Why you should time your algorithms
Almost every Speedcuber chooses their algorithms and solutions purely based on feel
Some common examples of this include, but are not limited to:
1. Doing OLLCP because EPLL algs "feel" nicer than "slow" diag PLL's
2. Avoiding Dot-OLL's absurdly often because they "feel" slow
3. Avoiding left hand dominant algorithms because they "feel" slow
4. Avoiding algorithms with high move count because they "feel" slow
5. Avoiding algorithms with LOW move count because the fingertricks "feel" weird (and also slow)
And guess what? I'm willing to bet that quite literally everyone in the top 100 for 3x3 average is currently doing a number of alg and solutions for cases that are actually slowing them down, for no other reason than because it "feels" better
Ever since I first started timing my basic F2L, OLL, and PLL algorithms more, I've realized that I need to revamp quite a lot of my solving style
Some algorithms that I thought were helping me are actually slowing me down
Other ones that I thought were gross and slow that I put aside, actually would be HELPING me quite a lot if I were to use them...
Now, what does this mean for you?
Well, it could mean a few things
Maybe you haven't pushed your PLL's to the absolute limit. Your U perms and H perm are feugo, and maybe you have some sick J perms...
...but maybe your G perms are hot garbage, and you also ought to replace that nasty N perm alg you're using
Maybe the algs are fine, but your fingertricks are holding you back because you’ve opted for a rudimentary turning style instead of incorporating some useful higher risk fingertricks like pinch U turns, OH style U layer pushes/flicks and D layer pushes, home grip index F/F’ turns etc
Maybe you avoid Dot-OLL's by doing silly stuff like R' D' r U r' D R2 U R', even though the time difference between R' D' r U r' D R2 U R' and a simple R U R', is MUCH GREATER than the time difference between a PERFECTED Dot-OLL and an ugly/slow Pi COLL case (or even a good OLL/COLL case for that matter)
Speaking of COLL, maybe that Pi COLL case you're using is way slower than just doing R' U2' R2 U R2' U R2 U2' R' (which I firmly believe is faster than R U2’ R2’ U’ R2 U’ R2’ U2’ R) every time, because your diagswap PLL's really aren't all that bad after all (Again, MY Vperm is literally better than my Zperm, and my Uperms are only .3-.5 of a second FASTER than my Nperms...)
Solving by feel is a natural part of the process, and I'm certainly not saying you should NEVER solve by feel
Far from it. Trusting your gut is a very useful skill to cultivate and has many benefits in regards to your mindset when you compete
Plus it's quite hard to know how fast an algorithm will be for you before you've really tried practicing it
However, if you REALLY want to make strides and break plateaus...
RUN. THE. NUMBERS.
Find out the numbers. Specifically, the numbers for how fast you can do x y and z algs
Run those numbers, and then run them down lower and lower by learning new algorithms, new fingertricks, even sometimes new ways of solving if you're overcompensating with silly EO tricks and OLLCP algorithms that just don't help ;)
Now, I would just end it here, but, there IS one final detail that I believe is quite important when you start “Running the numbers”
Don’t just use best single as your number
This is a mistake I’ve made myself, and it’s a very easy one to make
Instead, you want to use some sort of small average like avg5, mo3, or simply timing yourself doing the algorithm 3 times in a row before stopping the timer
Doing so will prevent you from making your alg choices based on best single, which can very easily lead you to using high risk, low reward algorithms
Being able to get a .6x for a random OLL is great, but if you only get .6x after 100+ attempts when there’s an algorithm you can consistently get .8x avg5’s with, then are you really saving time by using the former algorithm?
Probably not!
That’s not to say that I’m entirely against high risk algorithms, but unless you’re messing up because you haven’t practiced the case enough, or it involves a fingertrick that you could clearly improve at, then from a risk to reward standpoint it really doesn’t make sense to use a riskier algorithm in favor of a more consistent one
As such, I’m a big fan of avg5, mo3, and doing the alg 3 times in a row and timing that instead of timing the alg once (the 3 in a row one is also a very easy way to account for home grip most of the time, but it can be accounted for with the other timing methods as well if you pay attention to it, and also can be messed up with something like the x’ RUD Eperm for example)
Avg12 could work as well, but to be honest I feel like a sample size that large would prioritize safety way too much. On top of that, it’s also pretty time consuming and time that’s probably better spent just doing solves and learning new trick (aka the things that move the needle more than anything else over time)
Whatever you end up doing, just don’t choose your algorithms purely based on feel and nothing else