Validate incoming data before accepting it

A technically readable message is not necessarily a valid business record. This applies both to API responses and to information extracted from documents. Formats, allowed values and relationships need validation before acceptance.

Separate the validation stages

• Check format, character encoding and required fields first.
• Validate values against business rules and existing references.
• Route uncertain extraction results to targeted human review.
• Preserve provenance and corrections without distributing unnecessary raw data.

A date can be syntactically valid but inconsistent with its business operation. Likewise, a recognized product identifier becomes usable only when its mapping is unambiguous. Silent correction after import can conceal these issues.

Acceptance testing should therefore include contradictory, incomplete and duplicate inputs alongside valid examples. The application must show what it accepts, what it rejects and where a person needs to decide. This makes data quality a visible part of the workflow.