nutrient-extraction-samples

Maintainer checklist

This repository is public. Use this checklist when adding new demos or swapping source documents.

---

Adding a new demo

• Source PDF is public-domain, synthetic, or explicitly licensed for redistribution
• All names, identifiers, and personal values in the document are synthetic or anonymized
• Generator committed, docs.json schema defined, template renders correctly
• Cache seeded (python common/seed_cache.py) so the demo runs without an API key
• output/index.html committed and opens correctly in a browser
• Screenshot added to docs/screenshots/ and referenced in the README
• Per-demo README.md written with a tuning story or usage note
• PII review completed on all derivatives (see Privacy section in README)

Before any --refresh run

• API key is set in .env and is not committed
• Confirm the refresh will not capture real document values (synthetic source only)
• Check that metadata.json output does not include the usage field after the run

Secret scanning

Run against full history before any major release:

gitleaks detect --source . --log-opts="--all" --redact

Extraction API keys are pdf_live_-prefixed. The scanner covers all key shapes; the prefix check is a supplement only.

Known committed values

SSN-formatted value: 000-45-6789 appears in rma_extraction’s metadata.json, cache/*.json, and output/index.html. The 000- prefix is never issued by the SSA — this is a synthetic value from a sample form used for demonstration purposes only.

Billing data: The usage field (remainingCredits, price_composition) was stripped from all committed metadata.json files. The common/ cache layer excludes the usage field on all future runs.

This site is open source. Improve this page.