hyperbots-inc/synthetic-pii-redaction

Synthetic PII Redaction

100 page images of PII-heavy enterprise documents: KYC and account-opening forms, patient intake forms, insurance claims, vendor bank-detail forms, credit applications, payslips, bank and card statements, utility bills, offer letters, collection letters, lab reports, timesheets, expense reports and meeting minutes. Every PII value is labelled twice: as a character span in the page text and as pixel boxes in the image. Each row also has the redacted text and a de-identified version with consistent synthetic replacements.

The format follows HyperAPI POST /v1/redact, which offers redact (black boxes) and deidentify (synthetic replacement) modes and returns a per-type summary.

PII types

PERSON_NAME, EMAIL, PHONE, ADDRESS (personal addresses only; company addresses are not PII), DATE_OF_BIRTH, NATIONAL_ID, PASSPORT_NUMBER, ACCOUNT_NUMBER, IBAN (valid check digits). Signatures and logos are boxed separately.

Columns

Column	Description
image	1240 x 1754 JPEG
ocr_text	Page text in reading order
pii	List of {type, text, start, end, bboxes, synthetic_replacement}. start/end index ocr_text; bboxes is a list because a value can wrap across lines
signatures, logos	Boxes for image-level redaction (include_logos)
summary	JSON string of counts per type, e.g. {"PERSON_NAME": 3, "EMAIL": 1}
redacted_text	ocr_text with each PII value replaced by [REDACTED]
deidentified_text	ocr_text with type-consistent synthetic values

from datasets import load_dataset
ds = load_dataset("hyperbots-inc/synthetic-pii-redaction", split="test")
ex = ds[0]
for e in ex["pii"]:
    print(e["type"], repr(e["text"]), e["bboxes"])

Suggested metrics

Use span-level precision, recall and F1 per type (exact and overlap), pixel recall of predicted masks over gold boxes, and leak rate (the share of gold PII strings still present in your redacted output).

How the data was made

Every document is generated from a seeded random number generator by docgen.js and rendered on an HTML canvas in Chrome on macOS. Images use the system's print fonts and, for handwriting, the handwriting-style fonts installed with macOS (Bradley Hand, Noteworthy, Marker Felt, Chalkboard SE, Snell Roundhand, Apple Chancery, Savoye LET, Brush Script MT). Labels are taken from the generator itself, so there is no human annotation error. Field and block boxes are recorded at the moment each element is drawn.

All companies, people, addresses, account numbers and amounts are fictional. Phone numbers use ranges reserved for fiction where these exist (US 555-01xx, UK 07700 900xxx, AU 0491 570 1xx). Email addresses use the reserved example.com / example.org domains, and US ID numbers start with 9, which is never issued as an SSN.

To rebuild the dataset, load generator/docgen.js and generator/datasets.js in a browser page and run await DocSets.build('redaction'). Rendering depends on installed fonts, so pixels can differ on other systems, but text and labels are identical for the same seed.

Limitations

• The documents are synthetic. Their layouts, vocabulary and value ranges are much narrower than real enterprise traffic. Use them for benchmarking, regression tests and pipeline development, not as a replacement for evaluation on your own documents.
• All content is in English, with values formatted for 9 locales (US, UK, IN, CA, AU, DE, FR, SG, AE).
• Handwriting is rendered from fonts with per-character jitter. It is a stand-in for real handwriting, not a sample of it.

About HyperAPI

HyperAPI is Hyperbots' document intelligence API for enterprise documents. It provides parse, extract, classify, split and redact behind one endpoint and returns schema-validated JSON with a confidence score on every field. It has a REST API and a Python SDK, with an MCP server for agents available on request, and you can start free. See the HyperAPI Space for a walkthrough.

Part of the Hyperbots synthetic document suite

Dataset	Focus
synthetic-enterprise-doc-extraction	Field extraction across 31 enterprise document types
synthetic-doc-parsing	Layout blocks, reading order, Markdown and tables
synthetic-pii-redaction (this dataset)	PII spans and boxes for masking and de-identification
synthetic-doc-classification	31-class document classification
synthetic-doc-splitting	Multi-document bundles with true boundaries
synthetic-handwritten-forms	Forms filled in handwriting-style script
synthetic-stamped-documents	Stamps and seals over printed text
synthetic-bad-scans	Skew, blur, noise, shadows and compression at 3 severities
synthetic-dense-documents	Small type, multi-column layouts and long tables
synthetic-invoice-extraction	Invoice fields and line items (text + HTML)

License and citation

CC BY 4.0. Generator code is MIT.

@misc{hyperbots2026piiredaction,
  title  = {Synthetic PII Redaction},
  author = {Hyperbots Inc.},
  year   = {2026},
  url    = {https://huggingface.co/datasets/hyperbots-inc/synthetic-pii-redaction}
}