Let me first state that I am not anti-Colemak. If I'm anti-anything in respect of keyboard layouts, I'm most definitely anti-QWERTY. It's clear that Colemak is superior to QWERTY (let's face it, almost any rationally-designed layout is).
What I'm having trouble with is the assertion that Colemak is easier to learn than, say, Dvorak (note that it's https://colemak.com/Easy_to_learn that itself makes several comparisons with Dvorak, almost inviting counter-argument).
Apart from the shortcuts issue, which I agree is a valid point (although it can be addressed in other ways), the main justification for the 'easier to learn' claim appears to me to be due to the assertion that "Memorization isn't linear: fewer changes from the de-facto standard improves the ability to switch from it."
It seems to me that this assertion, that Colemak>Dvorak because "Colemak moves only 17 keys, while Dvorak moves 33 keys," while logical-sounding, is flawed. A layout has to be learnt as a whole - a gestalt. It's inappropriate to make the comparison with learning a linear sequence of characters, since each key in the layout has a dynamic relationship with every other key.
Even minor changes can throw things out of kilter. Consider the 'a' and 'o' keys on the Dvorak layout. Michael Capewell points out that these two letters should be switched, and I tend to agree, as I personally find it troublesome typing 'inboard stroke flow' and 'keyboard' on the ANSI Dvorak. (I'd love to know why Dr. Dvorak apparently broke the 'inboard stroke flow' rule on these two).
As an experiment, I tried switching just these two characters around on my own (ANSI Dvorak) layout. Immediately, my typing speed dropped by a third, while my error rate soared by a factor of three. These two characters had only moved a single position (cf. "The keys in Colemak are just a key or two away from their QWERTY position"), and yet this simple change caused chaos, with knock-on effects as my fingers, that had learnt certain combinations, now fumbled almost constantly, and the comfortable flow was completely lost.
I didn't persevere with the revision; I do acknowledge that it would take me less time to learn the revision than it did to learn Dvorak itself - but I'm only talking about two adjacent keys here, and it was clear that this apparently minor change had a substantial knock-on effect.
This experiment convinced me of two things:
1. Fine-tuning a '90% perfect' design isn't worth the effort (whereas switching to ANY sensibly-designed layout from QWERTY definitely is)
2. If there's this much difficulty with simply swapping the positions of two *adjacent* keys, then it's going to be no more difficult to learn a new layout that moves half the keys than one that moves all of them.
In a similar vein, it's claimed that the "hardest keys to learn are those that move between hands" (with another Dvorak comparison). This does not seem to me to hold water: I may be wrong (I quite often am). However, justification for this assertion would enhance credibility.
I hope that this post invites debate on this topic - if indeed it has already been addressed elsewhere, I'd be grateful if you would point me to that.