professorgoose.com• May 20, 2026• 1 min read
Show HN: how I fixed my ai goose tutor to stop punishing understandinga few weeks ago, I built professor goose, a socratic ai tutor built around the feynman idea that if you can’t explain it, you don’t understand it. the way professor goose works specifically is you pick a topic, rubber duck at a 3d goose, and instead of answering it asks follow up questions until it understands.that begs the question, how does the goose know it has understood?that’s when I thought of an understanding bar - always available to the user to help visualize how much the goose understands you, 0 -> 100%.the original logic powering the understanding bar went something like this: every turn, id send the convo to an llm and ask it to return a number 0-100 , with a rubric of brackets to make the output less volatile. 0-10 meant no real understanding. 11-20 named , but empty. 21-35 meant a partial understanding, and so on, up to 93-100 for the goose understanding your topic exceptionally. this approach worked. mostly. until I started looking at what came back once real users teste