Getting to sub-20
Where the twenty seconds go, what to measure, and why the answer is almost never that your hands are too slow.
By the end you will be able to
- Break your solve into four splits and say which one is costing you
- Tell a lookahead problem, a recognition problem and a turning problem apart
- Build a practice week that works on one thing at a time
- Say honestly which algorithms sub-20 requires and which it does not
A twenty-second solve is about sixty turns. Sixty turns in twenty seconds is three turns a second, which is an unhurried pace — you can watch someone do it and think they are being careful. Nothing about sub-20 requires fast hands.
What it requires is that the twenty seconds are nearly all turning. That is the whole problem, and every measurement below is a way of finding out where they are not.
The budget
| Step | Turns | Time | What it looks like |
|---|---|---|---|
| Cross | about 6 | 2 s | Planned in inspection, run without looking at it |
| F2L | about 28 | 10 s | Four pairs, no pause between them, at most two rotations |
| OLL | about 10 | 3 s | Case read during the last insertion |
| PLL | about 13 | 4 s | Corners read as the OLL algorithm finishes |
| Alignment turns | about 3 | 1 s | The turns nobody writes down |
What to measure
Averages tell you whether you are improving. They do not tell you what to do next. Four measurements do.
- Splits. Stop the timer when the cross is done, then again after F2L. Do it ten times and average each. Nothing else you measure is as useful as knowing your F2L time on its own.
- Pauses over half a second. Film one solve on a phone and count them. Most people at thirty seconds have six or seven; a sub-20 solve has one or two.
- Rotations. Count them in the same video. Five or six means the F2L is being solved wherever the pieces happen to be rather than where you decided.
- Turns per second. Count the turns in the video and divide. Almost everyone finds this number is already high enough, which is the useful part — it tells you to stop trying to move faster.
| You stop here | It is really | What to do about it |
|---|---|---|
| Before the first turn | The cross was not planned | Inspection drills: cross, then cross plus one pair |
| After the cross | No first pair planned | Track one corner through the cross during inspection |
| Between pairs | Lookahead | Slow solves at two turns a second with no pauses allowed |
| Halfway through a pair | You started it without a plan | Decide the whole insertion before the first turn of it |
| After the last pair | OLL recognition | Recognition-only drills, and read the case during the insertion |
| After OLL | PLL recognition | Learn to read the corners as the OLL algorithm finishes |
| After PLL | The alignment turn | Drill finishing: stop the clock only when the cube is solved |
A practice week
Practice that is all timed solves stops working somewhere around twenty-five seconds. The mix below is about forty minutes a day and it keeps one thing at a time under repair.
- Twenty slow solves. Two turns a second, no pauses permitted. This is the lookahead drill and it is the most valuable thing on the list.
- One average of twelve, timed properly. With inspection. This is your measurement, not your practice.
- One PLL time attack. All twenty-one in a row, against the clock.
- One OLL group. Whichever group you are learning, or whichever one the recognition list says you are slow on.
- Ten cross-plus-first-pair drills. Fifteen seconds of inspection, then cross and one pair, then stop.
- One rest day a week. Fingers and recognition both consolidate overnight, and a week with a day off in it beats a week without one.
What sub-20 feels like
It is quieter than it looks from below. The last layer stops being the exciting part and becomes a rest — two algorithms your hands run while your mind is already finished. The F2L stops feeling like four separate problems. And the solve stops having a moment where you look at the cube and think.
The next barrier is at about fifteen seconds, and it is a different animal: that one does want full OLL, and it wants F2L solutions chosen for what they leave behind rather than for what they cost. Before any of that, there is a good case for staying at twenty for a month and making it consistent, because an average of twelve at nineteen seconds with no solve above twenty-two is a much better foundation than one that swings between fourteen and twenty-eight.
Averages of five and twelve The numbers competitions use, and the only ones worth judging a fortnight of practice by. Drills, hardware and habits Short specific things to try, sorted by what you are working on. What comes after CFOP COLL and Winter Variation, Roux and ZZ, and the point at which you stop looking algorithms up.