README.md

---
license: cc-by-4.0
pretty_name: AriaDLP Persian/English Redaction Lab
task_categories:
  - token-classification
  - other
tags:
  - cybersecurity
  - pii-detection
  - redaction
  - document-analysis
  - synthetic
language:
  - en
  - fa
size_categories:
  - n<1K
---

# AriaDLP Persian / English redaction lab

Synthetic 80-document bilingual corpus for the AriaDLP workstation: text, CSV, and generated text PDFs with gold character spans. Split is by **document template**, not random rows. Regex-only is the baseline. Policy uses checksums, Persian digit folding, and a closed name lexicon.

Organization dataset and collection are public. Live Gradio is [alirezaaminzadeh/aria-dlp](https://huggingface.co/spaces/alirezaaminzadeh/aria-dlp) and the organization card is [AriaAICompany/aria-dlp](https://huggingface.co/spaces/AriaAICompany/aria-dlp). Runnable Space source is stored in `demo/`. Collection: [Aria AI — Cybersecurity](https://huggingface.co/collections/AriaAICompany/aria-ai-cybersecurity-6ab01faa3995a3bb737b58a7).

Generated data, seed 24. Mailboxes use `aria-lab.test`. PANs are Luhn-valid laboratory numbers. Names come from a closed lexicon (`Labian`, `Fixture`, `نمونه‌ای`, `آزمایشی`). No real identity records are included. Passing the suite is not a claim of complete production redaction.

## Files

- data/documents.jsonl
- data/gold_spans.json
- data/scenarios.jsonl
- data/schema.json
- data/splits.json
- data/protocol.md
- data/feature_card.md
- data/evaluation.json
- data/sample_preview.json
- demo/ — runnable Gradio Space source
- showcase/ — organization static card

## Split

Development templates: English support email, Persian HR letter, invoice CSV, onboarding worksheet.  
Test templates: Persian contract, PDF memo, hard-negative catalog, mixed ticket. Rewriting one document is not placed across the split.

## License

CC-BY-4.0. Keep the synthetic-data and laboratory-label notices.