What our fictional-shipment test checks

Measured 2 November 2026. These are development tests on fictional data, not evidence of accuracy on customer documents or independent production acceptance.

Calculated figures

We vary the first line's quantity and unit price across 12 fictional shipments. An independent decimal calculation in the test computes each invoice total and net weight directly from the inputs, including freight and insurance. The invoice and packing-list outputs must agree with those numbers.

Planted mismatches

In 20 separate comparisons, we change one party, shipment field, total or item value in a matching document pair. Each comparison must report that difference with both source values. The unchanged pair must produce no false alarm.

What this does not prove

It does not establish the accuracy of AI extraction, real-world input values, origin eligibility, customs acceptance or letter-of-credit compliance. Every draft still needs your review. AU Logistics does not lodge, certify or sign.

Read the measured results (JSON)

Timing and model-cost measurements: 20 fictional runs per task