How it works
Built around one idea: you understand what you can rebuild yourself.
Every quest follows the same loop, and every piece of it is generated, tested and checked before it reaches you.
The loop, idea by idea
- 1Situation
What you doA concrete scenario anchors the whole idea: a coin, a channel of water, a search query.
WhyIdeas stick to situations. The same scenario carries through the quest, so understanding builds instead of resetting.
- 2Predict
What you doBefore the lab unlocks, you commit to a guess about the situation and say how sure you are, or say you’re not sure yet.
WhyCommitting makes the surprise visible. A wrong prediction you can see is one of the strongest moments for learning.
- 3Explore
What you doYou push on a live model: drag a point, change a rate, step an algorithm, run code.
WhyDoing is how intuition forms. Every control changes something real in the model.
- 4Observe
What you doYou describe what happened in your own words, then the idea is explained.
WhyExplaining it yourself first, then comparing, is what turns an experience into a mental model.
- 5Check
What you doA short question on a fresh case, graded on our servers. Two tries, then the reasoning.
WhyRetrieval on a new case shows whether the idea transferred, not whether you remember the screen.
- 6Transfer
What you doYou apply the idea to a new situation, with feedback from the mentor if you want it.
WhyUsing an idea somewhere new is the clearest sign you understand it, and it makes it last.
- 7Go deeper
What you doAt any point, open the idea another way: /equation for the derivation, /code to run it, /diagram for the structure, /experiment for a new lab, or /test for five or six graded questions.
WhyPeople understand through different doors. Each view is built for the step you’re on, and test results count towards your progress.
At any point you can open the mentor for a hint scoped to where you are, or ask to see the idea another way: as a diagram, an equation, code or a different experiment.
The Lagrange engine
Plan, compose, verify, review.
Quests are built progressively. Your route appears first; then each step is built, and the next one is prepared while you work on the current one, so you never wait for a whole course. Several AI models do the work, each on what it does best: one builds the lab, another reviews it.
Plan
Your question becomes a small map of the concepts involved and how they depend on each other. From it, a route of three to five ideas in an order that builds, each with a clear outcome.
Compose
For each idea, the engine designs an experiment first: the insight, the misconception a smart beginner holds, the surprise that breaks it, and the controls that create it. Then it composes a live lab from a library of visual and computational building blocks.
Verify
Every lab is run before you see it. Every expression has to work, the challenge must be unsolved at the start and solvable by its own solution, and the misconception must actually appear in the model.
Review
An independent review checks accuracy and fit with your question. Problems with a clear fix are repaired automatically; anything else is regenerated.
Feynman School is AI-generated, and checks reduce mistakes rather than eliminate them. If something looks wrong, ask the mentor or tell us. Every model also lists its assumptions and limits, in “The model behind this quest”.
Evidence, not badges
We keep what you did. We never say you’ve “mastered” it.
Clicking through isn’t learning, and a badge can’t tell the difference. So we keep the evidence instead: what you predicted, what you saw when you pushed on the model, what you explained, and how your checks went.
Every quest and lab you explored, with your prediction, what you saw and the takeaway, ready to revisit.
Questions on fresh cases, graded on our servers: whether you got them right first time.
Spaced review. Ideas come back at growing intervals, because remembering something just as you’re about to forget it is what makes it last.
See it for yourself.