Document extraction glossary

Definitions

Verifiable document extraction

Document extraction in which every returned value can be traced back to its origin. For scans and photos, a field carries the text it was read from and, when OCR located it, the page and position; for structured e-invoices, the value is parsed from the source file. The point is that a reviewer or a downstream system can check a value instead of trusting it blind.

Source evidence

The proof attached to a field extracted from a scanned or photographed document: the matched text and, when OCR could locate the value, the page and bounding box. It lets someone confirm where a number came from without re-reading the whole page. Structured e-invoices are parsed from their data and carry no bounding box.

Match status

A per-field flag that says how a value was anchored to the source. The values are exact, fuzzy, llm_located, not_found, not_applicable, and user_entered. A value that OCR cannot anchor is returned as not_found rather than silently accepted.

Bounding box

The rectangle on the page image where OCR located a value. DocSolved draws it on the annotated PDF, colour-coded by the confidence scale. A box exists only when OCR evidence located the value; e-invoice fields have none by design.

Structured e-invoice

An invoice delivered as machine-readable data rather than a scan: Factur-X, ZUGFeRD and XRechnung (CII), UBL 2.1, and KSeF FA(3). DocSolved parses the values directly from the file, with no OCR and no AI guessing, so there is nothing to misread.

KSeF FA(3)

The XML e-invoice format of Poland's National e-Invoice System (KSeF). DocSolved reads the FA(3) file itself and validates it deterministically; there is no KSeF API connection or credentials involved.

OCR (optical character recognition)

Turning an image of text (a scan or photo) into machine-readable characters and their positions on the page. In DocSolved, OCR produces the tokens and boxes that later anchor each extracted field to its source.

Confidence score

A per-field score for values that AI extracted from a scan or photo; low-confidence values are highlighted for review. Structured e-invoice fields carry a dedicated trust status instead, because nothing about them was guessed.

Document type detection

Automatic identification of what a document is — invoice, receipt, purchase order, delivery note, CV, and more — with a confidence score. The detected type selects which fields DocSolved extracts.

Annotated PDF

A downloadable copy of the document with each extracted value boxed on the page and colour-coded by confidence, so the whole extraction can be checked at a glance.