I would use an ergonomic mod that let me keep the left wrist straight (mine is the one I call the WideAngle mod). For any layout, not just QWERTY.

Single swaps may not be worth a whole lot. Sure, you may reduce finger travel or some other parameter but that certainly isn't all. That being said, carpalx has a pretty nice-looking model for effort that uses trigraphs and several parameters; it's still only as good as the (subjective) weightings though. Determining the effort for different parameters (or trigraphs) is bound to be difficult since you can't give them proper testing before determining how hard they are, for instance - effort before and after learning is bound to differ in many cases.

I would think that the proposed K-E swap would put the K key in a too good position? That position is a lot better than its reputation because in reality it's an "almost-home-row" position for people who have decent-length middle fingers (i.e., almost everyone?). Just an illustration of the complexity of the issue.

Swap #4 (A-D) would probably need to more unnecessary confusion than real benefit. The S key moving one position to the right in Colemak is probably the most problematic change for people switching to Colemak. Realizing this, Shai first tried to keep the S key in place but at the end it wasn't quite possible.

Interestingly, the layout proposed by Jay Walker is nearly a partial Colemak. :) Just swap around the J-F-K (hehe!) keys and voila! I'd think that'd be a good idea at any rate since the QWERTY 'T' position isn't good enough for the F key in my opinion. One advantage to that would be that you could take one baby step with that layout and then more easily take the rest of the leap to Colemak later should you feel like that. (Swap L-U as well, and now we're even closer!?)

The 'CarpalxQ' layout by Jay Walker; image from Carpalx. (Note that the 'P' should've been red too.)

Okay, analyzing the above some more I'd like to propose a loop! Remember the Tarmak Transitional layout for learning Colemak one hand at a time (with a mini-step before even that should you wish one), as user Ezuk did?

Now, here's an alternative. Rather than focus on learning Colemak one hand at a time we'll focus on making as few changes as possible while maximizing their benefit, and then stepping up the effort if desired. Much like what CarpalX and Jay Walker did (the 'CarpalxQ' layout), but with a Colemak slant. Might be useful for those who would like to try out a half-baked Colemak because a full one seems daunting?

Actually, from the Tarmak remember that there is a little loop that can be implemented quite separately: L>U>I makes 3 keys change from their QWERTY to their Colemak positions without mucking up anything else! This is partially done in the Walker layout above (and if he'd only moved the U as well it'd be there).

The Tarmak had a loop that changed nearly the whole right hand after that one: E>K>N>J>Y>O>; (and then the semicolon got stuck temporarily in the old QWERTY 'E' position). Something similar is going to happen here too, but since the T and F deserve a good position if you're planning on using this for a while and the old E position is quite good they'll get to come out and play (also, the Y>O>; has less impact so we'll forgo that for now). Ready? Here goes: T>F>E>K>N>J and then stick the J in the old T position which should be okay since it's not a very good position and not a common letter. :)

q w F r J y u i o p
 a s d T g h N E l ;    The "LazyMak" loop (T>F>E>K>N>J)
  z x c v b K m

Oops, seems I've named it already! ;) This is a "lazy man's Colemak", right? Trying to get as much benefit as possible out of much less change.

If you want to go a bit further, consider the Y>O>;>P loop as well (one Y more than what the CarpalxQ layout does). That leaves the P in a wrong position, but puts the O on the right-hand home row where it should be:

q w e r t P u i Y ;
 a s d f g h j k l O    The Y>O>;>P mini-loop (alone)
  z x c v b n m

Combining the above I'd put the P on the left hand finally, and the J where it belongs.  It becomes pretty much a Tarmak layout (the whole right hand is changed to Colemak!), but with the added benefit of the semicolon, F and T being in place. This would be even more ambitious then, but fortunately it can be split up in baby steps. One thing that might appeal to layout learners is not having to handle the much complained about 'S' key movement of Colemak right away?

q w F r P J L U Y ;
 a s d T g h N E I O    Ambitious transitional layout (T>F>E>K>N>J>Y>O>;>P plus L>U>I)
  z x c v b K m

I'd recommend making at least the "LazyMak" plus one mini-loop of choice, but the latter would be a bit optional I guess and one step at a time would also be entirely possible if you wanted to (see Ezuk's experiences for instance).

People who really don't like the 'S' key moving might get away with leaving it at that? But I hope Shai or someone can tell us why it's a very very good idea to eventually take the final step and put in the last P>R>S>D>G loop to produce a full Colemak layout even if it seems a bit frightening to the newbie! :)

If you're really sceptical to large changes you might want to forgo a logical use of the QWERTY 'E' position so you could do a "most critical mini-loop" of E>K>N>J (with the J ending up at the QWERTY 'E' position). Then, the next step could be the T>F>E or J>Y>O>; loop depending on whether you wanted to take the LazyMak or Tarmak route first, eventually followed by L>U>I and finally ;>P>R>S>D>G (which I personally wouldn't split up).

Interesting?

On #3 you're missing a P and have double T ;) Good job. Maybe I'll persuade a buddy of mine to change to Colemak this time.

Heh, I hope Suzzy is happy about the intensity and thoroughness with which her question has been answered! :D

Thinking more about it, maybe then the last step should be reevaluated? Moving the P could wait to avoid confusion and the semicolon be our safer joker while pursuing worthier targets? To me it seems that D and G are likely candidates for a better Tarmak#4 (however, moving the D now would really mess up the home position so I'll avoid that):

4) The G>T>F>(;) loop, bringing the most essential keys on the left hand into place

 q  w {F} r {G} J  u  i  Y  p 
   a  s  d {T}{;} h  N  E  l  O     The "Tarmak#4[P5]" transitional layout (G>T>F>E>K>N>J>Y>O>; plus L>U>I)
     z  x  c  v  b  K  m 

Then again, that raises back up the question of whether to do this before or after the L>U>I change! :) Would this step be a lot more beneficial than L>U>I? If not (I haven't got much of an idea, myself) then adherence to the old Tarmak proposals takes precedence. I'll go with the above step as #4 for now.

Yeah. Looks pretty good now, I think. :)  I believe I should move my further musings and info on the Tarmak issue back to the first Tarmak thread now. Cluttered up this topic well and proper already, haven't I. ;)

About tests.
The problem is that all these tests we run are only as good as the model, and the models... frankly, they aren't real hands. They don't float, they don't make accommodations, and they are very, very subjective. For instance I use the keyboard layout analyzer (without a name) at http://www.codesharp.co.uk/dvorak/ as it is the best I can find that doesn't require installing python and/or perl PLUS a 5 week training course (referring to Carpalx and The New Keyboard Layout Project's optimizers). That one says the nj swap is preferable (6.6%) to the ek swap (5.8%), despite the fact that ek has a greater positive impact on distance and same finger ratio. On the other hand, nj has a greater impact on row jumps (On, in, un... think that is about it really) and less damaging to 'effort balance'.

Now if you use the prettier, easier to use, but significantly less capable Pat or JK's Keyboard Layout Analyzer, then its model will tell you that swapping keys like e and f is a wonderful idea, completely ignoring that while you lose the ed (1.47%) and ec(.4%) digraphs, you will kill yourself with the er(2.14%), et(.91%), be(.45%), ve(.55%) and ge(.44%) digraphs.

Taking the question this thread poses in another direction:
What is the single easiest way to improve on Qwerty?

And, as with the majority of my posts, I'm going to again suggest alternate home positions. Nothing too radical this time. Heck, I'm not even going to recommend any key swapping.
No, just rest your fingers on "asef jio;".

I tried rearranging the keys on the codesharp site to try to reflect the new positions and distances, (for instance the left middle finger becomes, e on home, d on bottom, c on number, 3 on top) the results... well, I think they should be better, with only 5.9% (update: 6.5%) effort improvement over Qwerty (and this is not even considering the possible penalties that this position might gain if it turns out to actually be more stressful than traditional placement). Still, it manages to just out preform the ek swap's 5.8%. On paper, the 'fixed' hand position gives less travel distance, better finger/hand balance, greater use of the 'home row', and particularly amusing, less row jumping. The ek swap out preforms in same finger ratio, dropping from qwerty's 4.8 to 3.4. The nj swap is still untouchable at 6.6%, but as stated before, this is mostly due to the 'balancing' statistics. Anyways, I've included the code below. (code is wrong, use new code further down)

`12c4567,.0-=
#qw3rtyu89p[]\
#asefghjio;'*N
*Lzxdvbnmkl/*R
##*S#
*L
#######&<>)_+
######YU**(P{}|
######HJIO:"
######NMKL?
#
*R
~!@C$%^
#QW*#RT
#ASEFG
#ZXDVB
#

Further improvements could probably be made by changing the hand positions to:
   - aser/aset nio;/hio; , uncomfortable for me, puts pointers out of position
   - aert or jiop , makes left or right shifting respectively damn near impossible. ...and I couldn't figure out how to quantify the strain for the e/s or a/s stretch.
   - jiol, ... the less said...

And sorry DreymaR, we're really not giving your Colemak 5-step the feedback it deserves.

Edit: CORRECTION! The codesharp site calculates 3 on the ring and 8 on the index so the code should go as follows

`123c5678,.-=
#qw4rtyu90p[]\
#asefghjio;'*N
*Lzxdvbnmkl/*R
##*S#
*L
#######&**<>_+
######YU()P{}|
######HJIO:"
######NMKL?
#
*R
~!@*#C%^
#QW$RT
#ASEFG
#ZXDVB
#

This change INCREASES the improvement over Qwerty from the previously calculated 5.9% to 6.5%, but decreases the finger effort balance from .8% improvement on qwerty to qwerty being  3.1% improvement on this one. Interestingly enough, switching c to the 4 but leaving the , and . on 8 and 9 (index and middle with the way the test is programmed) raises the improvement over qwerty to 7%, but drops the effort balance to 10.6% in favor of Qwerty.

Ok, as I understand it, there are two independant loops, and 3 dependent loops.

In order of fewest to most key swaps.

LUI independant. L
EKNJ independant. E
JYO: requires EKNJ. J
GTF: requires JYO:. G
PRSD: requires GTF:. P

L
E
E L
E J
E J L
E J G
E J G L
E J G P
E J G P L

If you want it broken down more than that, you are going to need to be a bit more explicit.

For these tests I am using an except (first 50,000 characters) from Benjamin Button. Don't like? Do your own tests.

Layout Codes

LUI

`1234567890-=
#qwertyluop[]\
#asdfghjki;'*N
*Lzxcvbnm,./*R
##*S#
*L
#######&**()_+
######YLUOP{}|
######HJKI:"
######NM<>?
#
*R
~!@*#$%^
#QWERT
#ASDFG
#ZXCVB
#

EKNJ

`1234567890-=
#qwjrtyuiop[]\
#asdfghnel;'*N
*Lzxcvbkm,./*R
##*S#
*L
#######&**()_+
######YUIOP{}|
######HNEL:"
######KM<>?
#
*R
~!@*#$%^
#QWJRT
#ASDFG
#ZXCVB
#

EKNJ LUI

`1234567890-=
#qwjrtyluop[]\
#asdfghnei;'*N
*Lzxcvbkm,./*R
##*S#
*L
#######&**()_+
######YLUOP{}|
######HNEI:"
######KM<>?
#
*R
~!@*#$%^
#QWJRT
#ASDFG
#ZXCVB
#

EKNJ JYO:

`1234567890-=
#qw;rtjuiyp[]\
#asdfghnelo'*N
*Lzxcvbkm,./*R
##*S#
*L
#######&**()_+
######JUIYP{}|
######HNELO"
######KM<>?
#
*R
~!@*#$%^
#QW:RT
#ASDFG
#ZXCVB
#

EKNJ JYO: LUI

`1234567890-=
#qw;rtjluyp[]\
#asdfghneio'*N
*Lzxcvbkm,./*R
##*S#
*L
#######&**()_+
######JLUYP{}|
######HNEIO"
######KM<>?
#
*R
~!@*#$%^
#QW:RT
#ASDFG
#ZXCVB
#

EKNJ JYO: GTF:

`1234567890-=
#qwfrgjuiyp[]\
#asdt;hnelo'*N
*Lzxcvbkm,./*R
##*S#
*L
#######&**()_+
######JUIYP{}|
######HNELO"
######KM<>?
#
*R
~!@*#$%^
#QWFRG
#ASDT:
#ZXCVB
#

EKNJ JYO: GTF: LUI

`1234567890-=
#qwfrgjluyp[]\
#asdt;hneio'*N
*Lzxcvbkm,./*R
##*S#
*L
#######&**()_+
######JLUYP{}|
######HNEIO"
######KM<>?
#
*R
~!@*#$%^
#QWFRG
#ASDT:
#ZXCVB
#

EKNJ JYO: GTF: PRSD:

`1234567890-=
#qwfpgjuiy;[]\
#arstdhnelo'*N
*Lzxcvbkm,./*R
##*S#
*L
#######&**()_+
######JUIY:{}|
######HNELO"
######KM<>?
#
*R
~!@*#$%^
#QWFPG
#ARSTD
#ZXCVB
#

EKNJ JYO: GTF: PRSD: LUI

`1234567890-=
#qwfpgjluy;[]\
#arstdhneio'*N
*Lzxcvbkm,./*R
##*S#
*L
#######&**()_+
######JLUY:{}|
######HNEIO"
######KM<>?
#
*R
~!@*#$%^
#QWFPG
#ARSTD
#ZXCVB
#

Now, I recommend you go to the codesharp site and use these codes yourself, because while it looks alright there, posted as code... this next bit is going to be ugly.

Results

Overall effort (see details of how effort is calculated)
Layout      Rank    % improvement over worst    Effort Score    
Qwerty       10         0.0 %                    147,896     
Loop L        9         2.6 %                    144,047     
Loop E        8         11.9 %                   130,290     
Loop EL       7         13.1 %                   128,585     
Loop EJ       6         13.2 %                   128,313     
Loop EJL      5         14.4 %                   126,547     
Loop EJG      4         19.1 %                   119,626     
Loop EJGL     2         20.3 %                   117,859     
Loop EJGP     3         20.1 %                   118,173     
Loop final    1         21.3 %                   116,422     

Effort imbalance between fingers (see details of how effort is calculated)
Layout        Rank    % improvement over worst    Effort Imbalance Score    
Qwerty        2         16.2 %                     475     
Loop L        8          8.3 %                     520     
Loop E        5         13.5 %                     491     
Loop EL       7          9.0 %                     516     
Loop EJ      10          0.0 %                     567     
Loop EJL      9          0.4 %                     565     
Loop EJG      6         10.6 %                     507     
Loop EJGL     4         13.8 %                     489     
Loop EJGP     3         13.8 %                     489     
Loop final    1         17.4 %                     469

Travel Distance (only moves between keys, excluding movement to actually press a key) , - computed for a distance between keys of 1.8 cm
Layout       LThumb   LPinky   LRing    LMiddle   LIndex    RIndex    RMiddle    RRing    RPinky    RThumb   Total
Qwerty        0.0 %    4.8 %    2.8 %    15.5 %    23.7 %    26.9 %     8.5 %     9.9 %     7.9 %    0.0 %    1,221 m
Loop L        0.0 %    5.0 %    3.0 %    16.3 %    24.9 %    26.9 %     5.2 %    10.4 %     8.3 %    0.0 %    1,162 m
Loop E        0.0 %    6.4 %    3.8 %     3.7 %    32.0 %    20.2 %    11.1 %    12.7 %    10.2 %    0.0 %     914 m
Loop EL       0.0 %    6.7 %    4.0 %     3.9 %    33.6 %    21.1 %     6.8 %    13.4 %    10.6 %    0.0 %     871 m
Loop EJ       0.0 %    7.3 %    4.4 %     3.8 %    36.6 %    17.7 %    12.6 %     6.0 %    11.7 %    0.0 %     798 m
Loop EJL      0.0 %    7.7 %    4.6 %     4.0 %    38.8 %    19.5 %     6.8 %     6.3 %    12.3 %    0.0 %     754 m
Loop EJG      0.0 %    8.4 %    5.0 %     8.6 %    22.3 %    20.6 %    14.6 %     6.9 %    13.5 %    0.0 %     687 m
Loop EJGL     0.0 %    9.0 %    5.3 %     9.1 %    23.9 %    22.9 %     8.0 %     7.4 %    14.4 %    0.0 %     642 m
Loop EJGP     0.0 %    8.6 %    5.1 %     8.9 %    22.9 %    21.1 %    15.0 %     7.1 %    11.2 %    0.0 %     669 m
Loop final    0.0 %    9.2 %    5.5 %     9.5 %    24.5 %    23.5 %     8.2 %     7.6 %    12.0 %    0.0 %     625 m

% Home keys (out of all character keys plus SHIFT and ENTER, but excluding keys operated by thumbs - for example, a s d f j k l ; : on Qwerty)
Qwerty        23.3 %    
Loop L        26.0 %    
Loop E        39.9 %    
Loop EL       42.5 %    
Loop EJ       46.7 %    
Loop EJL      49.4 %    
Loop EJG      53.0 %    
Loop EJGL     55.6 %    
Loop EJGP     54.2 %    
Loop final    56.8 %

% Keys typed with the same finger as the previous key (excluding repeating keys like ee)
Qwerty        4.8 %    
Loop L        4.5 %    
Loop E        3.2 %    
Loop EL       2.9 %    
Loop EJ       3.5 %    
Loop EJL      2.6 %    
Loop EJG      3.3 %    
Loop EJGL     2.4 %    
Loop EJGP     2.7 %    
Loop final    1.8 %

Damnable tabs. Still, should be readable. I hope.

ALSO, for what it is worth, N/J E/K gets better results than E>K>N>J by about .75%

EDIT: Straightened Tables

Same text, so scores are comparable to previous test. 50,000 character limitation is by tool's design, not my choice.

Layouts

Loop EG

`1234567890-=
#qwfrgyuiop[]\
#asdtjhnel;'*N
*Lzxcvbkm,./*R
##*S#
*L
#######&**()_+
######YUIOP{}|
######HNEL:"
######KM<>?
#
*R
~!@*#$%^
#QWFRG
#ASDTJ
#ZXCVB
#

Loop EGL

`1234567890-=
#qwfrgyluop[]\
#asdtjhnei;'*N
*Lzxcvbkm,./*R
##*S#
*L
#######&**()_+
######YLUOP{}|
######HNEI:"
######KM<>?
#
*R
~!@*#$%^
#QWFRG
#ASDTJ
#ZXCVB
#

Results

Overall effort (see details of how effort is calculated)
Layout       Rank    % improvement over worst    Effort Score    
Qwerty        7          0.0 %                    147,896     
Dvorak        4         18.9 %                    119,898     
Colemak       3         21.3 %                    116,422     
Workman       1         21.8 %                    115,595     
Dvorak U/I    2         21.4 %                    116,217     
Loop EG       6         17.7 %                    121,678     
Loop EGL      5         18.9 %                    119,973     

Effort imbalance between fingers (see details of how effort is calculated)
Layout       Rank    % improvement over worst    Effort Imbalance Score    
Qwerty        5          4.2 %                    475     
Dvorak        7          0.0 %                    496     
Colemak       4          5.5 %                    469     
Workman       1         21.9 %                    387     
Dvorak U/I    6          3.6 %                    478     
Loop EG       2         10.4 %                    444     
Loop EGL      3          7.9 %                    457

Finger Travel Distance (only moves between keys, excluding movement to actually press a key) , - computed for a distance between keys of 1.8 cm
Layout        LThumb  LPinky    LRing    LMiddle   LIndex    RIndex    RMiddle   RRing     RPinky    RThumb    Total
Qwerty        0.0 %     4.8 %    2.8 %    15.5 %    23.7 %    26.9 %     8.5 %     9.9 %     7.9 %    0.0 %     1,221 m
Dvorak        0.0 %    10.0 %    2.7 %     3.5 %    21.1 %    26.8 %     8.8 %    12.4 %    14.7 %    0.0 %       717 m
Colemak       0.0 %     9.2 %    5.5 %     9.5 %    24.5 %    23.5 %     8.2 %     7.6 %    12.0 %    0.0 %       625 m
Workman       0.0 %     7.7 %    9.7 %    17.1 %    21.4 %    16.5 %     8.2 %     5.8 %    13.5 %    0.0 %       622 m
Dvorak U/I    0.0 %    10.9 %    2.9 %     3.8 %    14.6 %    29.0 %     9.5 %    13.5 %    15.9 %    0.0 %       662 m
Loop EG       0.0 %     7.2 %    4.3 %     7.3 %    19.4 %    23.0 %    12.6 %    14.5 %    11.6 %    0.0 %       802 m
Loop EGL      0.0 %     7.6 %    4.5 %     7.7 %    20.5 %    24.3 %     7.8 %    15.4 %    12.2 %    0.0 %       758 m


% Home keys (out of all character keys plus SHIFT and ENTER, but excluding keys operated by thumbs - for example, a s d f j k l ; : on Qwerty)
Qwerty        23.3 %    
Dvorak        53.8 %    
Colemak       56.8 %    
Workman       57.3 %    
Dvorak U/I    57.3 %    
Loop EG       46.1 %    
Loop EGL      48.8 %    

:: UP ::
% Keys typed with the same finger as the previous key (excluding repeating keys like ee)
Qwerty        4.8 %    
Dvorak        2.1 %    
Colemak       1.8 %    
Workman       2.7 %    
Dvorak U/I    2.1 %    
Loop EG       3.0 %    
Loop EGL      2.7 %

Note: I haven't included the 'keys per finger' or 'effort per finger' charts. Figured I was already tossing around enough numbers. As I stated before, a test is only as good as its model, and no model can compare to a real hand. Looking at this data, Workman comes ahead. Look at Carpalx and Colemak is the clear winner. I suppose this is one of those times where synergy actually counts for something.

Edit: Update
I lied. I have more numbers for you.

Found a new competitive low move layouts. As follows:

Qwerty ek nj tf - 6 keys moved, 17.1% improvement, same finger 3.4%
q w K r F y u i o p
a s d T g h N E l ;
z x c v b J m , . /

Qwerty ek nj tf fk - 6 keys moved, 18% improvement, same finger 3.2%
q w F r K y u i o p
a s d T g h N E l ;
z x c v b J m , . /

Qwerty ek nj tf fk kj - 6 keys moved, 18% improvement, same finger 3%
q w F r J y u i o p
a s d T g h N E l ;
z x c v b K m , . /

I find this interesting, as the last one is supposedly 'better' than the Tarmak Loop EG where 7 keys moved, and the only difference is that J/G. I suppose it is like the balance drop seen around Loop EJ and EJL.

Results

Overall effort (see details of how effort is calculated)
Layout                    Rank    % improvement over worst    Effort Score    
Qwerty                     5          0.0 %                    147,896     
Tarmak Loop EG (7 keys)    3         17.7 %                    121,678     
EK NJ TF (6 keys)          4         17.1 %                    122,551     
EK NJ TF FK (6 keys)       1         18.0 %                    121,226     
EK NJ TF FK KJ (6 keys)    2         18.0 %                    121,348     

Effort imbalance between fingers (see details of how effort is calculated)
Layout                    Rank    % improvement over worst    Effort Imbalance Score    
Qwerty                     5         0.0 %                     475     
Tarmak Loop EG (7 keys)    3         6.5 %                     444     
EK NJ TF (6 keys)          4         2.1 %                     465     
EK NJ TF FK (6 keys)       1         6.8 %                     443     
EK NJ TF FK KJ (6 keys)    2         6.7 %                     443

Finger Travel Distance (only moves between keys, excluding movement to actually press a key) , - computed for a distance between keys of 1.8 cm
Layout                    LThumb   LPinky   LRing    LMiddle   LIndex    RIndex    RMiddle   RRing     RPinky    RThumb   Total
Qwerty                     0.0 %    4.8 %    2.8 %    15.5 %    23.7 %    26.9 %     8.5 %     9.9 %     7.9 %    0.0 %    1,221 m
Tarmak Loop EG (7 keys)    0.0 %    7.2 %    4.3 %     7.3 %    19.4 %    23.0 %    12.6 %    14.5 %    11.6 %    0.0 %      802 m
EK NJ TF (6 keys)          0.0 %    7.2 %    4.3 %     4.8 %    22.7 %    22.1 %    12.7 %    14.6 %    11.6 %    0.0 %      800 m
EK NJ TF FK (6 keys)       0.0 %    7.3 %    4.3 %     7.4 %    19.5 %    22.3 %    12.8 %    14.7 %    11.7 %    0.0 %      793 m
EK NJ TF FK KJ (6 keys)    0.0 %    7.3 %    4.3 %     7.4 %    18.6 %    23.3 %    12.8 %    14.7 %    11.7 %    0.0 %      794 m

% Home keys (out of all character keys plus SHIFT and ENTER, but excluding keys operated by thumbs - for example, a s d f j k l ; : on Qwerty)
Qwerty                     23.3 %    
Tarmak Loop EG (7 keys)    46.1 %    
EK NJ TF (6 keys)          46.1 %    
EK NJ TF FK (6 keys)       46.1 %    
EK NJ TF FK KJ (6 keys)    46.1 %    


% Keys typed with the same finger as the previous key (excluding repeating keys like ee)
Qwerty                     4.8 %    
Tarmak Loop EG (7 keys)    3.0 %    
EK NJ TF (6 keys)          3.4 %    
EK NJ TF FK (6 keys)       3.2 %    
EK NJ TF FK KJ (6 keys)    3.0 %

EDIT: Straightened Tables

Disclaimer: Shift home positions at your own risk.

Since I was curious about the distance reduction of alternate resting positions, I did a bit more analysis*. As the only metric that changes in a meaningful AND measurable way is distance, I chose to use PAT or JK's Keyboard Layout Analyzer for simplicity. Then I figured if I already have this data, and while not as complete as if I had used codesharp, there can't be any harm in posting it. (Then I thought better of it and added the disclaimer.)

The Test Models

(diagrams, not code this time)
Note: Number row is being used to represent keys a 2 row strech from home. Keys are otherwise kept in position relative to new fingering.

Qwerty (asdf jkl;)
 12345 67890-=
 qwert yuiop[]
 asdfg hjkl;'
 zxcvb nm,./

Left-hand

asef jkl;
 12c45 67890-=
 qw3rt yuiop[]
 asefg hjkl;'
 zxdvb nm,./
 
aser jkl;
 12cvb 67890-=
 qw345 yuiop[]
 asert hjkl;'
 zxdfg nm,./
 
aset jkl; (streching pointer over the r)
(left other positions the same, as they are still harder to reach)
 12cvb 67890-=
 qw345 yuiop[]
 asetr hjkl;'
 zxdfg nm,./
 
Right Hand

asdf jil;
 12345 67,90-=
 qwert yu8op[]
 asdfg hjil;'
 zxcvb nmk./
 
asdf jio;
 12345 67,.0-=
 qwert yu89p[]
 asdfg hjio;'
 zxcvb nmkl/
 
asdf jiop (swapped the apostrophe for more accurate representation)
 12345 67,./-=
 qwert yu890']
 asdfg hjiop[
 zxcvb nmkl;
 
Combined ... use your imagination

Results

Note: Results rounded to nearest percent. The number row represents the increased percentage of 2 row stretches. Also, I am discounting distance changes based on the right shift key. I'm a left shifter. I learned on an ANSI board, and on an ANSI board the left shift is half an inch closer to typing position (practically right under the left pinkie) than the right shift. And, you know, without the right shift's wrist breaking angle.

Layout Name    Meters     -%Meters  %Home  +%Number

Standard Home
asdf jkl;    1359.9615    0%       34%    0%

Left Adjustments
asef jkl;    1279.3306    6%       42%    2%
aser jkl;    1237.7218    9%       52%    6%
aset jkl;    1186.8195   13%       52%    6%

Right Adjustments
asdf jil;    1286.2888    5%       40%    1%
asdf jio;    1252.2520    8%       44%    3%
asdf jiop    1238.8844    9%       43%    3%

For Combined Hands, math up the Left and Right numbers. EX:
aser jiop               18%       62%    9%
aset jiop               22%       62%    9%

And that is that. Make of it what you will.

Footnote Rant

*When DreymaR used the word analysis over in the Tarmak thread, I remembered all of the times I have heard the "Colemak has no research" line. Mostly from Dvorak users. Upon thinking about it, it occurred to me that despite what me may think about Carpalx and all of our other fancy computer simulations, most of them are based on logically sound, reasonable, but mostly unproven hypothesis.

It is like the 'people who eat their greens live longer' line. While this may be true, it is a misleading statement that downplays the importance of other factors. Such as, people who make sure to eat balanced diets are more health conscious in the first place, and take better care of themselves. Or, vegetables are more expensive than junk food, in calories per dollar measure, and usually take more work to prepare. Thereby, people who buy veggies are people who can afford veggies, and have more time to do them up, probably make more money, have a higher quality of living, go to their yearly checkups and so on.

More relevant to this topic, people who use Dvorak exhibit a lower percentage of RSI. This might not have quite so much to do with Dvorak itself, and more to do with the type of person who is concerned enough with RSI to take the time to LEARN Dvorak. And, maybe do hand stretches, warm up exercises, take breaks, etc.

Also for what it is worth, supposedly RSI was almost unknown back in the days of typewriters. I have heard it suggested that the reason for this was due to the frequent micro-breaks forced on the typist by working the carriage return lever and changing paper. Again, discounting other relevant factors.

Point is: what we have here is Analysis of potentially relevant data. Pure number crunching. Not Research.

Update: Proof of unreliability.

I was just checking some numbers, and double check, I took the old asef jio; code, supposedly 14% distance reduction through PAT or JK's and ran it through codesharp again, as I had ignored distance percentages on previous runs. Found that while the effort reduction was still at the 6.5% I quoted earlier, apparently it scored 17.7% distance reduction over qwerty. ... I must be missing something, somewhere. Again.

@cevgar

Regarding your footnote rant.  Do you have faith in those statistics that you produced then?

The carpalx stats are strangely alluring at first, but the stats miss a lot of the subtleties of typing.

I think the kind of people that pick an alternative to Qwerty are the kind of people that may already be wrestling with RSI symptoms rather than newcomers to touch typing.

A typewriter is quite a different beast from a modern computer keyboard.  I wonder if there is a keying style that is economical for both?  I remember using a manual typewriter when I was young, and they were pretty tiring on the fingers.  Dvorak was originally designed for a typewriter.

Nice post, cevgar! I agree with your musings, and they're well written too. :)

I'd put more stock in same-finger than in distance, myself. But others see it differently (and the Workman creator seems not to mind same-finger much which makes me think he's a complete nutcase?).

Certainly not! The Tarmak loops are each named for the most important letter they bring to a better place, thus ETROI in the right sequence for the current Tarmak progression (and ETOIR for the old one). But these don't make for good names. They're just codes for marking the different variants. 'Tarmak' is a much better name as it's catchy and interestingly connotated.