TheoremPath does not just mark an answer right or wrong. It records the target, the attempted question, the feedback, the gap, and the next checkpoint.
The replay below shows one deterministic loop across a matrix shape ledger, evaluation, concentration, learning theory, and an ML systems target.
Checkpoint: ML foundations checkpoint. The examples are intentionally compact: they show the breadth of the system without turning this page into a lesson.
Step 1: Target
Start from the thing the learner wants to understand
feedback
Question
Target: read a finite-class generalization proof, then connect the same shape discipline to transformer and KV-cache systems work.
Learner answer
I want the shortest reliable path.
Feedback
The run starts by fixing the target. That target determines which checks matter and which topics can wait.
Gap detected
No gap yet
Next checkpoint
Transformer shape and KV-cache ledger
Locked until the learner clears the shape ledger and the probability bridge.