The bar you keep missing is usually one beat

There is a kind of practice that feels productive and is not: a loop around eight bars, thirty repetitions, and no improvement at the one moment that keeps going wrong. You are rehearsing your competence. The bar that beats you is in there, surrounded by material you already own, and every repetition spends most of its effort on that material.

Bisect it, the way you would a bug

The method comes from debugging and works for the same reason: the fastest way to find a fault in a long sequence is to cut the sequence in half and ask which half still fails.

1. 1Set the loop wideAround the whole phrase you cannot play. Four repetitions. Note which half breaks down, not precisely where.
2. 2Cut it in halfKeep the half that failed. Four more repetitions.
3. 3Cut againAnd again. Three or four cuts take an eight-bar phrase down to roughly a beat.
4. 4Stop when it stops failingThe last loop that still defeats you is the real problem. Everything wider was scenery.

Most people are surprised how small the answer is. An eight-bar run that has been unplayable for a week usually reduces to one position shift, a fingering that was never decided, or a single note being reached for late.

Settings · audio & playback

1Audio & playback

1aAudio quality1bSkip interval1cCount-in duration

Then widen back out in one move

Once the beat is solid, do not creep outwards a note at a time. Go straight back to the full phrase. If it holds, you were right about the fault. If it collapses, you narrowed onto the wrong beat, which is useful information that took a minute instead of a week.

The usual failure: the beat is fine alone but falls apart in context. That means the problem is the approach to the beat, not the beat. Move the loop one beat earlier and repeat.

Popup · loop and timeline

1Loop A–B

1aA / B points1bLoop on / off1cCount-in1dClear

2Timeline

2aA–B band

Why about a second is right

The count-in replays at every wrap from B back to A, so one attempt costs the count-in plus the loop. At 90 bpm a single beat is two thirds of a second, and the default count-in duration is 3 s — call it four seconds an attempt, or about fifteen a minute. Slowing down does not buy you more of them: the count-in is a wall-clock pause, so at 0.7× the beat itself stretches to about a second and the attempt gets slightly longer, not shorter. Drop the count-in to 2 s and you are back to roughly twenty. That density is the point: corrections stay in short-term memory between attempts, which is not true when a four-bar loop makes each repetition twelve seconds long.

Much shorter than a beat stops working. A half-beat loop gives no musical context, and you end up drilling a fragment you cannot place back into the phrase.

Slow it, but not as far as you think

Speed runs from 0.25× to 3×, and the instinct is to go to the floor. The useful window for learning a part is 0.6× to 0.8×. Below about 0.5× the pulse stops being felt as a pulse and you start playing a series of separate events instead of a phrase with a groove, which is a different skill and not the one you are building.

More on the speed control in slowing a video down without the chipmunk voice.

Where this stops helping

Bisecting finds where the problem is. It does not tell you what the problem is, and it will not help if the answer is a wrong fingering or a tempo beyond you today. If four cuts land on a beat that stays impossible at 0.6×, the loop has done its job and the answer is a different fingering.

It will not help with stamina, tone or intonation either. Those need long takes, not short loops.

A twenty-minute version