A lot of course material is not really text. Scanned readings, slides exported as images, and handwritten worked examples all arrive as PDFs, but the pages are pictures. Until today Polya brought those files in and then had nothing to say about them, because there were no words to search.
Polya now reads those pages. When a page has no text in it, Polya transcribes that page on its own, so a passage it later cites still points at the right page number. Pages that already have text cost nothing extra and are read the way they always were. Very large scans are handled as one file instead of page by page, so a large scanned reader no longer stalls the import.
Two days ago we fixed a quieter failure. A single malformed character in a PDF, or in a lecture transcript, could make the whole file fail to save, and you would see a source that never finished. Polya now cleans those characters out before saving, so one bad glyph costs you one glyph. Imports that ran long were also being cut off partway; they now pick up where they stopped.
We also walked through Polya as a student would, on a phone and on a laptop, and fixed what got in the way:
- A New conversation button. Before, the only way to start over after a question was to reload the page.
- The question box grows as you type, so a pasted problem stays visible.
- A long course name no longer pushes Study and the options menu off the edge of a phone screen.
- Long links and code wrap inside a chat message, and your own line breaks are kept.
- A cited PDF always has an Open the PDF link, because phone browsers do not show PDFs inside the page.
- If your courses fail to load, you see a retry notice instead of an empty list telling you to bring in your first course.