Usage¶
Send a supported XML or PDF document and inspect every returned field against the source. Keep monetary values as decimal strings.
Request¶
Use raw application/xml, raw application/pdf, or JSON:
{"document_base64":"<canonical-base64-document-bytes>","content_type":"application/pdf","country_hint":"MX"}
Replace the illustrative placeholder with strict canonical base64. XML can also be a UTF-8 string. Optional country_hint accepts MX or BR; contradictions reject. JSON instances preserves input order and per-record errors. JSON Lines accepts one record per line.
Bounds and concurrency¶
| Bound | Limit |
|---|---|
| XML | 1 MiB decoded |
| 4 MiB decoded | |
| HTTP body | 6 MiB |
| Records | 16 |
| PDFs per request | 2 |
| PDF pages | 3 |
| Active PDF workers | 1 |
Queue PDF work or retry pdf_busy with bounded backoff. Unsupported single inputs return 415; invalid/resource-limited documents return 400. Batch responses retain ordered per-record errors. Examples include pdf_page_limit, pdf_time_limit, unsupported_pdf_active_content and unsupported_or_ambiguous_pdf_layout.
Response¶
The output schema defines canonical fields, nullable values, decimal strings and provenance. Parse strings with a decimal type, not binary floating-point arithmetic. XML precision is preserved; supported PDF separators are normalized without inventing a balancing amount. Unknown currency and unreadable/conflicting values remain missing.
Every PDF requires source review, including identifiers and amounts with high recognition confidence. Unreadable rows may be omitted with review reasons. No external OCR service, taxpayer lookup or runtime download is used. Protect both supplied documents and extracted outputs.
See limits and evaluation for supported layouts and retained errors.
Last updated November 15, 2026 · Suggest an improvement