Results updated 19 Sep 2026

Ensembles

Explore the masking tradeoffs of fixed detector compositions.

Dataset slice
FIXED REFERENCE COMPOSITION

PPLX + Fastino + mmBERT + BardsAI

One fixed union across the same dataset slice.

94.40%fully hidden annotations
3,254untouched / 227,466

Composition outcomes

Manually selected compositions evaluated on this benchmark; no independent selection holdout.

CompositionUntouched ↓Fully hidden ↑Extra masking ↓Char F1 ↑
20.95%74.92%11.42%0.716
13.19%79.86%5.88%0.763
3.46%91.94%15.27%0.736
2.09%92.96%18.84%0.701
1.87%94.08%17.10%0.714
1.43%94.40%19.82%0.685
Partial coverage · 40/41 datasets27.17%64.40%6.37%0.713

Detection and full hiding

A larger union can hide more annotations while masking more unannotated text.

Fully hiddenPartially hiddenUntouched

Ensemble costs in the source report are derived from member throughput, not measured end-to-end latency. Compositions containing BardsAI have a mixed-device GPU estimate.

PII Arena · Frozen 09 Sep 2026Publication v1.0.3 · MIT licensed