How this tool computes its result
The checker parses the PDF in your browser with PDF.js and audits each page independently. It measures text-layer coverage, extracts positioned text, identifies larger text runs as heading candidates, flags repeated column-like spacing as table-shaped content, and detects parallel left/right text bands as a possible multi-column reading-order risk. It also reads link annotations, document metadata, top-level outline entries, total words, and estimated tokens. The score deducts points for image-only pages, missing title or language metadata, absent structure in longer documents, complex reading-order patterns, table-heavy layouts, and very large extracted context. These are deterministic file checks: the score does not claim that a model has fetched, understood, recommended, or cited the PDF.
