A silent persistent learning guardian that detects when the user repeats the same conceptual error and intervenes with calibrated questions to break the cycle. Use automatically — no user invocation needed.
