Technology1 min read

Show HN: Claimseed, synthetic Medicare docs with planted errors and answer keys

By · Published by Everything Blog

In short

Software engineers evaluating automated document review platforms often face severe privacy hurdles when testing models against sensitive real-world healthcare records. To solve this friction, the new utility outputs artificially constructed billing statements equipped with verifiable defect maps, eliminating patient data liabilities. The approach empowers developers to thoroughly benchmark automated agents and data extraction pipelines under realistic operational conditions. Specifically, release version 0.6.0 offers 1,022 blind documents across 34 distinct scenarios, including Medicare Summary Notices, hospital bills, and commercial explanations of benefits.

Key points

  • Engineering teams can rigorously stress-test extraction models and billing agents without…: Engineering teams can rigorously stress-test extraction models and billing agents without exposing genuine patient personal health information.
  • Developers can deploy the software as a command-line tool, an ECMAScript library, or a Mo…: Developers can deploy the software as a command-line tool, an ECMAScript library, or a Model Context Protocol server.
  • Built-in canary strings protect machine learning benchmarks by preventing reference answe…: Built-in canary strings protect machine learning benchmarks by preventing reference answer keys from leaking into training corpora.

Sources checked by our analysis pipeline: 1

Seeded, fully synthetic U.S. medical-billing specimen documents — Medicare Summary Notices, hospital itemized bills and Medicare Advantage explanations of benefits — rendered to PDF with ground-truth JSON that says exactly which billing errors were injected and where. For testing document extractors, bill-review tools and agents without ever touching a real person's paperwork.

- Every name, identifier, address, phone number and amount is generated from a seed. Nothing in this repository

is, or is derived from, a real document; see

docs/synthetic-data-policy.md

. - Every rendered page says so: a specimen band (text, in the foot margin), a diagonal watermark (drawn as shapes, so it never leaks letters into text extraction) and PDF metadata, stamped into the finished PDF by the renderer itself — no CLI flag, library call or data object switches them off.

- Same seed, same document: all content is derived from the seed, fonts are bundled and PDF metadata is fixed, so two renders on the same Chrome build (Linux, as in CI) are byte-identical; other platforms produce the same content, with bytes that may differ by Chrome…

Technology