PPLX detector
pplx+ov100 PII detection benchmark
Measured on 11 of 41 retained datasets under the same frozen scoring and masking protocol.
61.81%fully hidden annotations
64.12%detected / overlapped
8.18%extra masking
—CPU characters / second
Strengths
Highest complete-masking rates by normalized data type.
Logins & usernames100.00%7,794 / 7,794
Network identifiers99.21%17,359 / 17,498
Phone numbers & email97.98%5,631 / 5,747
Watch list
Lowest observed complete-masking rates; inspect before deployment.
Organizations2.74%360 / 13,123
Addresses & locations22.00%3,374 / 15,334
Other sensitive attributes49.94%444 / 889
Language coverage
The overall leader can differ from the best choice for one language slice.
| Language | Datasets | Fully hidden ↑ | Detected ↑ | Extra masking ↓ |
|---|---|---|---|---|
| English | 6 | 71.19% | 76.77% | 9.07% |
| Russian | 5 | 59.61% | 61.17% | 2.32% |
Configuration, source and version
- Family
- PPLX
- Revision
- f1f90a53823f5df0a1344c1e137d9fffdaab54d6
- Eligible datasets
- 11 / 41
- Character F1
- 0.675
- Untouched diagnostic
- 35.88% · 33,130 / 92,332
- Benchmark version
- 1.0.3 · 2026-09-09
- CPU seconds / 10k chars
- Not measured
- GPU seconds / 10k chars
- 1.928
Results by dataset
Complete masking is primary; detection shows whether any character overlapped.
| Dataset | Gold annotations | Fully hidden ↑ | Detected ↑ | Extra masking ↓ |
|---|---|---|---|---|
| nerel | 24,369 | 15.27% | 15.66% | — |
| factrueval | 7,966 | 16.92% | 16.96% | — |
| tab-echr | 3,830 | 29.66% | 44.23% | — |
| dialogpii-en | 3,087 | 71.36% | 75.64% | 6.13% |
| jayguard | 1,195 | 72.30% | 83.93% | 2.16% |
| secrets-issues | 288 | 75.69% | 96.18% | 9.77% |
| leak-museum | 101 | 86.14% | 88.12% | 3.90% |
| nemotron-pii | 9,391 | 86.50% | 88.25% | 6.93% |
| creddata | 776 | 86.60% | 94.20% | 7.55% |
| redact-ru | 9,230 | 87.25% | 90.83% | 8.63% |
| synth-ru-tickets | 32,099 | 95.45% | 97.30% | — |