"I would say that that's a poor way of comparing the layouts. You're allowing Maltron to have one more homerow key than either Colemak or Dvorak which is the only reason why it appears to be "better" according to your metrics. This is problematic for two reasons:
1. It introduces a hardware dependency for the Maltron layout.
2. It does not consider what the Colemak or Dvorak layout could gain using the same hardware."
I'm not "allowing" anything. It's just better design. It's like complaining that a single speed bicycle doesn't perform as well as a 27 speed bicycle because it has extra gears. That's the intention. The "extra" home key (E) is there simply because unlike QWERTY, where the hands are constantly moving away from the home rows, Malt does not need to have each thumb striking the space bar. With Malt, the hands very seldom move far or frequently enough to warrant this feature, so one of the thumbs would be left with nothing to do. Although I don't use either Dvorak or Colemak, if the thumbs were being under-worked because of the "efficiency" of each layout, then that would seem to be an opportunity missed to improve them. It would seem a simple enough test, just try to use only one thumb for the space bar. If your typing speed does not change, then you have an "extra" home key waiting to be employed.
The "hardware dependency" is immaterial since the Maltron shell design is quite a separate function from the Malt key distribution. The two videos show that the Malt layout is the ONLY variable involved, as the same physical keyboard shape is used in both videos.
It's of no importance what Colemak or Dvorak MIGHT gain. Dvorak and Colemak are what they are, and that's what I measured. I took them as I found them. If you want to change them, then they cease to be Dvorak or Colemak. However, if you feel that this is important, then you should approach Maltron because they offer Dvorak as an alternative key distribution, and I'm sure they could supply your Colemak request adequately. (You would need to buy four keyboards, QWERTY, Dvorak, Colemak and Malt to have a truly unbiased test - or at least two, because they are switchable between different distributions, with the key caps dual engraved.)
Because my method of measurement of the key distribution is completely independent of human factors (eg comfort, strength/length of individual fingers, distance hands have to move over the keyboard, typing speed, pressure to operate keys) it's unbiased and can be applied to any theoretical key distribution straight away, without the necessity of an operator having to familiarise him or herself with the distribution - ie there is no practice effect. The only variable in my method is the initial list of words used. But as long as the same list of words is used for each distribution pattern analysis then it ceases to be a variable and becomes a constant across all distributions tested.
To see how simple and accurate this test is it's not even necessary to have a physical keyboard. Create the word list you wish to use. Just draw on a piece of paper a proposed key distribution. Make a list showing all the keys that are NOT on the home rows. I use the basic word processing function to find and replace each of these letters with a marker of some sort (I use #), then I use a short macro to go through the list and find any word which contains this marker and delete the word. What I am left with is a list of words which contain ONLY letters which are typed without moving the hands from the home rows. Since even a single marker in a word will result in that word being deleted, there will simply be a "go/no-go" test to pass. This could be automated to an even greater degree by combining both steps into the one via a recurring macro, when one could enter all the letters not on home rows and in a single instance, determine the score. Whilst I don't perform this often enough to warrant creating such a macro, I would suggest that it would not be a difficult task to analyse a couple of dozen different theoretical layouts in a day, assuming they were already created ahead of time.
This to me is far simpler, cheaper and quicker than building (or modifying) a keyboard, then practising with it until a pre-determined level of proficiency is reached (say a random wpm speed) which could take weeks and only applies to that one operator.
Joe