According to my AdnW analyser, the best 2-letter swap is L - E   Mind, this is according to the AdnW-ideals, which are partly the same as Colemak's, but differ in that
- AdnW wants to avoid " adjacent keys" (qwerty AS is seen as bad, AD is seen as better)
- AdnW prefers alternation over rolls.

Anyway, here are the metrics (tested on English text corpus)

QWERTY           576.383 Gesamtaufwand  338.719 Lageaufwand        links rechts
                   6.804 Kollisionen      6.299 Shift-Kollisionen  ob 28.0 20.2
  qwert yuiop[    52.755 Handwechsel     41.474 Shift-Handwechsel  mi 22.1  9.5
  asdfg hjkl;'     1.080 Ein-/Auswärts   11.226 IndirKollision     un  6.8 13.3
  zxcvb nm,./     21.628 benachbart      12.586 Shift-benachbart  sum 56.9 43.1
                  9.1  8.4 18.5 20.9 --.- --.- 18.4  8.9 12.1  3.6 Sh  1.1  1.7

QwLrty       497.365 Gesamtaufwand  306.693 Lageaufwand        links rechts
                   3.897 Kollisionen      6.070 Shift-Kollisionen  ob 20.2 20.2
  qwlrt yuiop[    58.837 Handwechsel     40.532 Shift-Handwechsel  mi 22.1 17.3
  asdfg hjke;'     1.075 Ein-/Auswärts   11.858 IndirKollision     un  6.8 13.3
  zxcvb nm,./     12.121 benachbart      19.351 Shift-benachbart  sum 49.2 50.8
                  9.1  8.4 10.7 20.9 --.- --.- 18.4  8.9 19.9  3.6 Sh  1.1  1.7

QwLrty is better than Qwerty:
- Kollisionen (same finger% ) is down from 6.8% to 3.9%
- Alternation is up from 52.7% to 58.8%
- The percentage of key strokes on adjacent keys is down from 21.6% to 12.1%. This is due to the very common digrams WE and ER. Those are on adjacent keys in Qwerty, but no longer in QwLrty
- hand balance is much better: from 56.9 (left) / 43.1 (right)  to 49.2 (left) / 50.8 (right)
- it's still not ideal of course (how could it with only 2 keys swapped).

@cevgar The idea is that especially pinky-ring (or ring-middle) is cumbersome, because the lower independency of these two fingers.  ER is less of a problem, since index and middle are both strong and they are independent.

For me, I agree with the AdnW/Dvorak "adjacent finger theory" - to me, ring-index feels better than middle-index. But I also found that it highly depends on the keyboard. On flat keyboards with little travel, such as laptop keyboards, adjacent keys do not feel bad.  In other words, on a modern laptop, rolls feel nice. But on keyboards with more travel, rolls feel less nice.

I think it is one of those things you should really try out. I have tried several layouts, some installed on my laptop (Colemak, various MTGAP versions, Dvorak, AdnW), some I tried in the browser, such as Capewell and the Carpalx versions. To me, AdnW felt best, followed by Dvorak. 

tl;dr: it depends on your hands and on the keyboard you type on.

Edit: I remember reading somewhere that it has to do with "rolling"  the hand. If you type (qwerty) AD, you may slightly ' wiggle' your hands, in addition to the fingers moving. The combination of those movements makes that pinky-middle combo easier than AS, where you have less "lever" since the fingers are closer.

That is 4-2 > 3-2 (like you say) I suppose, right ?  I agree with most, only to me 2-4 is better than 2-3-4

To add one more element, how do you feel about 'waves'? This is what I call a movement that flows left- right- left over the keyboard. On a Colemak board a word like INTASTE or, less extreme, words like PRINT and LIST.  In these cases, a "mixed roll' such as 2-4-3 does not bother me so much, as long as the "wave" keeps rolling in that direction.....

Mixed rolls you can put in a model (I believe it already IS partly in the AdnW-model, and also in the MTGAP model). But waves are harder to model. You need 4, 5 or even 6-gram analysis  (is that tetragram, pentagram and hexagram?) that is, you would need to look at 4, 5 or 6-letter combinations. In the analysers I know of (AdnW, Mtgap and carpalx) only letterfrequencies (=1-grams), 2-grams and sometimes 3-grams are considered.

At the same time I think such an analysis would not add much. As some of you may know, keyboards are compromises. I could design a keyboard with many inward rolls (actually, the Mtgap software makes this sort of experiments easy).  But it will come at a price: bad load balance over fingers and hands, more home row jumps, more same finger, etcetera. Likewise, i could design the lowest same finger keyboard in the world. I'm sure it would suck at for instance alternation, or have all rolls outward, etc.

There is a league of good layouts, that have found some middle ground, that is, that perform good on some aspects and well enough on others. Colemak, Dvorak, MTGAP, Carpalx, Aus der Neo Welt (AdnW), Balance12.....   The 'battle' between layout fans is a nice pastime, in which I sure like to participate! Because, you know, of course my custom AdnW layout is muuuuch better than your stock Colemak !!!!  ;-)    But to be honest, it's like fighting if Volvo is better than Audi or the other way around ! ...

Edit: typos

Interesting, both the wave concept (to which I agree at least partially) and the Gentner result!

No points for guessing that I'm a Prius person.

I partially agree with you - pinky use should in general be avoided, but home-row pinky usage is acceptable. The real problem comes when you have to stretch to reach awkward keys with pinkies, but in Colemak the top and bottom row pinky keys are Q Z ; /  -- none of which are very frequent.

Here are my key ratings:

However since creating that, my love of the lower-row index finger keys (qwerty C/M or the D/H keys in my mod) has only grown. I could probably argue for these keys to be more favourable than the home row pinkies even!

The problem with the JA switch is you end up with a lot of common letters on the left index finger: A, N, H, M, U - not to mention a lot of awkward bigrams. And you still end up with J still being in a too good a position. For minimal swaps involving J, I would sooner swap J with N.

Index finger key stretching diagonally up is not pleasant, the T is very badly positioned in Qwerty, much worse than E in my view also. Consequently nowadays I think the best single change in Qwerty is the F-T switch.

People need to sooner realize and accept the fact that the home pinky is not that great. Empirical and anecdotal evidence suggests that it is: slow, weak, inflexible; awkward for rolls, double letters, and long one-hand sequences; easily tired and stressed.

And yet the bigrams are possibly faster than pressing the home pinky once.

You can see the timing of each finger by trying this typing speed test yourself: http://patorjk.com/typing-speed-test/ . Type as fast as you can using your favorite layout for at least 2 minutes. After each trial, it shows you which fingers are fastest and slowest. Do it several times and note the averages for each finger.

My own scores reveal that the home pinky is slower than the top and bottom rows for ring and index fingers by a wide margin. It's even slower than index finger reaching to the inside column. The bottom row ring and index are surprisingly fast and should not be underestimated.

Greatly disagree with your chart. My chart looks like this since I've done the speed test above:

The values take into account the speed and roll potential of the fingers. Lower scores are better. As you can see, the 10 best keys on each hand do not involve the pinky at all.

You need to do a few throw-away trials before each session to warm up and loosen up the fingers. As you do more trials on the same session, the stats become more evident. For me, the pinky never gets below three digits of milliseconds, while the other fingers can consistently get down to 65-90 ms.

It's always interesting to see other people's assessment of the relative strength for fingers/keys. But even so, as you would probably guess, I disagree with a lot of what you have there. In particular:

- you have top-row ring finger better than bottom row index finger. I disagree particularly strongly with this.
- there is only a small penalty for the middle column (Qwerty G and H), you are rating these the same as bottom row index finger.
- you have diagonally-up index fingers (Qwerty T and Y) as better than home-row pinky.

Your diagram seem to be assuming a matrix-like layout, so I don't know if you are allowing for standard staggered design. If you are using a matrix keyboard that might account for your different experience to some extent, but I can't imagine it would make *that* much difference.

I do think you may have a point the home-row pinkies have historically been given slightly too favourable a rating, but what you have there doesn't match my experience at all.

That's part of the fun of keyboard layouts though - each to their own!

I had something of a brainwave a couple of weeks ago which - if not particularly good keyswap - I feel is worth sharing.

My thought process went something like this:

   - T and R cause too many consecutive finger keystrokes. Neither should be on an index finger.
   - Typing R with L.Middle doesn't help because then you trade TR GR BR FR (and variants) for ER RE ERE and CR.
   - unless you type E with L.Ring. Which is an awkward reach from the home position on S.
   - None of which fixes that T is too far away for the second most frequent letter.
   - Unfortunately the Left Hand simply can't be moved because A is so blasted important.
   - It isn't as if Q and Z matter, they are almost never used.
   - *Sigh*
   - Y'know, ; is also almost never used.
   - Funny. Because it's, like, almost the exact opposite setup of the Left hand position. Moving the left or right hands inward would fix a lot of same finger problems. The left can't move because of the importance of A dispite the useless surronding keys, and you want to move the right because of the uselessness of ; but can't because it has useful surrounding keys.
   - ...
   - So, I'll just swap A and ; then?

Result of A ; swap.

Pros:
- Distance to T Y and B are reduced.
- L.Hand bottom row stagger is fixed.
- L.Index samefinger digraphs broken up.
- L.Alt is easier to hit with L.Pinkie
- Only swaps two keys! Heck, with the infrequency of ; I'd go so far as to call it 1.5 keys. In theory, on regular qwerty this would be an easy to read cypher.

Cons:
- Doesn't fix ED
- Doesn't fix UN
- Increases distance to L.Shift
- E is now on L.Ring.
- Ctrl+A is now a R.Hand shortcut.
- L.Alt is harder to hit with L.Thumb.
- Swaps keys. You'd probably get fired from your workplace.
- Requires relearning entire Left Hand.

Of course, since this is a mini-mod, you could always tack others on top of it. Like the E/L swap for instance.

Interesting idea. Also, because you are only swapping A with semicolon (same row), you can more easily switch keycaps. 

I find it a bit weird to have only one centre column though. If you don't mind about the Q, Z and semicolon (since they are rare), isn't what you have essentially the same as this:

It's basically the same layout home-key wise, but you get to keep the fingers in the usual location. Requires pulling up more keys though.

What do you mean by 'adaptive' fingering? I didn't see it in the link.

Do you mean based on the context of the word and surrounding letters? If so, are you saying you ALWAYS use R3 for 'U'? I think thats a big mistake to always use that finger for U. I only use R3 for U in UH,UB,UN combos, and R2 (standard) otherwise. How would you type "guide" then?

I could be convinced that adapting E based on context isn't as good, but U seems pretty straightforward to use index as default and middle on occasion

The E/L swap does look nice, as does E/K, and may be worthwhile for me to consider both of those

Yes certainly for the avg typist, using the same finger for an alt fingering will definitely be a lower bottleneck. I don't think its optimal though.

You make a good point about U though. Using middle for U is really only a problem for UI and UK (luck, duck) and the UK is extremely rare and has a letter in between. So U was a bad example. I already used the RT example, could also be used for FT as in 'shift'. You'd def want to use middle for F there, but not in most other cases.

I'm curious why you've changed your thoughts on that?