REDAX / BENCH
Benchmarks
Understand what Redax measures and how to reproduce a run before making a performance claim.
WHYKeep raw customer text outside the model boundary.
WHATA local API that detects, replaces, and records safely.
HOWRun it, send text, pass only returned text onward.
No release-grade result yet
Current release position
Run it on your workload
regex baseline · pinned Guardrails-PII local adapter
No current accuracy or latency number is presented as a product promise. Record the detector, model revision and digest, corpus, runtime, hardware, thresholds, and Redax commit before comparing runs.
How to read the score
RedactionBench combines character-level coverage, strict overlap, and format precision. It is a comparative engineering signal—not a guarantee that every possible entity in every domain will be detected.
Recall
Did we cover the sensitive characters?
Precision
Did the span boundaries stay tight?
Latency
How quickly can the boundary respond?
Reproduce it
The benchmark fixtures and runner live in the repository so you can evaluate a policy or detector against the same corpus before shipping.
See deployment options →