 (91.6%) | 471 (62.1%) |
| davlan-xlmr | 3 (0.8%) | 6 (3.3%) | 126 (89.4%) | 768 (98.8%) | 545 (71.9%) |
| davlan-mbert | 1 (0.3%) | 5 (2.7%) | 116 (82.3%) | 693 (89.2%) | 670 (88.4%) |
| ru-legal-ner+cpu-int8 | 240 (62.8%) | 172 (93.5%) | 5 (3.5%) | 646 (83.1%) | 460 (60.7%) |
| ru-legal-ner | 262 (68.6%) | 171 (92.9%) | 3 (2.1%) | 649 (83.5%) | 467 (61.6%) |
| davlan-mbert+cpu-int8 | 2 (0.5%) | 8 (4.3%) | 129 (91.5%) | 715 (92.0%) | 704 (92.9%) |
| ner-ru-gherman | 0 (0.0%) | 8 (4.3%) | 113 (80.1%) | 775 (99.7%) | 678 (89.4%) |
| ner-ru-gherman-onnx | 0 (0.0%) | 8 (4.3%) | 116 (82.3%) | 776 (99.9%) | 685 (90.4%) |
| ner-ru-gherman+cpu-int8 | 0 (0.0%) | 11 (6.0%) | 122 (86.5%) | 777 (100.0%) | 707 (93.3%) |
| davlan-xlmr+cpu-int8 | 67 (17.5%) | 41 (22.3%) | 141 (100.0%) | 776 (99.9%) | 747 (98.5%) |
| rules-ru | 382 (100.0%) | 184 (100.0%) | 21 (14.9%) | 572 (73.6%) | 717 (94.6%) |
| fef2-secret-ru | 315 (82.5%) | 180 (97.8%) | 32 (22.7%) | 764 (98.3%) | 665 (87.7%) |
| fef2-secret-ru+cpu-int8 | 293 (76.7%) | 181 (98.4%) | 34 (24.1%) | 767 (98.7%) | 684 (90.2%) |
| gliner-pii-edge+cpu-int8 | 376 (98.4%) | 176 (95.7%) | 120 (85.1%) | 769 (99.0%) | 657 (86.7%) |
| spacy-ru-lg | 343 (89.8%) | 176 (95.7%) | 139 (98.6%) | 755 (97.2%) | 706 (93.1%) |
| natasha | 322 (84.3%) | 169 (91.8%) | 141 (100.0%) | 764 (98.3%) | 757 (99.9%) |
| gliner-pii-base+cpu-int8 | 379 (99.2%) | 184 (100.0%) | 136 (96.5%) | 777 (100.0%) | 751 (99.1%) |
| gliner-multi-v21+cpu-int8 | 382 (100.0%) | 184 (100.0%) | 141 (100.0%) | 777 (100.0%) | 758 (100.0%) |
| gliner-nvidia+cpu-int8 | 382 (100.0%) | 184 (100.0%) | 141 (100.0%) | 777 (100.0%) | 758 (100.0%) |
| gliner-urchade+cpu-int8 | 382 (100.0%) | 184 (100.0%) | 141 (100.0%) | 777 (100.0%) | 758 (100.0%) |
| spacy-alrosait | 382 (100.0%) | 184 (100.0%) | 141 (100.0%) | 777 (100.0%) | 758 (100.0%) |

## Char recall by gold type

| type | group | pplx+cpu-int8 | gliner2-hivetrace-omni | gliner2-fastino | nuner-zero | gliner-stream-pii | gliner-nvidia | pplx | gliner-urchade | mmbert32k | gliner2-large | gliner-pii-base | gliner-pii-edge | gliner25-fastino | openmed-multilingual | kalyan-ettin | gliner-multi-v21 | openmed-nemotron | nym-base | opf-ru | apararti | pii-shield-onnx | nym-base+cpu-int8 | nym-small | openai-base | openai-base-onnx | bardsai-eu | mmbert32k+cpu-int8 | opf-kz-ru | stanza-ru | traciora | gravitee-small | gravitee-small+cpu-int8 | opf-ru-v2 | ru-pii-ner | ner-ru-yqelz | gliner2-hivetrace-uni | gliner2-vladlinv | kalyan-ettin+cpu-int8 | davlan-xlmr | davlan-mbert | ru-legal-ner+cpu-int8 | ru-legal-ner | davlan-mbert+cpu-int8 | ner-ru-gherman | ner-ru-gherman-onnx | ner-ru-gherman+cpu-int8 | davlan-xlmr+cpu-int8 | rules-ru | fef2-secret-ru | fef2-secret-ru+cpu-int8 | gliner-pii-edge+cpu-int8 | spacy-ru-lg | natasha | gliner-pii-base+cpu-int8 | gliner-multi-v21+cpu-int8 | gliner-nvidia+cpu-int8 | gliner-urchade+cpu-int8 | spacy-alrosait |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| email | CONTACT | 0.998 | 1.000 | 1.000 | 1.000 | 0.834 | 0.971 | 0.990 | 0.854 | 0.908 | 1.000 | 0.963 | 0.955 | 1.000 | 1.000 | 0.990 | 0.245 | 0.996 | 1.000 | 0.994 | 1.000 | 0.990 | 1.000 | 1.000 | 0.975 | 0.975 | 0.773 | 0.422 | 1.000 | 0.052 | 1.000 | 0.998 | 0.998 | 0.956 | 0.872 | 0.042 | 1.000 | 0.929 | 0.862 | 0.051 | 0.132 | 0.854 | 0.862 | 0.058 | 0.098 | 0.092 | 0.065 | 0.000 | 1.000 | 0.926 | 0.880 | 0.078 | 0.018 | 0.000 | 0.008 | 0.000 | 0.000 | 0.000 | 0.000 |
| emp_id | ACCOUNT | 1.000 | 1.000 | 0.987 | 0.937 | 0.987 | 0.984 | 0.981 | 0.735 | 0.653 | 0.961 | 0.992 | 1.000 | 0.384 | 1.000 | 0.654 | 0.432 | 0.462 | 1.000 | 0.161 | 0.884 | 0.431 | 0.995 | 0.995 | 0.565 | 0.571 | 0.618 | 0.554 | 0.848 | 0.298 | 0.270 | 0.000 | 0.000 | 0.262 | 0.873 | 0.012 | 0.168 | 0.431 | 0.220 | 0.000 | 0.000 | 0.357 | 0.363 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0.450 | 0.333 | 0.157 | 0.023 | 0.000 | 0.053 | 0.000 | 0.000 | 0.000 | 0.000 |
| hostname | NET | 0.878 | 0.965 | 0.794 | 0.605 | 0.601 | 0.434 | 0.567 | 0.575 | 0.486 | 0.437 | 0.797 | 0.533 | 0.401 | 0.139 | 0.026 | 0.344 | 0.022 | 0.000 | 0.028 | 0.112 | 0.945 | 0.000 | 0.002 | 0.036 | 0.036 | 0.205 | 0.227 | 0.049 | 0.694 | 0.028 | 0.001 | 0.001 | 0.022 | 0.192 | 0.300 | 0.035 | 0.004 | 0.008 | 0.014 | 0.066 | 0.008 | 0.009 | 0.042 | 0.001 | 0.001 | 0.000 | 0.001 | 0.000 | 0.015 | 0.008 | 0.005 | 0.037 | 0.024 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 |
| ip | NET | 1.000 | 1.000 | 1.000 | 1.000 | 0.947 | 0.941 | 0.992 | 0.881 | 0.826 | 0.946 | 0.976 | 0.770 | 0.992 | 0.997 | 0.916 | 0.840 | 0.971 | 0.296 | 0.616 | 0.986 | 0.640 | 0.353 | 0.087 | 0.880 | 0.881 | 0.408 | 0.483 | 0.975 | 0.034 | 0.764 | 0.992 | 0.993 | 0.988 | 0.470 | 0.005 | 0.070 | 0.131 | 0.084 | 0.000 | 0.000 | 0.242 | 0.228 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0.878 | 0.000 | 0.000 | 0.016 | 0.008 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 |
| location | ADDRESS | 0.794 | 0.924 | 0.892 | 0.766 | 0.646 | 0.982 | 0.257 | 0.632 | 0.181 | 0.794 | 0.689 | 0.961 | 0.645 | 0.786 | 0.731 | 0.801 | 0.621 | 0.758 | 0.031 | 0.050 | 0.183 | 0.562 | 0.841 | 0.014 | 0.014 | 0.945 | 0.159 | 0.031 | 0.829 | 0.012 | 0.882 | 0.818 | 0.006 | 0.036 | 0.965 | 0.009 | 0.012 | 0.220 | 0.969 | 0.975 | 0.056 | 0.064 | 0.961 | 0.965 | 0.963 | 0.947 | 0.779 | 0.000 | 0.014 | 0.014 | 0.049 | 0.043 | 0.080 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 |
| person | PERSON | 0.997 | 0.999 | 1.000 | 0.991 | 0.945 | 0.973 | 0.995 | 0.997 | 0.986 | 0.994 | 0.314 | 0.369 | 0.999 | 0.914 | 0.828 | 0.968 | 0.912 | 0.915 | 0.945 | 0.957 | 0.558 | 0.904 | 0.915 | 0.961 | 0.960 | 0.993 | 0.617 | 0.906 | 0.974 | 0.896 | 0.985 | 0.985 | 0.928 | 0.959 | 0.952 | 0.982 | 0.959 | 0.599 | 0.993 | 0.997 | 0.294 | 0.257 | 0.985 | 0.918 | 0.917 | 0.916 | 0.776 | 0.000 | 0.165 | 0.224 | 0.012 | 0.096 | 0.156 | 0.009 | 0.000 | 0.000 | 0.000 | 0.000 |
| phone | CONTACT | 1.000 | 1.000 | 0.986 | 1.000 | 0.675 | 0.931 | 0.972 | 0.924 | 0.993 | 1.000 | 0.979 | 0.924 | 0.910 | 0.986 | 0.824 | 0.927 | 0.986 | 1.000 | 1.000 | 1.000 | 0.896 | 1.000 | 1.000 | 0.958 | 0.958 | 0.896 | 0.910 | 1.000 | 0.014 | 0.810 | 0.927 | 0.900 | 0.990 | 0.827 | 0.069 | 0.855 | 0.730 | 0.543 | 0.000 | 0.000 | 0.927 | 0.927 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0.211 | 0.076 | 0.062 | 0.398 | 0.000 | 0.000 | 0.187 | 0.000 | 0.000 | 0.000 | 0.000 |
| username | ACCOUNT | 0.968 | 0.863 | 0.954 | 0.859 | 0.939 | 0.929 | 0.941 | 0.806 | 0.705 | 0.730 | 0.835 | 0.910 | 0.768 | 0.733 | 0.794 | 0.659 | 0.738 | 0.858 | 0.872 | 0.628 | 0.217 | 0.718 | 0.673 | 0.650 | 0.643 | 0.315 | 0.298 | 0.508 | 0.089 | 0.560 | 0.253 | 0.272 | 0.361 | 0.313 | 0.320 | 0.581 | 0.474 | 0.370 | 0.338 | 0.142 | 0.376 | 0.360 | 0.090 | 0.135 | 0.124 | 0.085 | 0.020 | 0.075 | 0.062 | 0.058 | 0.129 | 0.072 | 0.001 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 |

## char F1 by domain

| domain | n | pplx+cpu-int8 | gliner2-hivetrace-omni | gliner2-fastino | nuner-zero | gliner-stream-pii | gliner-nvidia | pplx | gliner-urchade | mmbert32k | gliner2-large | gliner-pii-base | gliner-pii-edge | gliner25-fastino | openmed-multilingual | kalyan-ettin | gliner-multi-v21 | openmed-nemotron | nym-base | opf-ru | apararti | pii-shield-onnx | nym-base+cpu-int8 | nym-small | openai-base | openai-base-onnx | bardsai-eu | mmbert32k+cpu-int8 | opf-kz-ru | stanza-ru | traciora | gravitee-small | gravitee-small+cpu-int8 | opf-ru-v2 | ru-pii-ner | ner-ru-yqelz | gliner2-hivetrace-uni | gliner2-vladlinv | kalyan-ettin+cpu-int8 | davlan-xlmr | davlan-mbert | ru-legal-ner+cpu-int8 | ru-legal-ner | davlan-mbert+cpu-int8 | ner-ru-gherman | ner-ru-gherman-onnx | ner-ru-gherman+cpu-int8 | davlan-xlmr+cpu-int8 | rules-ru | fef2-secret-ru | fef2-secret-ru+cpu-int8 | gliner-pii-edge+cpu-int8 | spacy-ru-lg | natasha | gliner-pii-base+cpu-int8 | gliner-multi-v21+cpu-int8 | gliner-nvidia+cpu-int8 | gliner-urchade+cpu-int8 | spacy-alrosait |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| correspondence | 148 | 0.712 | 0.822 | 0.802 | 0.717 | 0.741 | 0.759 | 0.777 | 0.740 | 0.646 | 0.655 | 0.737 | 0.724 | 0.693 | 0.630 | 0.619 | 0.590 | 0.593 | 0.611 | 0.566 | 0.574 | 0.618 | 0.593 | 0.590 | 0.585 | 0.586 | 0.595 | 0.455 | 0.562 | 0.318 | 0.589 | 0.607 | 0.608 | 0.605 | 0.644 | 0.376 | 0.551 | 0.559 | 0.429 | 0.398 | 0.380 | 0.319 | 0.321 | 0.376 | 0.413 | 0.412 | 0.399 | 0.297 | 0.301 | 0.247 | 0.223 | 0.068 | 0.087 | 0.107 | 0.012 | 0.000 | 0.000 | 0.000 | 0.000 |
| diagnostics | 745 | 0.600 | 0.712 | 0.726 | 0.623 | 0.683 | 0.622 | 0.619 | 0.646 | 0.563 | 0.601 | 0.694 | 0.652 | 0.616 | 0.495 | 0.504 | 0.620 | 0.474 | 0.482 | 0.467 | 0.460 | 0.508 | 0.471 | 0.434 | 0.468 | 0.466 | 0.436 | 0.354 | 0.416 | 0.244 | 0.495 | 0.418 | 0.428 | 0.500 | 0.427 | 0.281 | 0.460 | 0.450 | 0.288 | 0.281 | 0.262 | 0.279 | 0.266 | 0.233 | 0.277 | 0.272 | 0.253 | 0.176 | 0.233 | 0.169 | 0.171 | 0.100 | 0.046 | 0.045 | 0.009 | 0.000 | 0.000 | 0.000 | 0.000 |
| resolution | 321 | 0.662 | 0.756 | 0.769 | 0.683 | 0.709 | 0.676 | 0.688 | 0.694 | 0.618 | 0.621 | 0.683 | 0.690 | 0.661 | 0.604 | 0.604 | 0.597 | 0.573 | 0.621 | 0.554 | 0.518 | 0.573 | 0.581 | 0.582 | 0.546 | 0.547 | 0.542 | 0.465 | 0.484 | 0.300 | 0.571 | 0.546 | 0.544 | 0.539 | 0.615 | 0.345 | 0.604 | 0.562 | 0.429 | 0.421 | 0.384 | 0.409 | 0.403 | 0.389 | 0.416 | 0.413 | 0.400 | 0.319 | 0.223 | 0.268 | 0.315 | 0.094 | 0.085 | 0.049 | 0.013 | 0.000 | 0.000 | 0.000 | 0.000 |
| root_cause | 121 | 0.022 | 0.016 | 0.021 | 0.000 | 0.009 | 0.000 | 0.000 | 0.012 | 0.030 | 0.000 | 0.023 | 0.000 | 0.060 | 0.021 | 0.000 | 0.036 | 0.018 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0.023 | 0.000 | 0.001 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0.121 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0.029 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 |
| ticket | 117 | 0.719 | 0.867 | 0.875 | 0.766 | 0.801 | 0.829 | 0.786 | 0.766 | 0.613 | 0.767 | 0.746 | 0.713 | 0.804 | 0.675 | 0.672 | 0.677 | 0.633 | 0.671 | 0.545 | 0.607 | 0.551 | 0.658 | 0.658 | 0.583 | 0.585 | 0.613 | 0.460 | 0.587 | 0.280 | 0.598 | 0.609 | 0.601 | 0.631 | 0.725 | 0.288 | 0.576 | 0.601 | 0.449 | 0.430 | 0.402 | 0.363 | 0.346 | 0.392 | 0.496 | 0.494 | 0.491 | 0.423 | 0.270 | 0.326 | 0.298 | 0.077 | 0.040 | 0.121 | 0.020 | 0.000 | 0.000 | 0.000 | 0.000 |

## 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 |
|---|---|---|---|---|
| gliner2-hivetrace-omni | 120 (5.4%) | 45 (2.0%) | 31 (1.4%) | 12 (0.5%) |
| gliner2-fastino | 160 (7.1%) | 97 (4.3%) | 60 (2.7%) | 36 (1.6%) |
| nuner-zero | 347 (15.5%) | 187 (8.3%) | 148 (6.6%) | 95 (4.2%) |
| gliner-stream-pii | 353 (15.7%) | 219 (9.8%) | 166 (7.4%) | 100 (4.5%) |
| gliner-nvidia | 365 (16.3%) | 283 (12.6%) | 233 (10.4%) | 183 (8.2%) |
| gliner-urchade | 492 (21.9%) | 308 (13.7%) | 232 (10.3%) | 134 (6.0%) |
| mmbert32k | 498 (22.2%) | 169 (7.5%) | 133 (5.9%) | 131 (5.8%) |
| gliner2-large | 533 (23.8%) | 363 (16.2%) | 260 (11.6%) | 120 (5.4%) |
| gliner-pii-base | 544 (24.3%) | 76 (3.4%) | 11 (0.5%) | 0 (0.0%) |
| gliner-pii-edge | 615 (27.4%) | 86 (3.8%) | 30 (1.3%) | 13 (0.6%) |
| gliner25-fastino | 616 (27.5%) | 541 (24.1%) | 504 (22.5%) | 458 (20.4%) |
| openmed-multilingual | 634 (28.3%) | 565 (25.2%) | 559 (24.9%) | 559 (24.9%) |
| kalyan-ettin | 711 (31.7%) | 606 (27.0%) | 601 (26.8%) | 601 (26.8%) |
| gliner-multi-v21 | 793 (35.4%) | 382 (17.0%) | 200 (8.9%) | 95 (4.2%) |
| openmed-nemotron | 811 (36.2%) | 758 (33.8%) | 755 (33.7%) | 755 (33.7%) |
| nym-base | 828 (36.9%) | 823 (36.7%) | 823 (36.7%) | 823 (36.7%) |
| opf-ru | 893 (39.8%) | 840 (37.5%) | 831 (37.1%) | 831 (37.1%) |
| apararti | 898 (40.1%) | 886 (39.5%) | 886 (39.5%) | 886 (39.5%) |
| pii-shield-onnx | 923 (41.2%) | 833 (37.2%) | 831 (37.1%) | 829 (37.0%) |
| nym-base+cpu-int8 | 938 (41.8%) | 930 (41.5%) | 930 (41.5%) | 930 (41.5%) |
| nym-small | 994 (44.3%) | 991 (44.2%) | 991 (44.2%) | 991 (44.2%) |
| openai-base | 1008 (45.0%) | 1005 (44.8%) | 1005 (44.8%) | 1005 (44.8%) |
| openai-base-onnx | 1012 (45.1%) | 1010 (45.0%) | 1010 (45.0%) | 1010 (45.0%) |
| bardsai-eu | 1013 (45.2%) | 933 (41.6%) | 929 (41.4%) | 929 (41.4%) |
| mmbert32k+cpu-int8 | 1033 (46.1%) | 408 (18.2%) | 301 (13.4%) | 284 (12.7%) |
| opf-kz-ru | 1036 (46.2%) | 1023 (45.6%) | 1023 (45.6%) | 1023 (45.6%) |
| traciora | 1130 (50.4%) | 1118 (49.9%) | 1115 (49.7%) | 1115 (49.7%) |
| gravitee-small | 1172 (52.3%) | 1153 (51.4%) | 1153 (51.4%) | 1153 (51.4%) |
| gravitee-small+cpu-int8 | 1173 (52.3%) | 1134 (50.6%) | 1134 (50.6%) | 1134 (50.6%) |
| opf-ru-v2 | 1220 (54.4%) | 1205 (53.7%) | 1203 (53.7%) | 1203 (53.7%) |
| ner-ru-yqelz | 1285 (57.3%) | 1267 (56.5%) | 1267 (56.5%) | 1267 (56.5%) |
| gliner2-hivetrace-uni | 1311 (58.5%) | 877 (39.1%) | 575 (25.6%) | 255 (11.4%) |
| gliner2-vladlinv | 1364 (60.8%) | 1348 (60.1%) | 1332 (59.4%) | 1321 (58.9%) |
| kalyan-ettin+cpu-int8 | 1395 (62.2%) | 1289 (57.5%) | 1268 (56.6%) | 1266 (56.5%) |
| davlan-xlmr | 1448 (64.6%) | 1420 (63.3%) | 1419 (63.3%) | 1419 (63.3%) |
| davlan-mbert | 1485 (66.2%) | 1484 (66.2%) | 1484 (66.2%) | 1484 (66.2%) |
| ru-legal-ner+cpu-int8 | 1523 (67.9%) | 1119 (49.9%) | 1081 (48.2%) | 1075 (47.9%) |
| ru-legal-ner | 1552 (69.2%) | 1146 (51.1%) | 1106 (49.3%) | 1100 (49.1%) |
| davlan-mbert+cpu-int8 | 1558 (69.5%) | 1557 (69.4%) | 1557 (69.4%) | 1557 (69.4%) |
| ner-ru-gherman | 1574 (70.2%) | 1567 (69.9%) | 1567 (69.9%) | 1567 (69.9%) |
| ner-ru-gherman-onnx | 1585 (70.7%) | 1578 (70.4%) | 1578 (70.4%) | 1578 (70.4%) |
| ner-ru-gherman+cpu-int8 | 1617 (72.1%) | 1612 (71.9%) | 1612 (71.9%) | 1612 (71.9%) |
| davlan-xlmr+cpu-int8 | 1772 (79.0%) | 1712 (76.4%) | 1705 (76.0%) | 1705 (76.0%) |
| fef2-secret-ru | 1956 (87.2%) | 1955 (87.2%) | 1955 (87.2%) | 1955 (87.2%) |
| fef2-secret-ru+cpu-int8 | 1959 (87.4%) | 1956 (87.2%) | 1956 (87.2%) | 1956 (87.2%) |
| gliner-pii-edge+cpu-int8 | 2098 (93.6%) | 1380 (61.6%) | 839 (37.4%) | 312 (13.9%) |
| gliner-pii-base+cpu-int8 | 2227 (99.3%) | 1949 (86.9%) | 1374 (61.3%) | 420 (18.7%) |
| gliner-multi-v21+cpu-int8 | 2242 (100.0%) | 2242 (100.0%) | 2242 (100.0%) | 2241 (100.0%) |
| gliner-nvidia+cpu-int8 | 2242 (100.0%) | 2242 (100.0%) | 2242 (100.0%) | 2235 (99.7%) |
| gliner-urchade+cpu-int8 | 2242 (100.0%) | 2242 (100.0%) | 2241 (100.0%) | 2183 (97.4%) |

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.