The upload now knows what a good note looks like

The upload screen used to accept whatever it was given and say nothing. That sounds harmless until you look at what came out of it: notes titled "Notes", descriptions of four words, and files whose type the app had guessed from the label rather than from the file. Those notes are effectively invisible — nobody searching for anything will ever surface a document called "Notes" — and the person who uploaded it has no way of knowing that, because at the moment of uploading, the app told them everything was fine.

So the upload now coaches instead of just accepting. It asks for a description in the range that actually works, tells you what it is doing to your file while it does it, and warns you before you upload something the library already has. The API scores every note for how findable it is and stores that score, which means the ordering on browse pages can put well-described notes in front of poorly described ones instead of sorting purely by date. The score is shown to the person who wrote the note and to nobody else — it is there to help you improve your own upload, not to rank students against each other in public.

Underneath, the app stopped guessing what kind of file it was holding. It had been reading the file kind from a label rather than from the file itself, which is why some documents opened to a blank viewer: the app was confidently handing a file to a renderer that could not display it. It now decides from the file, and refuses to hand anything to the viewer it cannot actually render, showing a download instead of a white screen.

The help centre moved out of the app binary and into the database. Previously every correction to a help answer needed a new app release and a store review to reach anybody. Now the answers are served, cached for offline reading, and can be fixed the same afternoon somebody points out that one of them is wrong.