Hi maintainers (cc @sroy9675 @pselvana if you're interested),
We maintain a small, fully replayable PII-detection benchmark — 18 synthetic corpora, 916 cases, ~3,400 annotated truths, across 14 languages, all synthetic identities, no real user data. It lives at packages/redact/bench in our repo (I'm the founder of OpenMasq, disclosing that once and moving on). We've already run the same harness against Microsoft Presidio's own evaluation corpus, and published the results including the cases where our engine loses.
We'd like to add DataFog as a column, run the exact same corpora and scoring script against your detector, and publish your numbers next to ours — as-is, whatever they show. This isn't a pitch: no downloads, no pricing, nothing to install beyond your own library.
What it costs you: one script that wraps DataFog's detection call to match our simple input/output contract, and a review pass on the resulting column before it goes live. You keep full control of your numbers, the corpora are inspectable line by line, and everything is reproducible from a clean checkout.
If someone here wants to write the adapter, or just wants to see the harness first, happy to open a draft PR against your repo or point you to the exact interface.
Hi maintainers (cc @sroy9675 @pselvana if you're interested),
We maintain a small, fully replayable PII-detection benchmark — 18 synthetic corpora, 916 cases, ~3,400 annotated truths, across 14 languages, all synthetic identities, no real user data. It lives at
packages/redact/benchin our repo (I'm the founder of OpenMasq, disclosing that once and moving on). We've already run the same harness against Microsoft Presidio's own evaluation corpus, and published the results including the cases where our engine loses.We'd like to add DataFog as a column, run the exact same corpora and scoring script against your detector, and publish your numbers next to ours — as-is, whatever they show. This isn't a pitch: no downloads, no pricing, nothing to install beyond your own library.
What it costs you: one script that wraps DataFog's detection call to match our simple input/output contract, and a review pass on the resulting column before it goes live. You keep full control of your numbers, the corpora are inspectable line by line, and everything is reproducible from a clean checkout.
If someone here wants to write the adapter, or just wants to see the harness first, happy to open a draft PR against your repo or point you to the exact interface.