More reliable invoice and receipt parsing

Receipts that include an invoice number keep their receipt classification. When the fast model is uncertain about a document’s type, a larger model checks it. Long invoices and receipts can use source-backed headers without hitting unused reference-candidate limits; uncertain financial facts still need review.

Invoice and receipt fields can be parsed before you choose a contract. If OCR cannot confirm a PDF’s readable text, its parsed fields remain available for review and recording stays blocked. Complete, verified documents still need the ordinary contract, ownership and recording checks.