Practicing the passage you cannot play
The four bars that never improve are usually two beats you have never isolated. How to find the difficulty, work it under tempo, and put it back without it collapsing.
There is a passage in the piece you are learning that has not improved in three weeks. You know exactly where it is. You brace two bars before it arrives, it comes apart in roughly the same place every time, you carry on to the end, and tomorrow you do the whole thing again.
Playing the piece from the top is not practice on that passage. It is a performance with one bad moment in it, repeated daily until the bad moment is the most rehearsed thing in the piece. What follows is the alternative. It is the same method on any instrument and it is deeply unglamorous.
The difficulty is smaller than you think
You have been calling it "bars 34 to 37" because that is the region where it goes wrong. Bars 34 to 37 is not the difficulty. The difficulty is usually two beats: one string crossing, one finger substitution, one leap the hand has never actually measured. The rest of those four bars is fine, and you have been practicing the fine part four times a session for a month.
So go and find the two beats. Play one beat before the join and one beat after, and nothing else. If that fragment is easy, move the window over by a beat and try again. Somewhere in there is the moment the hand does not know what to do, and it is shorter than the passage you have been blaming.
This is worth doing slowly and in writing. Put a bracket in the score around the beats that actually fail. Most of the value of the next twenty minutes comes from getting that bracket in the right place.
Work it at a tempo where it is easy
Not "slower". Easy. If the fragment is at 92 and it fails one time in four, 72 is not the answer, because one time in four is still a rehearsal of the failure. Halve it. If it is still not easy at 46, halve it again and stop feeling silly about it. Nobody is listening, and a tempo you can play accurately is the only tempo that teaches your hand anything.
The test is not whether you get through it. The test is whether you knew, before each note, what the hand was about to do. If you are finding out at the same moment the note sounds, you are still too fast.
Five correct repetitions in a row, and a mistake resets the count
Four good ones and a scrappy fifth is not five. Four good ones and a scrappy fifth followed by "well, the fifth was nearly right" is how a passage stays broken for a month. Count out loud, or count on your fingers, or make five marks in pencil in the margin and cross them off. The counting matters because the honesty is the hard part, not the playing.
Five is not a magic number. It is small enough that resetting to zero costs you about a minute, which is what makes the rule survivable.
Four clicks, then stop when the quality drops
Once you have five in a row, put the metronome up by four clicks and go again. Five more in a row at the new tempo. Then four clicks more.
You will hit a tempo where the quality drops. Stop there, in the middle of the session, even if that is two increments in and you have been at this for six minutes. The session ends when the passage stops being correct, not when the clock says twenty minutes. Continuing past that point is how you spend a half hour teaching your hand the wrong movement with great commitment.
There is a metronome at /tools/metronome that needs no account, and it will do the counting while you do the honest part.
Tomorrow, start twelve clicks below where you finished
You will not keep yesterday's top tempo. Nobody does. Starting twelve clicks below where you stopped gives you three or four easy increments before you are back at the edge, and those easy increments are not wasted time. They are the part where yesterday's work gets confirmed rather than gambled.
Over a week this looks like: 60, 76, 68, 84, 76, 92. The line goes up and it goes down, and it is still going up.
Rhythm variations, when it is nearly there
When the fragment is reliable but uneven, change the rhythm rather than the tempo. Play it dotted: long-short, long-short. Then reverse it: short-long. Then group it in threes against whatever the written grouping is. Each variation puts the fast notes and the pauses in different places, so no single set of fingers gets to coast on being habitual.
If a variation is a mess, that is where the unevenness lives. That is now the two beats.
Putting it back
Rejoin the passage to the piece from further back than feels necessary. Start eight bars early, at the fragment's tempo, and play through it and four bars out the other side. The failure usually starts before you hear it: a rushed bar 30 sets up a hand position that cannot get through bar 35.
Then leave it. One pass in context is enough. If it holds at the reduced tempo in context, you have done the week's work on that passage, and playing it ten more times at performance tempo to see whether it breaks is not practice either. It is checking, and checking is what the run-through at the end of the week is for.
The short version
- Find the actual difficulty. It is usually two beats, not four bars.
- Work it at a tempo where it is genuinely easy.
- Five correct repetitions in a row. A mistake resets the count.
- Raise the metronome by four clicks. Stop the moment quality drops.
- Next day, start twelve clicks below where you finished.
If that list looks familiar, it is one of the example assignments that ships with MakesPerfect, written the way a teacher would write it into a student's notebook. It works because it is small, it is countable, and it removes the option of getting through the passage on luck.