File size: 722 Bytes
eef4eed | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 | # AriaDLP laboratory protocol
Synthetic bilingual DLP corpus. Seed 24. Split by template.
## What is labeled
Person names from a closed laboratory lexicon, mailbox addresses on `aria-lab.test`,
Iranian mobile numbers, checksum-valid national IDs, Luhn-valid test PANs, IBANs,
and `ARIA-24xxx` employee ids.
## Hard negatives
Invoice / SKU tokens, ISBN, semantic versions, `Mr. Protocol`, dates, and numbers
that fail national-ID / Luhn / IBAN checks.
## PDF
`en_pdf_memo` is a generated text PDF. Gold includes body spans and metadata fields.
Acceptance is content removal plus metadata wipe, not drawing a rectangle.
## Limits
This corpus does not prove production recall. Unknown items stay on the review list.
|