The numbers you need are in a table on a page with no export button, and retyping them by hand invites errors.
Table extraction takes a capture of a table and returns structured CSV or Markdown. Because the input is pixels rather than the DOM, it works on tables that scrapers cannot touch: canvas-rendered dashboards, images, and PDFs rendered in-page. The model reconstructs rows and columns, including headers.
Take a region capture over the table, run the extract-as-table action, and paste the result where you need it. If a column comes back misread, capture a tighter region with larger text; images are downscaled to a 1,568 pixel long edge before being sent, so bigger text survives better.