Use the two-failure reset rule: `/clear` and restart after correcting the same issue twice

SmartScope·low signal

Context contamination is self-reinforcing — once a session has a failed approach in its window, repeated corrections accelerate failure recurrence rather than fixing it. The operational threshold practitioners use: if you're correcting the same issue a second time, a `/clear` and clean restart is faster than a third correction. Checkpoints make this cheaper — you can roll back only the conversation while keeping the code (or vice versa) when the discussion was valuable but the implementation approach was wrong.