# russian-pii-66k - ru / pii (1500 rows, 4805 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 0.298 (70.2% 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 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| opf-ru | **3** | 0.1% [0.0%, 0.1%] | 99.9% | 0.995 [0.995, 0.995] | 0.991 | 1.000 | 0.822 | 1.000 | 0.957 | 0/0 | 0 | 0 | 155 | 1.4B | train |
| nym-base | **3** | 0.1% [0.0%, 0.1%] | 99.4% | 0.993 [0.992, 0.994] | 0.987 | 0.999 | 0.942 | 0.995 | 0.660 | 0/0 | 0 | 0 | 2 | 308M |  |
| pplx | **11** | 0.2% [0.1%, 0.4%] | 97.9% | 0.955 [0.952, 0.959] | 0.920 | 0.993 | 0.648 | 0.990 | 0.853 | 0/0 | 0 | 0 | 51 | 596M |  |
| nym-small | **14** | 0.3% [0.1%, 0.5%] | 99.4% | 0.991 [0.990, 0.992] | 0.985 | 0.997 | 0.941 | 0.993 | 0.645 | 0/0 | 0 | 0 | 23 | - |  |
| mmbert32k | **127** | 2.6% [2.2%, 3.1%] | 76.0% | 0.942 [0.938, 0.945] | 0.958 | 0.926 | 0.474 | 0.971 | 0.637 | 0/0 | 0 | 0 | 3 | 308M |  |
| bardsai-eu | **165** | 3.4% [2.9%, 4.0%] | 81.6% | 0.930 [0.926, 0.934] | 0.946 | 0.915 | 0.526 | 0.971 | 0.372 | 0/0 | 0 | 0 | 242 | - |  |
| gliner-nvidia-ru | **168** | 3.5% [2.9%, 4.0%] | 91.9% | 0.961 [0.957, 0.964] | 0.967 | 0.955 | 0.850 | 0.976 | 0.930 | 0/0 | 0 | 0 | 22 | 445M |  |
| ru-pii-ner | **175** | 3.6% [3.1%, 4.3%] | 96.2% | 0.958 [0.954, 0.962] | 0.949 | 0.967 | 0.708 | 0.975 | 0.388 | 0/0 | 0 | 0 | 54 | 358M |  |
| gliner-nvidia | **189** | 3.9% [3.4%, 4.5%] | 91.1% | 0.936 [0.932, 0.940] | 0.929 | 0.944 | 0.755 | 0.971 | 0.918 | 0/0 | 0 | 0 | 5 | 445M |  |
| gliner2-fastino | **243** | 5.1% [4.3%, 5.8%] | 92.3% | 0.965 [0.962, 0.969] | 0.983 | 0.948 | 0.884 | 0.972 | 0.936 | 0/0 | 0 | 0 | 7 | 307M |  |
| gliner2-fastino-ru | **248** | 5.2% [4.4%, 6.0%] | 92.1% | 0.960 [0.956, 0.964] | 0.976 | 0.945 | 0.767 | 0.972 | 0.928 | 0/0 | 0 | 0 | 18 | 307M |  |
| ru-legal-ner | **287** | 6.0% [5.3%, 6.7%] | 86.7% | 0.939 [0.935, 0.943] | 0.944 | 0.934 | 0.598 | 0.959 | 0.801 | 0/0 | 0 | 0 | 2 | 29M |  |
| gliner2-hivetrace-omni-ru | **294** | 6.1% [5.5%, 6.8%] | 87.5% | 0.950 [0.946, 0.954] | 0.982 | 0.920 | 0.771 | 0.966 | 0.914 | 0/0 | 0 | 0 | 4 | 307M |  |
| gliner2-hivetrace-omni | **317** | 6.6% [6.0%, 7.3%] | 86.8% | 0.951 [0.947, 0.955] | 0.989 | 0.916 | 0.867 | 0.964 | 0.926 | 0/0 | 0 | 0 | 7 | 307M |  |
| openai-base | **375** | 7.8% [7.0%, 8.7%] | 89.8% | 0.901 [0.896, 0.906] | 0.867 | 0.938 | 0.528 | 0.950 | 0.767 | 0/0 | 0 | 0 | 255 | 1.4B |  |
| apararti | **430** | 8.9% [8.1%, 9.8%] | 88.5% | 0.902 [0.896, 0.907] | 0.879 | 0.925 | 0.551 | 0.942 | 0.767 | 0/0 | 0 | 0 | 54 | 1.4B |  |
| traciora | **515** | 10.7% [9.8%, 11.6%] | 86.2% | 0.906 [0.901, 0.911] | 0.908 | 0.903 | 0.598 | 0.935 | 0.832 | 0/0 | 0 | 0 | 222 | 1.4B |  |
| openmed-multilingual | **574** | 11.9% [11.0%, 12.9%] | 80.2% | 0.934 [0.929, 0.939] | 0.981 | 0.893 | 0.789 | 0.922 | 0.906 | 0/0 | 0 | 0 | 138 | 1.4B |  |
| opf-ru-v2 | **614** | 12.8% [11.8%, 13.7%] | 85.5% | 0.910 [0.905, 0.915] | 0.932 | 0.889 | 0.618 | 0.928 | 0.846 | 0/0 | 0 | 0 | 410 | 1.4B |  |
| nuner-zero | **645** | 13.4% [12.2%, 14.6%] | 82.1% | 0.839 [0.832, 0.845] | 0.802 | 0.879 | 0.396 | 0.909 | 0.726 | 0/0 | 0 | 0 | 10 | 449M |  |
| opf-kz-ru | **646** | 13.4% [12.3%, 14.6%] | 79.6% | 0.874 [0.868, 0.880] | 0.871 | 0.877 | 0.482 | 0.909 | 0.736 | 0/0 | 0 | 0 | 51 | 1.4B |  |
| pii-shield-onnx | **690** | 14.4% [13.2%, 15.5%] | 78.9% | 0.922 [0.917, 0.927] | 0.960 | 0.886 | 0.632 | 0.915 | 0.698 | 0/0 | 0 | 0 | 54 | - |  |
| fef2-secret-ru | **893** | 18.6% [17.4%, 19.7%] | 67.3% | 0.867 [0.860, 0.874] | 0.947 | 0.799 | 0.451 | 0.884 | 0.249 | 0/0 | 0 | 0 | 2 | 177M |  |
| openmed-nemotron | **972** | 20.2% [19.2%, 21.5%] | 66.5% | 0.790 [0.781, 0.799] | 0.838 | 0.748 | 0.581 | 0.824 | 0.630 | 0/0 | 0 | 0 | 141 | 1.4B |  |
| gliner2-vladlinv-ru | **1006** | 20.9% [19.4%, 22.4%] | 79.0% | 0.878 [0.869, 0.886] | 0.969 | 0.802 | 0.659 | 0.881 | 0.799 | 0/0 | 0 | 0 | 7 | 287M |  |
| gliner-stream-pii | **1015** | 21.1% [19.9%, 22.4%] | 64.6% | 0.816 [0.806, 0.824] | 0.946 | 0.717 | 0.663 | 0.853 | 0.786 | 0/0 | 0 | 0 | 41 | 677M |  |
| gliner-pii-edge | **1019** | 21.2% [19.8%, 22.5%] | 73.9% | 0.844 [0.837, 0.853] | 0.896 | 0.798 | 0.602 | 0.827 | 0.579 | 0/0 | 0 | 0 | 10 | 45M |  |
| gliner25-fastino-ru | **1067** | 22.2% [20.9%, 23.4%] | 75.5% | 0.858 [0.850, 0.866] | 0.962 | 0.774 | 0.592 | 0.873 | 0.814 | 0/0 | 0 | 0 | 2 | 287M |  |
| gliner2-large | **1081** | 22.5% [21.1%, 23.8%] | 76.1% | 0.892 [0.885, 0.899] | 0.970 | 0.825 | 0.720 | 0.867 | 0.799 | 0/0 | 0 | 0 | 11 | 486M |  |
| kalyan-ettin | **1113** | 23.2% [21.8%, 24.6%] | 61.8% | 0.820 [0.811, 0.829] | 0.945 | 0.724 | 0.545 | 0.843 | 0.710 | 0/0 | 0 | 0 | 7 | 68M |  |
| gliner-multi-v21-ru | **1365** | 28.4% [27.2%, 29.6%] | 60.7% | 0.704 [0.693, 0.714] | 0.856 | 0.598 | 0.330 | 0.802 | 0.655 | 0/0 | 0 | 0 | 7 | 289M |  |
| gliner-urchade | **1370** | 28.5% [27.1%, 29.9%] | 70.4% | 0.876 [0.869, 0.884] | 0.953 | 0.811 | 0.698 | 0.829 | 0.859 | 0/0 | 0 | 0 | 5 | 289M |  |
| gliner2-vladlinv | **1447** | 30.1% [28.3%, 31.8%] | 69.8% | 0.824 [0.813, 0.835] | 0.978 | 0.711 | 0.618 | 0.820 | 0.690 | 0/0 | 0 | 0 | 2 | 287M |  |
| gravitee-small | **1461** | 30.4% [28.9%, 32.1%] | 63.0% | 0.651 [0.641, 0.659] | 0.597 | 0.716 | 0.315 | 0.743 | 0.489 | 0/0 | 0 | 0 | 3 | 29M |  |
| gliner-urchade-ru | **1497** | 31.2% [29.8%, 32.5%] | 68.0% | 0.858 [0.851, 0.866] | 0.959 | 0.777 | 0.688 | 0.812 | 0.850 | 0/0 | 0 | 0 | 2 | 289M |  |
| gliner2-hivetrace-uni | **1546** | 32.2% [31.0%, 33.5%] | 62.6% | 0.863 [0.856, 0.870] | 0.981 | 0.771 | 0.695 | 0.803 | 0.683 | 0/0 | 0 | 0 | 15 | 147M |  |
| gliner25-fastino | **1622** | 33.8% [32.3%, 35.2%] | 63.9% | 0.787 [0.777, 0.797] | 0.969 | 0.663 | 0.661 | 0.796 | 0.711 | 0/0 | 0 | 0 | 2 | 287M |  |
| gliner-multi-v21 | **1672** | 34.8% [33.5%, 36.2%] | 54.8% | 0.653 [0.641, 0.665] | 0.867 | 0.524 | 0.330 | 0.757 | 0.597 | 0/0 | 0 | 0 | 3 | 289M |  |
| stanza-ru | **2423** | 50.4% [49.2%, 51.7%] | 39.1% | 0.551 [0.538, 0.565] | 0.849 | 0.408 | 0.194 | 0.626 | 0.177 | 0/0 | 0 | 0 | 25 | - |  |
| gliner-pii-base | **2759** | 57.4% [56.0%, 58.8%] | 39.2% | 0.627 [0.613, 0.642] | 0.964 | 0.465 | 0.519 | 0.589 | 0.609 | 0/0 | 0 | 0 | 2 | 166M |  |
| ner-ru-gherman | **2781** | 57.9% [56.5%, 59.4%] | 28.9% | 0.409 [0.395, 0.422] | 0.964 | 0.259 | 0.359 | 0.592 | 0.121 | 0/0 | 0 | 0 | 2 | 177M |  |
| ner-ru-yqelz | **2854** | 59.4% [58.2%, 60.6%] | 32.5% | 0.432 [0.418, 0.446] | 0.844 | 0.291 | 0.167 | 0.549 | 0.192 | 0/0 | 0 | 0 | 2 | 559M |  |
| davlan-xlmr | **2932** | 61.0% [59.6%, 62.6%] | 35.5% | 0.403 [0.387, 0.418] | 0.876 | 0.262 | 0.139 | 0.555 | 0.204 | 0/0 | 0 | 0 | 2 | 277M |  |
| spacy-ru-lg | **2987** | 62.2% [60.9%, 63.4%] | 29.3% | 0.436 [0.422, 0.452] | 0.920 | 0.286 | 0.140 | 0.539 | 0.203 | 0/0 | 0 | 0 | 2 | - |  |
| spacy-alrosait | **3026** | 63.0% [61.4%, 64.7%] | 36.1% | 0.373 [0.355, 0.389] | 0.822 | 0.241 | 0.066 | 0.540 | 0.171 | 0/0 | 0 | 0 | 2 | - |  |
| davlan-mbert | **3062** | 63.7% [62.4%, 65.1%] | 31.3% | 0.383 [0.366, 0.398] | 0.877 | 0.245 | 0.117 | 0.519 | 0.202 | 0/0 | 0 | 0 | 2 | 177M |  |
| gliner2-hivetrace-uni-ru | **3101** | 64.5% [63.2%, 66.0%] | 33.9% | 0.415 [0.400, 0.431] | 0.967 | 0.264 | 0.319 | 0.522 | 0.299 | 0/0 | 0 | 0 | 13 | 147M |  |
| rules-ru | **3178** | 66.1% [64.8%, 67.4%] | 33.6% | 0.623 [0.609, 0.639] | 0.996 | 0.454 | 0.498 | 0.506 | 0.621 | 0/0 | 0 | 0 | 0 | - |  |
| natasha | **3358** | 69.9% [68.6%, 71.2%] | 23.6% | 0.298 [0.285, 0.311] | 0.870 | 0.180 | 0.060 | 0.454 | 0.214 | 0/0 | 0 | 0 | 1 | - |  |
| betterleaks | **4805** | 100.0% [100.0%, 100.0%] | 0.0% | 0.000 [0.000, 0.000] | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0/0 | 0 | 0 | 0 | - |  |
| gitleaks | **4805** | 100.0% [100.0%, 100.0%] | 0.0% | 0.000 [0.000, 0.000] | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0/0 | 0 | 0 | 0 | - |  |

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`: opf-ru ≈ nym-base; pplx ≈ nym-small; bardsai-eu ≈ gliner-nvidia-ru; gliner-nvidia-ru ≈ ru-pii-ner; ru-pii-ner ≈ gliner-nvidia; gliner2-fastino ≈ gliner2-fastino-ru; gliner2-fastino-ru ≈ ru-legal-ner; ru-legal-ner ≈ gliner2-hivetrace-omni-ru; gliner2-hivetrace-omni-ru ≈ gliner2-hivetrace-omni; openmed-multilingual ≈ opf-ru-v2; opf-ru-v2 ≈ nuner-zero; nuner-zero ≈ opf-kz-ru; opf-kz-ru ≈ pii-shield-onnx; openmed-nemotron ≈ gliner2-vladlinv-ru; gliner2-vladlinv-ru ≈ gliner-stream-pii; gliner-stream-pii ≈ gliner-pii-edge; gliner-pii-edge ≈ gliner25-fastino-ru; gliner25-fastino-ru ≈ gliner2-large; gliner2-large ≈ kalyan-ettin; gliner-multi-v21-ru ≈ gliner-urchade; gliner-urchade ≈ gliner2-vladlinv; gliner2-vladlinv ≈ gravitee-small; gravitee-small ≈ gliner-urchade-ru; gliner-urchade-ru ≈ gliner2-hivetrace-uni; gliner2-hivetrace-uni ≈ gliner25-fastino; gliner25-fastino ≈ gliner-multi-v21; gliner-pii-base ≈ ner-ru-gherman; spacy-ru-lg ≈ spacy-alrosait; spacy-alrosait ≈ davlan-mbert; davlan-mbert ≈ gliner2-hivetrace-uni-ru; gliner2-hivetrace-uni-ru ≈ rules-ru; betterleaks ≈ gitleaks

Legacy runs on this set: betterleaks, pii-shield-onnx. Their meta carries no `protocol` and no `bench_sha256`, so the version of the text they were taken from is confirmed only by the row and character counts, not by a fingerprint.

## Missed by group

| model | PERSON | ADDRESS | CONTACT | ID | ACCOUNT | SECRET | DATE |
|---|---|---|---|---|---|---|---|
| spans in gold | 1151 | 901 | 793 | 1232 | 311 | 194 | 223 |
| opf-ru | 2 (0.2%) | 1 (0.1%) | 0 (0.0%) | 0 (0.0%) | 0 (0.0%) | 0 (0.0%) | 0 (0.0%) |
| nym-base | 2 (0.2%) | 1 (0.1%) | 0 (0.0%) | 0 (0.0%) | 0 (0.0%) | 0 (0.0%) | 0 (0.0%) |
| pplx | 4 (0.3%) | 6 (0.7%) | 0 (0.0%) | 0 (0.0%) | 1 (0.3%) | 0 (0.0%) | 0 (0.0%) |
| nym-small | 5 (0.4%) | 5 (0.6%) | 0 (0.0%) | 0 (0.0%) | 2 (0.6%) | 0 (0.0%) | 2 (0.9%) |
| mmbert32k | 32 (2.8%) | 32 (3.6%) | 0 (0.0%) | 5 (0.4%) | 42 (13.5%) | 16 (8.2%) | 0 (0.0%) |
| bardsai-eu | 5 (0.4%) | 62 (6.9%) | 1 (0.1%) | 57 (4.6%) | 30 (9.6%) | 8 (4.1%) | 2 (0.9%) |
| gliner-nvidia-ru | 8 (0.7%) | 79 (8.8%) | 5 (0.6%) | 40 (3.2%) | 35 (11.3%) | 1 (0.5%) | 0 (0.0%) |
| ru-pii-ner | 13 (1.1%) | 136 (15.1%) | 12 (1.5%) | 4 (0.3%) | 6 (1.9%) | 4 (2.1%) | 0 (0.0%) |
| gliner-nvidia | 5 (0.4%) | 57 (6.3%) | 1 (0.1%) | 124 (10.1%) | 2 (0.6%) | 0 (0.0%) | 0 (0.0%) |
| gliner2-fastino | 52 (4.5%) | 120 (13.3%) | 0 (0.0%) | 62 (5.0%) | 6 (1.9%) | 3 (1.5%) | 0 (0.0%) |
| gliner2-fastino-ru | 31 (2.7%) | 142 (15.8%) | 0 (0.0%) | 60 (4.9%) | 12 (3.9%) | 3 (1.5%) | 0 (0.0%) |
| ru-legal-ner | 47 (4.1%) | 91 (10.1%) | 10 (1.3%) | 7 (0.6%) | 54 (17.4%) | 44 (22.7%) | 34 (15.2%) |
| gliner2-hivetrace-omni-ru | 11 (1.0%) | 24 (2.7%) | 5 (0.6%) | 228 (18.5%) | 23 (7.4%) | 2 (1.0%) | 1 (0.4%) |
| gliner2-hivetrace-omni | 34 (3.0%) | 29 (3.2%) | 1 (0.1%) | 186 (15.1%) | 65 (20.9%) | 2 (1.0%) | 0 (0.0%) |
| openai-base | 162 (14.1%) | 151 (16.8%) | 0 (0.0%) | 1 (0.1%) | 59 (19.0%) | 2 (1.0%) | 0 (0.0%) |
| apararti | 172 (14.9%) | 177 (19.6%) | 0 (0.0%) | 0 (0.0%) | 80 (25.7%) | 1 (0.5%) | 0 (0.0%) |
| traciora | 83 (7.2%) | 198 (22.0%) | 1 (0.1%) | 6 (0.5%) | 72 (23.2%) | 3 (1.5%) | 152 (68.2%) |
| openmed-multilingual | 231 (20.1%) | 275 (30.5%) | 0 (0.0%) | 14 (1.1%) | 53 (17.0%) | 1 (0.5%) | 0 (0.0%) |
| opf-ru-v2 | 90 (7.8%) | 186 (20.6%) | 0 (0.0%) | 0 (0.0%) | 114 (36.7%) | 1 (0.5%) | 223 (100.0%) |
| nuner-zero | 298 (25.9%) | 142 (15.8%) | 4 (0.5%) | 195 (15.8%) | 5 (1.6%) | 1 (0.5%) | 0 (0.0%) |
| opf-kz-ru | 225 (19.5%) | 291 (32.3%) | 0 (0.0%) | 0 (0.0%) | 125 (40.2%) | 5 (2.6%) | 0 (0.0%) |
| pii-shield-onnx | 368 (32.0%) | 238 (26.4%) | 4 (0.5%) | 12 (1.0%) | 63 (20.3%) | 5 (2.6%) | 0 (0.0%) |
| fef2-secret-ru | 23 (2.0%) | 299 (33.2%) | 181 (22.8%) | 133 (10.8%) | 46 (14.8%) | 0 (0.0%) | 211 (94.6%) |
| openmed-nemotron | 149 (12.9%) | 260 (28.9%) | 5 (0.6%) | 494 (40.1%) | 58 (18.6%) | 6 (3.1%) | 0 (0.0%) |
| gliner2-vladlinv-ru | 49 (4.3%) | 555 (61.6%) | 6 (0.8%) | 283 (23.0%) | 89 (28.6%) | 24 (12.4%) | 0 (0.0%) |
| gliner-stream-pii | 115 (10.0%) | 387 (43.0%) | 50 (6.3%) | 405 (32.9%) | 32 (10.3%) | 22 (11.3%) | 4 (1.8%) |
| gliner-pii-edge | 116 (10.1%) | 457 (50.7%) | 23 (2.9%) | 358 (29.1%) | 5 (1.6%) | 24 (12.4%) | 36 (16.1%) |
| gliner25-fastino-ru | 46 (4.0%) | 191 (21.2%) | 0 (0.0%) | 565 (45.9%) | 175 (56.3%) | 29 (14.9%) | 61 (27.4%) |
| gliner2-large | 408 (35.4%) | 375 (41.6%) | 3 (0.4%) | 229 (18.6%) | 52 (16.7%) | 13 (6.7%) | 1 (0.4%) |
| kalyan-ettin | 102 (8.9%) | 492 (54.6%) | 25 (3.2%) | 436 (35.4%) | 42 (13.5%) | 16 (8.2%) | 0 (0.0%) |
| gliner-multi-v21-ru | 12 (1.0%) | 147 (16.3%) | 212 (26.7%) | 815 (66.2%) | 34 (10.9%) | 138 (71.1%) | 7 (3.1%) |
| gliner-urchade | 1051 (91.3%) | 93 (10.3%) | 0 (0.0%) | 201 (16.3%) | 22 (7.1%) | 3 (1.5%) | 0 (0.0%) |
| gliner2-vladlinv | 60 (5.2%) | 806 (89.5%) | 7 (0.9%) | 486 (39.4%) | 85 (27.3%) | 3 (1.5%) | 0 (0.0%) |
| gravitee-small | 548 (47.6%) | 132 (14.7%) | 21 (2.6%) | 353 (28.7%) | 282 (90.7%) | 125 (64.4%) | 0 (0.0%) |
| gliner-urchade-ru | 1030 (89.5%) | 98 (10.9%) | 0 (0.0%) | 268 (21.8%) | 99 (31.8%) | 1 (0.5%) | 1 (0.4%) |
| gliner2-hivetrace-uni | 898 (78.0%) | 348 (38.6%) | 9 (1.1%) | 100 (8.1%) | 77 (24.8%) | 113 (58.2%) | 1 (0.4%) |
| gliner25-fastino | 150 (13.0%) | 361 (40.1%) | 0 (0.0%) | 812 (65.9%) | 238 (76.5%) | 6 (3.1%) | 55 (24.7%) |
| gliner-multi-v21 | 19 (1.7%) | 208 (23.1%) | 236 (29.8%) | 1005 (81.6%) | 39 (12.5%) | 163 (84.0%) | 2 (0.9%) |
| stanza-ru | 2 (0.2%) | 376 (41.7%) | 722 (91.0%) | 831 (67.5%) | 237 (76.2%) | 32 (16.5%) | 223 (100.0%) |
| gliner-pii-base | 770 (66.9%) | 598 (66.4%) | 41 (5.2%) | 1105 (89.7%) | 41 (13.2%) | 81 (41.8%) | 123 (55.2%) |
| ner-ru-gherman | 33 (2.9%) | 191 (21.2%) | 685 (86.4%) | 1232 (100.0%) | 223 (71.7%) | 194 (100.0%) | 223 (100.0%) |
| ner-ru-yqelz | 14 (1.2%) | 444 (49.3%) | 626 (78.9%) | 1114 (90.4%) | 244 (78.5%) | 189 (97.4%) | 223 (100.0%) |
| davlan-xlmr | 5 (0.4%) | 221 (24.5%) | 789 (99.5%) | 1232 (100.0%) | 268 (86.2%) | 194 (100.0%) | 223 (100.0%) |
| spacy-ru-lg | 9 (0.8%) | 493 (54.7%) | 780 (98.4%) | 1014 (82.3%) | 286 (92.0%) | 182 (93.8%) | 223 (100.0%) |
| spacy-alrosait | 69 (6.0%) | 216 (24.0%) | 788 (99.4%) | 1226 (99.5%) | 311 (100.0%) | 193 (99.5%) | 223 (100.0%) |
| davlan-mbert | 9 (0.8%) | 314 (34.9%) | 788 (99.4%) | 1232 (100.0%) | 303 (97.4%) | 193 (99.5%) | 223 (100.0%) |
| gliner2-hivetrace-uni-ru | 359 (31.2%) | 740 (82.1%) | 787 (99.2%) | 875 (71.0%) | 121 (38.9%) | 185 (95.4%) | 34 (15.2%) |
| rules-ru | 1151 (100.0%) | 901 (100.0%) | 0 (0.0%) | 424 (34.4%) | 310 (99.7%) | 169 (87.1%) | 223 (100.0%) |
| natasha | 28 (2.4%) | 592 (65.7%) | 793 (100.0%) | 1229 (99.8%) | 311 (100.0%) | 182 (93.8%) | 223 (100.0%) |
| betterleaks | 1151 (100.0%) | 901 (100.0%) | 793 (100.0%) | 1232 (100.0%) | 311 (100.0%) | 194 (100.0%) | 223 (100.0%) |
| gitleaks | 1151 (100.0%) | 901 (100.0%) | 793 (100.0%) | 1232 (100.0%) | 311 (100.0%) | 194 (100.0%) | 223 (100.0%) |

## Char recall by gold type

| type | group | opf-ru | nym-base | pplx | nym-small | mmbert32k | bardsai-eu | gliner-nvidia-ru | ru-pii-ner | gliner-nvidia | gliner2-fastino | gliner2-fastino-ru | ru-legal-ner | gliner2-hivetrace-omni-ru | gliner2-hivetrace-omni | openai-base | apararti | traciora | openmed-multilingual | opf-ru-v2 | nuner-zero | opf-kz-ru | pii-shield-onnx | fef2-secret-ru | openmed-nemotron | gliner2-vladlinv-ru | gliner-stream-pii | gliner-pii-edge | gliner25-fastino-ru | gliner2-large | kalyan-ettin | gliner-multi-v21-ru | gliner-urchade | gliner2-vladlinv | gravitee-small | gliner-urchade-ru | gliner2-hivetrace-uni | gliner25-fastino | gliner-multi-v21 | stanza-ru | gliner-pii-base | ner-ru-gherman | ner-ru-yqelz | davlan-xlmr | spacy-ru-lg | spacy-alrosait | davlan-mbert | gliner2-hivetrace-uni-ru | rules-ru | natasha | betterleaks | gitleaks |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| ACCOUNTNUM | ID | 1.000 | 1.000 | 1.000 | 1.000 | 1.000 | 0.949 | 0.943 | 1.000 | 0.936 | 0.919 | 0.872 | 0.997 | 0.953 | 0.997 | 1.000 | 1.000 | 1.000 | 0.960 | 1.000 | 0.990 | 1.000 | 1.000 | 0.923 | 0.421 | 0.754 | 0.761 | 1.000 | 0.825 | 0.929 | 0.822 | 0.448 | 0.956 | 0.370 | 0.653 | 0.929 | 0.973 | 0.508 | 0.178 | 0.835 | 0.222 | 0.000 | 0.340 | 0.000 | 0.721 | 0.000 | 0.000 | 0.054 | 0.195 | 0.000 | 0.000 | 0.000 |
| BUILDINGNUM | ADDRESS | 0.998 | 0.996 | 1.000 | 0.994 | 0.994 | 0.960 | 0.894 | 0.947 | 0.955 | 0.773 | 0.741 | 0.917 | 0.992 | 0.989 | 0.970 | 0.949 | 0.962 | 0.435 | 0.938 | 0.972 | 0.790 | 0.864 | 0.200 | 0.556 | 0.363 | 0.454 | 0.308 | 0.726 | 0.546 | 0.293 | 0.741 | 0.796 | 0.081 | 0.964 | 0.926 | 0.117 | 0.556 | 0.701 | 0.096 | 0.493 | 0.926 | 0.042 | 0.592 | 0.000 | 0.972 | 0.176 | 0.004 | 0.000 | 0.000 | 0.000 | 0.000 |
| CITY | ADDRESS | 1.000 | 0.992 | 0.919 | 0.982 | 0.755 | 0.810 | 0.746 | 0.721 | 0.733 | 0.786 | 0.769 | 0.815 | 0.712 | 0.692 | 0.618 | 0.523 | 0.546 | 0.424 | 0.500 | 0.556 | 0.360 | 0.665 | 0.747 | 0.433 | 0.285 | 0.344 | 0.454 | 0.680 | 0.559 | 0.387 | 0.666 | 0.828 | 0.091 | 0.649 | 0.754 | 0.614 | 0.576 | 0.664 | 0.727 | 0.349 | 0.673 | 0.827 | 0.890 | 0.709 | 0.675 | 0.827 | 0.197 | 0.000 | 0.556 | 0.000 | 0.000 |
| CREDITCARDNUMBER | ID | 1.000 | 1.000 | 1.000 | 1.000 | 1.000 | 0.920 | 0.987 | 1.000 | 0.721 | 1.000 | 1.000 | 0.993 | 0.993 | 0.987 | 1.000 | 1.000 | 0.974 | 1.000 | 1.000 | 0.993 | 1.000 | 1.000 | 0.775 | 0.774 | 0.450 | 0.993 | 1.000 | 0.258 | 0.940 | 0.855 | 0.033 | 0.933 | 0.973 | 0.763 | 0.702 | 0.855 | 0.112 | 0.026 | 0.013 | 0.180 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0.106 | 0.993 | 0.000 | 0.000 | 0.000 |
| DATEOFBIRTH | DATE | 1.000 | 0.998 | 1.000 | 0.991 | 1.000 | 0.985 | 1.000 | 1.000 | 1.000 | 1.000 | 1.000 | 0.693 | 0.996 | 1.000 | 1.000 | 1.000 | 0.185 | 0.999 | 0.000 | 1.000 | 0.998 | 0.996 | 0.033 | 0.999 | 1.000 | 0.982 | 0.839 | 0.726 | 0.996 | 0.996 | 0.969 | 1.000 | 1.000 | 1.000 | 0.996 | 0.996 | 0.753 | 0.991 | 0.000 | 0.448 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0.848 | 0.000 | 0.000 | 0.000 | 0.000 |
| DRIVERLICENSENUM | ID | 0.998 | 0.998 | 1.000 | 0.997 | 0.785 | 0.919 | 0.954 | 1.000 | 0.954 | 0.985 | 1.000 | 0.941 | 0.847 | 0.977 | 0.992 | 0.997 | 0.987 | 0.973 | 0.997 | 0.877 | 0.984 | 0.971 | 0.758 | 0.495 | 0.908 | 0.149 | 0.649 | 0.183 | 0.947 | 0.242 | 0.046 | 0.962 | 0.924 | 0.795 | 0.916 | 0.893 | 0.153 | 0.214 | 0.038 | 0.000 | 0.000 | 0.013 | 0.000 | 0.010 | 0.010 | 0.000 | 0.450 | 1.000 | 0.000 | 0.000 | 0.000 |
| EMAIL | CONTACT | 1.000 | 1.000 | 0.999 | 1.000 | 0.997 | 0.800 | 1.000 | 1.000 | 1.000 | 1.000 | 1.000 | 1.000 | 1.000 | 1.000 | 1.000 | 1.000 | 1.000 | 0.999 | 0.998 | 1.000 | 1.000 | 1.000 | 1.000 | 1.000 | 0.998 | 0.981 | 0.995 | 1.000 | 0.997 | 0.999 | 0.513 | 0.999 | 0.998 | 0.978 | 1.000 | 1.000 | 1.000 | 0.436 | 0.176 | 1.000 | 0.138 | 0.001 | 0.005 | 0.029 | 0.003 | 0.008 | 0.000 | 1.000 | 0.000 | 0.000 | 0.000 |
| GIVENNAME | PERSON | 0.997 | 0.997 | 0.995 | 0.993 | 0.988 | 0.992 | 0.984 | 0.981 | 0.994 | 0.921 | 0.957 | 0.953 | 0.982 | 0.961 | 0.830 | 0.809 | 0.897 | 0.747 | 0.890 | 0.751 | 0.795 | 0.658 | 0.968 | 0.839 | 0.947 | 0.867 | 0.923 | 0.944 | 0.534 | 0.935 | 0.989 | 0.077 | 0.938 | 0.476 | 0.092 | 0.299 | 0.830 | 0.978 | 0.995 | 0.468 | 0.955 | 0.983 | 0.993 | 0.985 | 0.908 | 0.990 | 0.702 | 0.000 | 0.963 | 0.000 | 0.000 |
| IDCARDNUM | ID | 1.000 | 0.997 | 1.000 | 1.000 | 0.808 | 0.897 | 0.940 | 1.000 | 0.930 | 0.945 | 0.951 | 0.989 | 0.095 | 0.667 | 0.987 | 0.981 | 0.998 | 0.990 | 0.999 | 0.604 | 0.969 | 0.789 | 0.706 | 0.428 | 0.532 | 0.144 | 0.816 | 0.269 | 0.638 | 0.447 | 0.045 | 0.234 | 0.896 | 0.937 | 0.169 | 0.846 | 0.721 | 0.199 | 0.005 | 0.070 | 0.000 | 0.001 | 0.000 | 0.000 | 0.006 | 0.000 | 0.607 | 1.000 | 0.000 | 0.000 | 0.000 |
| PASSWORD | SECRET | 1.000 | 1.000 | 1.000 | 1.000 | 0.808 | 0.927 | 0.993 | 0.977 | 1.000 | 0.971 | 0.977 | 0.663 | 0.989 | 0.989 | 0.991 | 0.995 | 0.987 | 0.995 | 0.993 | 0.988 | 0.973 | 0.974 | 0.999 | 0.936 | 0.882 | 0.866 | 0.823 | 0.833 | 0.888 | 0.863 | 0.274 | 0.977 | 0.987 | 0.337 | 0.992 | 0.402 | 0.952 | 0.131 | 0.735 | 0.569 | 0.000 | 0.017 | 0.000 | 0.065 | 0.007 | 0.005 | 0.045 | 0.102 | 0.037 | 0.000 | 0.000 |
| SOCIALNUM | ID | 1.000 | 1.000 | 1.000 | 1.000 | 0.832 | 0.911 | 0.984 | 1.000 | 0.978 | 0.878 | 0.979 | 0.968 | 0.963 | 0.809 | 0.994 | 0.991 | 0.997 | 0.961 | 1.000 | 0.713 | 0.992 | 0.989 | 0.962 | 0.492 | 0.979 | 0.241 | 0.180 | 0.830 | 0.968 | 0.193 | 0.186 | 0.931 | 0.452 | 0.943 | 0.957 | 0.946 | 0.277 | 0.090 | 0.013 | 0.053 | 0.000 | 0.035 | 0.000 | 0.004 | 0.000 | 0.000 | 0.355 | 0.021 | 0.000 | 0.000 | 0.000 |
| STREET | ADDRESS | 1.000 | 0.998 | 0.997 | 0.998 | 0.897 | 0.955 | 0.920 | 0.925 | 0.951 | 0.812 | 0.774 | 0.959 | 0.950 | 0.920 | 0.963 | 0.906 | 0.917 | 0.727 | 0.900 | 0.913 | 0.583 | 0.700 | 0.689 | 0.728 | 0.384 | 0.465 | 0.227 | 0.834 | 0.682 | 0.156 | 0.911 | 0.983 | 0.104 | 0.909 | 0.972 | 0.677 | 0.722 | 0.883 | 0.591 | 0.061 | 0.623 | 0.325 | 0.988 | 0.256 | 0.990 | 0.943 | 0.000 | 0.000 | 0.157 | 0.000 | 0.000 |
| SURNAME | PERSON | 1.000 | 1.000 | 0.999 | 1.000 | 0.955 | 1.000 | 1.000 | 1.000 | 0.996 | 0.994 | 0.996 | 0.981 | 0.997 | 0.980 | 0.905 | 0.911 | 0.972 | 0.843 | 0.971 | 0.714 | 0.834 | 0.715 | 0.998 | 0.914 | 0.973 | 0.942 | 0.874 | 0.984 | 0.766 | 0.883 | 0.991 | 0.098 | 0.966 | 0.599 | 0.112 | 0.116 | 0.928 | 0.989 | 1.000 | 0.167 | 0.998 | 0.993 | 1.000 | 0.998 | 0.983 | 0.998 | 0.665 | 0.000 | 0.993 | 0.000 | 0.000 |
| TAXNUM | ID | 1.000 | 1.000 | 1.000 | 1.000 | 1.000 | 0.977 | 1.000 | 0.985 | 0.852 | 0.992 | 0.966 | 0.992 | 0.981 | 0.705 | 0.996 | 1.000 | 0.992 | 1.000 | 1.000 | 0.795 | 1.000 | 0.996 | 0.951 | 0.458 | 0.936 | 0.402 | 0.519 | 0.561 | 0.568 | 0.481 | 0.867 | 0.977 | 0.390 | 0.280 | 0.939 | 0.924 | 0.133 | 0.322 | 0.508 | 0.038 | 0.000 | 0.011 | 0.000 | 0.000 | 0.000 | 0.000 | 0.288 | 1.000 | 0.011 | 0.000 | 0.000 |
| TELEPHONENUM | CONTACT | 1.000 | 1.000 | 1.000 | 1.000 | 0.900 | 0.997 | 0.990 | 0.967 | 0.998 | 1.000 | 1.000 | 0.971 | 0.990 | 0.998 | 1.000 | 1.000 | 0.991 | 0.975 | 0.999 | 0.989 | 1.000 | 0.986 | 0.398 | 0.973 | 0.987 | 0.780 | 0.946 | 1.000 | 0.995 | 0.790 | 0.570 | 1.000 | 0.984 | 0.923 | 1.000 | 0.975 | 1.000 | 0.607 | 0.000 | 0.904 | 0.000 | 0.258 | 0.000 | 0.000 | 0.009 | 0.000 | 0.012 | 1.000 | 0.000 | 0.000 | 0.000 |
| USERNAME | ACCOUNT | 1.000 | 1.000 | 0.997 | 0.992 | 0.882 | 0.885 | 0.887 | 0.983 | 0.992 | 0.979 | 0.962 | 0.847 | 0.928 | 0.784 | 0.820 | 0.754 | 0.784 | 0.826 | 0.643 | 0.980 | 0.598 | 0.835 | 0.848 | 0.811 | 0.717 | 0.895 | 0.985 | 0.449 | 0.832 | 0.875 | 0.885 | 0.929 | 0.736 | 0.094 | 0.683 | 0.745 | 0.239 | 0.865 | 0.245 | 0.871 | 0.296 | 0.215 | 0.129 | 0.091 | 0.000 | 0.026 | 0.598 | 0.004 | 0.000 | 0.000 | 0.000 |
| ZIPCODE | ADDRESS | 1.000 | 1.000 | 0.993 | 1.000 | 0.953 | 0.687 | 0.833 | 0.887 | 0.833 | 0.993 | 0.987 | 0.727 | 0.973 | 0.973 | 0.820 | 0.907 | 0.740 | 0.967 | 0.947 | 0.867 | 0.900 | 0.433 | 0.347 | 0.913 | 0.573 | 0.747 | 0.767 | 0.753 | 0.380 | 0.793 | 0.753 | 0.907 | 0.133 | 0.873 | 0.900 | 0.680 | 0.227 | 0.500 | 0.000 | 0.240 | 0.000 | 0.260 | 0.007 | 0.000 | 0.287 | 0.000 | 0.460 | 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 |
|---|---|---|---|---|
| opf-ru | 3 (0.1%) | 2 (0.0%) | 2 (0.0%) | 2 (0.0%) |
| nym-base | 3 (0.1%) | 3 (0.1%) | 3 (0.1%) | 3 (0.1%) |
| nym-small | 14 (0.3%) | 13 (0.3%) | 13 (0.3%) | 13 (0.3%) |
| mmbert32k | 127 (2.6%) | 12 (0.2%) | 10 (0.2%) | 10 (0.2%) |
| bardsai-eu | 165 (3.4%) | 69 (1.4%) | 55 (1.1%) | 54 (1.1%) |
| gliner-nvidia-ru | 168 (3.5%) | 94 (2.0%) | 64 (1.3%) | 31 (0.6%) |
| gliner-nvidia | 189 (3.9%) | 93 (1.9%) | 37 (0.8%) | 16 (0.3%) |
| gliner2-fastino | 243 (5.1%) | 124 (2.6%) | 93 (1.9%) | 47 (1.0%) |
| gliner2-fastino-ru | 248 (5.2%) | 170 (3.5%) | 124 (2.6%) | 76 (1.6%) |
| ru-legal-ner | 287 (6.0%) | 85 (1.8%) | 75 (1.6%) | 72 (1.5%) |
| gliner2-hivetrace-omni-ru | 294 (6.1%) | 113 (2.4%) | 22 (0.5%) | 5 (0.1%) |
| gliner2-hivetrace-omni | 317 (6.6%) | 83 (1.7%) | 31 (0.6%) | 12 (0.2%) |
| openai-base | 375 (7.8%) | 368 (7.7%) | 368 (7.7%) | 368 (7.7%) |
| apararti | 430 (8.9%) | 418 (8.7%) | 418 (8.7%) | 418 (8.7%) |
| traciora | 515 (10.7%) | 504 (10.5%) | 502 (10.4%) | 502 (10.4%) |
| openmed-multilingual | 574 (11.9%) | 339 (7.1%) | 317 (6.6%) | 315 (6.6%) |
| opf-ru-v2 | 614 (12.8%) | 614 (12.8%) | 612 (12.7%) | 612 (12.7%) |
| nuner-zero | 645 (13.4%) | 189 (3.9%) | 97 (2.0%) | 39 (0.8%) |
| opf-kz-ru | 646 (13.4%) | 628 (13.1%) | 628 (13.1%) | 628 (13.1%) |
| fef2-secret-ru | 893 (18.6%) | 782 (16.3%) | 779 (16.2%) | 779 (16.2%) |
| openmed-nemotron | 972 (20.2%) | 646 (13.4%) | 617 (12.8%) | 615 (12.8%) |
| gliner2-vladlinv-ru | 1006 (20.9%) | 965 (20.1%) | 935 (19.5%) | 896 (18.6%) |
| gliner-stream-pii | 1015 (21.1%) | 606 (12.6%) | 408 (8.5%) | 230 (4.8%) |
| gliner-pii-edge | 1019 (21.2%) | 28 (0.6%) | 3 (0.1%) | 0 (0.0%) |
| gliner25-fastino-ru | 1067 (22.2%) | 995 (20.7%) | 978 (20.4%) | 955 (19.9%) |
| gliner2-large | 1081 (22.5%) | 587 (12.2%) | 363 (7.6%) | 176 (3.7%) |
| kalyan-ettin | 1113 (23.2%) | 546 (11.4%) | 480 (10.0%) | 472 (9.8%) |
| gliner-multi-v21-ru | 1365 (28.4%) | 404 (8.4%) | 127 (2.6%) | 18 (0.4%) |
| gliner-urchade | 1370 (28.5%) | 1193 (24.8%) | 1094 (22.8%) | 975 (20.3%) |
| gliner2-vladlinv | 1447 (30.1%) | 1414 (29.4%) | 1385 (28.8%) | 1338 (27.8%) |
| gravitee-small | 1461 (30.4%) | 1217 (25.3%) | 1201 (25.0%) | 1201 (25.0%) |
| gliner-urchade-ru | 1497 (31.2%) | 1202 (25.0%) | 1037 (21.6%) | 831 (17.3%) |
| gliner2-hivetrace-uni | 1546 (32.2%) | 865 (18.0%) | 637 (13.3%) | 330 (6.9%) |
| gliner25-fastino | 1622 (33.8%) | 1471 (30.6%) | 1424 (29.6%) | 1370 (28.5%) |
| gliner-multi-v21 | 1672 (34.8%) | 392 (8.2%) | 87 (1.8%) | 7 (0.1%) |
| gliner-pii-base | 2759 (57.4%) | 545 (11.3%) | 105 (2.2%) | 6 (0.1%) |
| ner-ru-gherman | 2781 (57.9%) | 2758 (57.4%) | 2758 (57.4%) | 2758 (57.4%) |
| ner-ru-yqelz | 2854 (59.4%) | 2772 (57.7%) | 2772 (57.7%) | 2772 (57.7%) |
| davlan-xlmr | 2932 (61.0%) | 2929 (61.0%) | 2929 (61.0%) | 2929 (61.0%) |
| davlan-mbert | 3062 (63.7%) | 3059 (63.7%) | 3059 (63.7%) | 3059 (63.7%) |
| gliner2-hivetrace-uni-ru | 3101 (64.5%) | 2120 (44.1%) | 1550 (32.3%) | 742 (15.4%) |

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.