When AI fills regulated passport fields from uploaded documents — datasheets, test reports, declarations — not every extraction is equally certain. A clear GTIN read off a label is high-confidence; a recycled-content percentage inferred from an ambiguous paragraph is not. The confidence score makes that uncertainty visible per field rather than presenting every value as equally trustworthy.
This is what makes AI extraction safe for compliance data, where a wrong figure carries legal weight. Paired with source attribution — the exact document and passage each value came from — the confidence score lets a reviewer focus their attention: approve the high-confidence fields quickly, scrutinise the low-confidence ones, and never publish a number they can't trace. In TracePass a human always approves before a passport goes live.
Frequently asked
Does a high confidence score mean the value is guaranteed correct?
No — it indicates how sure the AI is, not a guarantee. That is exactly why a human review step remains mandatory: the score directs attention, but a person approves every regulated field before publication, so accountability stays with the economic operator.
How does a confidence score relate to source attribution?
They work as a pair. The confidence score says how certain the AI is about a value; source attribution shows where the value came from — the exact document and passage. Together they let a reviewer verify any field quickly instead of re-keying it from scratch.