Rubik's TrainerCFOP practice for sub-20 solvers

Expected times#

A solve goal like "sub-20" is not a single number you practise. It is four numbers that add up. This page says what those four numbers are for each goal, where they came from, and — just as importantly — how much you should trust them.

The trainer uses them for exactly one purpose: to turn a measured case time into a score. Hitting the phase target for your goal maps to 80; hitting half the target time maps to 100.

What this app ships#

These are the values in targets.splits in app/config/config.yml. Seconds, per phase, including recognition.

Goal Cross F2L OLL PLL Total
Sub-30 4.0 15.0 5.5 5.5 30.0
Sub-20 2.5 9.5 4.0 4.0 20.0
Sub-15 2.0 7.0 3.0 3.0 15.0
Sub-12 1.5 5.5 2.5 2.5 12.0
Sub-10 1.2 4.6 2.2 2.0 10.0

As proportions of the whole solve:

Goal Cross F2L OLL PLL
Sub-30 13% 50% 18% 18%
Sub-20 13% 48% 20% 20%
Sub-15 13% 47% 20% 20%
Sub-12 13% 46% 21% 21%
Sub-10 12% 46% 22% 20%

Your instance may differ — an administrator can change any of these, add goals, or remove them. See Configuration.

The honest caveat#

There is no authoritative published table of CFOP split targets. I went looking for one and it does not exist. What exists is:

  • Feliks Zemdegs' CFOP Solve Splits Tool, a spreadsheet that compares your splits against averaged data for solvers between 60 and 8 seconds — the closest thing to a reference dataset, but the underlying table is not published as prose;
  • scattered self-reported splits on the speedsolving forums;
  • the Speedsolving wiki's progression guides, such as Road to Sub-15 Second Averages, which say which techniques to learn at each level and deliberately give no per-step time targets.

The CubeSkills article makes the point directly. Two solvers can both average 20 seconds with completely different profiles: it gives 4-11-2-3 (a solver with full OLL/PLL but inefficient F2L) versus 2-9-4-5 (a solver with good F2L and lookahead but a four-look last layer), and calls the second "a vastly different profile."1 Its recommendation is not "aim for these splits" but:

Based on my current splits of A-B-C-D and the current techniques and algorithms that I have learned, which step(s) of my solve can be improved most easily?

So: treat the table above as a yardstick, not a prescription. Its job is to give the scoring model a stable denominator so that "you got 20% faster at this OLL" is a meaningful statement. It is not a claim that your splits are wrong if they do not match.

What real splits look like#

Data points I could actually verify, for calibration:

Source Overall Cross F2L OLL PLL
CubeSkills "solver A"1 ~20 4 11 2 3
CubeSkills "solver B"1 ~20 2 9 4 5
Forum self-report, sub-19 avg2 18.6 2.50 10.49 2.43 3.21
Forum self-report, ~25s avg3 ~24 5.0 11.3 3.75 4.0

Two things fall out of that spread:

F2L is the longest stage by a wide margin. In every real profile above it is 45–56% of the solve, and Wikipedia's CFOP article states plainly that F2L "is the longest stage of the method."5 The shipped targets put it at 46–50%, at the low end of the observed range. If you are going to miss one target, miss this one.

The last layer is where the variance is. Solver A spends 5 seconds on the whole last layer; solver B spends 9. That gap is almost entirely how many algorithms you know: full OLL is 57 cases and full PLL is 21, versus 10 + 6 for the two-look versions.5 Learning algorithms buys last-layer time directly and predictably. Nothing else in the solve improves that mechanically.

Where to actually spend your effort#

The forum advice is consistent and it is not about the last layer:

  • Cross first, if yours is slow. A cross can always be solved in eight moves or fewer,5 and the standard advice to a 25-second solver with a 5-second cross is to get it to 2–3 seconds — planning the whole cross in inspection, aiming for around seven moves.3 J Perm puts the practical ceiling at "less than 5 seconds" with the right practice.4 A slow cross is the cheapest fix in the solve because it costs no algorithm learning at all.
  • Then F2L, and it is not about turning faster. F2L time is lookahead time. Rotationless insertions, multislotting, and tracking the next pair while you finish the current one are what move it; drilling the 41 cases individually (which is what this app does well) fixes recognition, but you still have to practise the transitions in full solves.
  • Full OLL and PLL somewhere around sub-25 to sub-20. The community consensus is that people approach advanced cross, advanced F2L, full OLL and full PLL around sub-30 averages.4 Below sub-20, full OLL is described as "an ongoing project most cubers finish on the way to sub-15."[^ss-sub15]

The 10-second rule of thumb

If your F2L is more than about 55% of your solve, work on lookahead. If your last layer is more than about 40%, learn algorithms. If your cross is more than about 20%, plan it in inspection. Those three sentences cover almost every intermediate solver.

How the trainer uses these numbers#

Two places, and only two:

  1. Case scores. A case belongs to a phase. Your trimmed mean time for that case is compared against a per-case share of the phase target — an OLL is compared against the OLL target, since a single OLL execution is the whole OLL phase. Target time → 80. Half the target time → 100. See Assessment.
  2. The projected solve time on the assessment overview, where a phase you have no data for falls back to its target.

Your goal is per profile, not per instance, so a coach can keep a sub-30 and a sub-12 student on the same machine and both see sensible colours.

Measuring your own splits#

Once you have enough full solves, your own data beats any table. The timer records phase splits: tap (or press the spacebar) at the end of the cross, the end of F2L, and the end of OLL, and the solve is stored with cross_ms, f2l_ms, oll_ms and pll_ms alongside the total.

Practical notes:

  • Splits include recognition. The time you spend looking at the last layer deciding which OLL it is belongs to OLL. That is how every source above measures it, and it is why a "2-second OLL" is a genuinely fast OLL.
  • Split by hand for the first twenty solves and then stop worrying. You need the shape, not precision. Ten to twenty splits tell you which phase is your problem; after that, the phase averages on the overview keep updating on their own.
  • Tapping costs you time. Expect your split-tracked solves to run a few tenths slower than your untracked ones. Compare splits against splits.

Cross has no case library

Cross is timed and scored as a phase, but it has no algorithm set — it is solved intuitively. You will see it in the timer, in the phase targets and on the spider graph, but not in the library.

References#

  1. Feliks Zemdegs, CFOP Solve Splits Tool, CubeSkills. The tool covers solvers between 60 and 8 seconds; faster solvers use extended crosses and other techniques that make a clean four-way split hard to measure. ↩↩↩

  2. User dodecicosidodecahedron, "Strange Splits (Sub-19 CFOP)", Speedsolving forums — averaged over roughly a dozen recorded solves, recognition included. ↩

  3. "Is my F2L time OK for my level?", Speedsolving forums. ↩↩

  4. J Perm, CFOP Speedsolving Method. ↩↩

  5. CFOP method, Wikipedia — 41 F2L, 57 OLL and 21 PLL algorithms for the full method; 10 + 6 for two-look OLL and PLL; the cross "can always be solved in 8 moves or fewer". ↩↩↩

Source: docs/user/expected-times.md