Reading order: how reviewers actually scan a document

The default review screen for a document pipeline shows the page on the left and the extracted fields on the right, in the order they appear on the form. It looks tidy. It is not how anyone reviews.

Reviewers start with the field that decides the case.

On an insurance claim that is the policy number and the date of loss; on a referral it is the patient identifier and the referring provider. Everything else is checked only if those pass.

So the screen leads with them.

We order fields by decision weight, not by position, and we jump the page view to the source region of the field in focus. Low-confidence fields are grouped at the top with the reason they were flagged.

The effect is measurable.

On the Cobalt Mutual engagement, the reordered screen cut median review time per flagged claim from four minutes to under two, with no change to the model at all.