Skip to content

Glossary

What is OCR grounding in document AI?

OCR grounding constrains document extraction to the text actually recognized on the page: if a value is not in the OCR output, the model must leave the field empty rather than infer it. For legal documents like CMR consignment notes, this is the difference between a blank a human fills and a fabricated party name that gets filed.

In freight operations

Ungrounded extractors score well on naive metrics — fewer blanks — precisely because they invent. Grounded extractors gate uncertainty to a human instead.

4logist AI measures its document reader on grounding violations, not blank counts, and chose its model accordingly.

How it relates to 4logist AI

4logist AI is built for SMB freight forwarders who want AI drafts with human approval. Glossary terms below link to agents and connectors when they map directly to product surfaces.

Frequently asked questions

Why is a blank field better than a guessed one?

A blank gates the document for human review. A plausible guess passes review and gets filed — turning an extraction error into a legal-document error.

Related