How should these claim forms be classified?

A scanned page is unstructured: the amounts and addresses are there for a human to read but are not addressable as named fields, which is why document processing exists - it takes unstructured content from documents and turns it into structured data, meaning specific fields suitable for a database. Until that extraction runs there is nothing for a query to average by region. Calling it structured mistakes a consistent template for a schema; a shared layout does not make a field queryable. Semi-structured is the tempting near miss, but formats like JSON and XML carry machine-readable tags around their values, and a scan carries pixels. Labeled versus unlabeled is a different axis altogether, about whether outcomes are marked.