You enroll in a course. The agent picks a problem at your current level. You
solve it in main.go, type
check, and the agent grades
the solution: bumps the concept's level, logs whatever you got wrong, generates
review cards from your mistakes, and schedules the problem to come back later.
$ check
✓ Solved.
Concept hash-maps: L1 → L2
Created 4 review cards.
Re-solve scheduled in 7 days.
You don't peek at the solution. There is no solution to peek at.
You type I don't know. The
agent breaks the problem down: a smaller version, a building block, a
sub-problem you can solve right now. You solve it. Then the next one. Then the
original.
You never accidentally read the answer. You never give up either.
Two things happen. The mistakes you made become
review cards — small
concept-level questions on their own schedule. And the problem itself enters a
Leitner schedule:
7,
21,
60,
180,
365 days.
Seven days later, two-sum-variant
comes back. Your old solution is hidden. You re-solve it from scratch.
Clean re-solve, the next due date pushes further out. Need scaffolding, the step
holds. Two consecutive failed re-solves on the same concept drop its level by
one and the agent reroutes you to easier problems on that concept until it
sticks.