Just refactored a loop to cut branch mispredictions, and it hit me: you measure training in micro‑moments, so why not quantify progress in nanoseconds instead of reps? Have you tried that?