PPLX detector
pplx+homoglyph PII detection benchmark
Measured on 3 of 41 retained datasets under the same frozen scoring and masking protocol.
81.93%fully hidden annotations
89.11%detected / overlapped
7.62%extra masking
—CPU characters / second
Strengths
Highest complete-masking rates by normalized data type.
Phone numbers & email90.24%703 / 779
People's names88.91%1,323 / 1,488
Documents & identifiers88.23%2,369 / 2,685
Watch list
Lowest observed complete-masking rates; inspect before deployment.
Network identifiers59.56%215 / 361
Addresses & locations64.81%919 / 1,418
Bank accounts & cards79.03%294 / 372
Language coverage
The overall leader can differ from the best choice for one language slice.
| Language | Datasets | Fully hidden ↑ | Detected ↑ | Extra masking ↓ |
|---|---|---|---|---|
| English | 1 | 75.69% | 95.83% | 10.28% |
| Russian | 2 | 82.18% | 88.84% | 1.56% |
Configuration, source and version
- Family
- PPLX
- Revision
- f1f90a53823f5df0a1344c1e137d9fffdaab54d6
- Eligible datasets
- 3 / 41
- Character F1
- 0.552
- Untouched diagnostic
- 10.89% · 815 / 7,486
- Benchmark version
- 1.0.3 · 2026-09-09
- CPU seconds / 10k chars
- Not measured
- GPU seconds / 10k chars
- 2.153
Results by dataset
Complete masking is primary; detection shows whether any character overlapped.
| Dataset | Gold annotations | Fully hidden ↑ | Detected ↑ | Extra masking ↓ |
|---|---|---|---|---|
| corrupt-secrets-issues | 288 | 75.69% | 95.83% | 10.28% |
| corrupt-redmadrobot | 5,531 | 78.67% | 86.53% | 1.58% |
| corrupt-hivetrace | 1,667 | 93.82% | 96.52% | 1.45% |