A small discrepancy. A decision you can inspect.

Follow the example in four steps.

1. Check the source. The initial sample contains 60 cartons of Product A and 40 of Product B, but its printed total is 98.
2. Resolve the inconsistency. The line quantities total 100. Use the matching sample to see a source with a printed total of 100, or edit the review fields to try your own synthetic values.
3. Review and approve. Matching counts enable the approval checkbox. Changing a quantity or switching samples clears an earlier approval.
4. Export locally. An approved, matching sample can be downloaded as CSV. The export is generated in the browser; these quantities are not uploaded to a server.

What has been checked?

The validation component is evaluated against 12 declared synthetic cases. In this version, 12 of 12 outputs match the expected outcomes. Cases cover matching and mismatched totals, missing fields, negative and fractional quantities, malformed values, whitespace, zero, scientific notation and the demo’s upper limit.

These results describe deterministic quantity checks only. They do not measure OCR accuracy, AI extraction correctness, document authenticity, physical receiving, production approvals, integration performance or customer savings.

Inspect the complete test-case JSON

How does a real pilot build on this?

Agree representative documents and expected fields, evaluate capture, define business checks and staff review rules, then connect one approved output destination. A production deployment must separately specify reviewer identity, access controls, approval history, monitoring and recovery behavior.