In this post I go into more detail about analysis, naming and history. It's non-essential stuff, but some may find it interesting.

TARMAK ANALYSIS

Here is some analysis of Tarmak, using analysis programs available online. Keep in mind though, that such analyzers are only as good as their design. The maker of CodeSharp (and the Workman layout) for instance, feels that the Colemak D and H positions are very bad for him, and that hand alternation is a lot better than finger rolls. While I've come to appreciate the first sentiment, I don't agree on all the rest. I also feel that same-finger bigrams are very bad; these are more common in the Workman layout than in Colemak and I think that's a horrible idea. Be that as it may, the tools take into consideration many relevant parameters and seem to be well designed as such.

CodeSharp "overall effort"
The following analysis was done by user Cevgar in 2011 using the CodeSharp analyzer. This tool seems to be no longer available online after the Workman layout moved to a new site?

This graphic shows the number of keys that each successive layout step changes, with the Workman and Dvorak layouts included as "further steps" after Colemak since they change 21 and 33 of the 35 letter block keys. The line in the graph represents how much "effort reduction per key" you gain for each step of more moved keys. Think of it as "bang for buck" if you will!

This was done for the older Tarmak(ETOIR) steps, but the results are very similar for the new one.

The "effort reduction per key" was assessed using the "overall effort" percentage from the CodeSharp analyzer, divided by the number of extra keys moved from the previous layout (4-3-3-3-4-4-12 keys). As seen, effort gain per key increases the most for the first Tarmak steps. Switching to Tarmak 2–4 really gains you a lot of benefit for your troubles. The first step may seem ineffective in comparison, but it bears the penalty of having to displace the J key; if that step too had moved only 3 keys it'd have done better but it's still clear that the second step is very important too. After the last Tarmak step, the final 4-key change to Colemak doesn't produce much modeled benefit compared to the previous changes. Since the mini-loops of Tarmak have been sorted using this model that may be expected; however, I personally believe that the Colemak has significant important benefits that the model are blind to! The full Colemak has a lot less same-finger bigrams and gives other benefits compared to the last Tarmak step which the analyzer won't quite pick up.

The Workman layout moves 4 more keys but doesn't do significantly better in the model for that trouble (21.8% vs 21.3% for Colemak) even though it was designed in tandem with the Codesharp analysis model. For a more updated analysis, check out the Colemakmods analysis page.

The Dvorak layout plummets because it changes nearly twice as many keys as Colemak without doing better in the analysis at all. In the name of good sportsmanship, I show the "UI" variant of Dvorak which scored much better than the garden-variety Dvorak (21.4% vs 18.9%; the latter is comparable to Tarmak3).

Patorjk's Keyboard Layout Analyzer
Here's a more recent analysis of both the old ETOIR and the current ETROI progressions. I did a side-by-side comparison on Patorjk's analyzer of a big chunk of text: A couple of chapters from Pride and Prejudice (found at Project Gutenberg) containing more than ten thousand strokes of the letter E should be representative for good English. The ETROI analysis is the basis for the heat map animation shown below.

    The current Tarmak(ETROI): http://patorjk.com/keyboard-layout-anal … d/MVB2CBnK
    The older Tarmak(ETOIR): http://patorjk.com/keyboard-layout-anal … d/2b1Bz5t9

The measly 141 instances of J in this sample prove that it is indeed a rare letter – the rarest of those that are moved by Colemak. Even the colon/semicolon key and Z had more presses and only X and Q of all the letters less (241/142/102/83 presses, respectively). This is an argument for moving only the J around, as a letter that rare won't be used enough to settle in your motor memory before the next step.

Patorjk favorises hand alternation which brings Colemak to somewhat of a disadvantage per se since Colemak focuses on same-hand rolls instead. I don't know whether this bias will play into the present analysis but I shouldn't think it matters a lot?

SteveP99's Switching Difficulty Index and Layout Analyzer

stevep99, the originator of the Colemak-DH a.k.a. Curl(DH) mod, has a page where he analyzes the difficulty of switching to other layouts, and a page where he analyzes their quality. His models are quite interesting, so have a gander if you're into modeling and analysis!

Layout  | Moves |  ∑m   | Effort ||  SDI  |  LEI  | "B4B" |
-----------------------------------------------------------
QWERTY  |   0   |  0.0  |  2.37  ||    0  |    0  |  1.0  |
Tarmak1 |   4   |  2.5  |  2.10  ||    8  |   13  |  1.6  |
Tarmak2 |   7   |  3.3  |  1.92  ||   12  |   23  |  2.0  |
Tarmak3 |  10   |  4.8  |  1.90  ||   19  |   25  |  1.3  |
Tarmak4 |  14   |  6.5  |  1.78  ||   25  |   33  |  1.3  |
Colemak |  17   |  8.0  |  1.72  ||   31  |   37  |  1.2  |
CmkCurl |  19   |  8.8  |  1.64  ||   33  |   45  |  1.4  |
-----------------------------------------------------------
Dvorak  |  28   | 22.3  |  1.88  ||   61  |   26  |  0.4  |
-----------------------------------------------------------

SteveP99's Switching Difficulty Index (S.D.I.) and Effort Per Key translated into Layout Ease and "Bang4Buck" indices (L.E.I. & "B4B").

I've fiddled a bit with Steve's raw results. His effort parameter is lower for better layouts, so I made a "Layout Ease Index" similar to his "Layout Switching Index" for direct comparison with QWERTY: The L.E.I. is defined as the percentual increase in ratio of QWERTY's "overall score effort per key" to the layout's (using Angle mods). Thus, it's 0 for QWERTY(Angle) and higher as we get to better layouts according to Steve's analysis model. Then a conjectural "Bang4Buck" parameter was calculated as the ratio of L.E.I.(="Bang") to S.D.I.(="Buck"). Don't take that too literally, but it should provide some clue as to how much gain you get for your effort.

CAVEAT: Do not take the above to mean that I think you should stop at Tarmak2 just because the analysis shows the most "Bang4Buck" there! It's true that switching seven or so keys gives a lot of model effort reduction for the cost – cf. the Minimak-8 layout which is an attempt to get the most out of the fewest key changes. But I really don't think that most typists will be happy typing on a "half-baked" optimization! Also, there are subtle benefits to full optimization that the model doesn't catch as mentioned above.

Also, the low "B4B" score of Dvorak in no way means that you shouldn't switch to Dvorak just because it doesn't give any "B4B" over QWERTY! Rather, it means that you should indeed be careful about interpreting these indices. Dvorak is a fine layout, well optimized and well tried out. It has a few blemishes but it's an excellent layout overall. Switching to Dvorak from QWERTY is well worth the effort in my opinion, and I should know because I've done it. It's just that with Colemak(-DH) you get something just as good and in my opinion even better, at a much lower cost! To which the above analysis agrees.

The numbers would be a bit different for Tarmak-DH/Colemak-DH. These layouts will have a sharper learning curve but end up at better analysis scores.

THE NAMING OF TARMAK

The name 'Tarmak' is obviously inspired by 'Colemak'. Thus I quipped when I thought this up:

To separate the Tarmak variants, I use a naming scheme that refers to the most important letter brought to its Colemak home row position by each step. Of the most common letters in English (ETAOINSHR, sorted by frequency), QWERTY only has A, S and H on the home row! So the first standard Tarmak step is Tarmak1(E) which brings the E to the home row (and N as well, but E is a lot more common!), then comes Tarmak2(ET) which adds T and so on. The whole progression to Colemak is Tarmak(ETROI). It's a bit of a toss-up between R and S frequencywise and they're both sorted in the same step but 'R' is brought to the home row while 'S' is only shuffled about.

THE HISTORY OF TARMAK

For those interested in how this came about or the former Tarmak variants, here's a run-through.

2008: Tarmac(EO-I-TR)
Back in 2008, there was a forum discussion about how any layout change will break our "gestalt" feeling and typing flow but then the flow is regained at a speed which depends on the complexity of the changes made. That made me think about learning Colemak in steps, and then I realized that the Colemak layout has some interesting properties because of Shai Coleman's design philosophies: Only two keys switch hands (P and E), and the L>U>I keys form a closed loop of changes. Also, the large loop of changes can be split fairly down the middle (P>R>S>D>G>T>F and E>K>N>J>Y>O>;). Even if I knew Colemak myself by then I set about devising a stepwise learning approach.

There is an older Tarmak topic, in which it started out as a 3-step approach starting with the LUI loop as  a "warm-up exercise" and then a larger one for each hand. This 'Tarmac' (as I spelt it then) was successfully tried out by user ezuk who was very happy with it, and there was much rejoicing.

   
    Spectrally color-coded Tarmak(EO-I-TR, 2008). The red/orange and blue/purple steps were merged.

2011: Tarmak(E-T-O-I-R)
Then in 2011, some useful model analysis by user cevgar (using the "CodeSharp" layout analyzer) on smaller loop chunks made me change the recommended loop progression to maximize the gain in typing model efficiency for each short step of 3–4 moved keys. Getting the G>T>F loop in place sailed up as much more important than fixing one hand at a time, since QWERTY puts the T in a needlessly poor position. This led to the 5–step Tarmak(ETOIR – earlier called 'ETOIS') progression that split each hand loop in half and demoted the LUI loop to the next-to-last step.

It's possible to do the "most effective" key movements 3-4 keys at a time, each step bringing an important key or two to the Colemak home position while moving less central keys out of the way to their proper positions.

Here are the Tarmak(ETOIR) Transitional Colemak Layout in sorted "lumps" for learning Colemak in 2-5 steps as desired:

 q  w {J} r  t  y  u  i  o  p 
   a  s  d  f  g  h {N}{E} l  ;     The "Tarmak1(E)" transitional layout (E>K>N>J)
     z  x  c  v  b {K} m 

 q  w {F} r {G} y  u  i  o  p 
   a  s  d {T}{J} h  N  E  l  ;     The "Tarmak2(ET)" transitional layout (G>T>F>E>K>N>J)
     z  x  c  v  b  K  m 

 q  w  F  r  G {J} u  i {Y} p 
   a  s  d  T {;} h  N  E  l {O}    The "Tarmak3(ETO)" transitional layout (G>T>F>E>K>N>J>Y>O>;)
     z  x  c  v  b  K  m 

 q  w  F  r  G  J {L}{U} Y  p 
   a  s  d  T  ;  h  N  E {I} O     The "Tarmak4(ETOI)" transitional layout (G>T>F>E>K>N>J>Y>O>; & L>U>I)
     z  x  c  v  b  K  m 

 q  w  F {P} G  J  L  U  Y {;}
   a {R}{S} T {D} h  N  E  I  O     The Colemak layout (P>R>S>D>G>T>F>E>K>N>J>Y>O>; & L>U>I)
     z  x  c  v  b  K  m 

The Tarmak(ETOIR) layouts have been tried successfully by several people, including forum reports by users ghaz, bph and Coldmak. The reports suggest that if you give each step a little time to settle in, you can keep your productivity while learning Colemak.

   
    Spectrally color-coded Tarmak(ETOIR, 2011). Red > orange > green > blue > purple.

2014: Tarmak(E-T-R-O-I)
Now, in April 2014 Karl Nordstrom (user karl) presented a new Tarmak variant in an excellent forum topic. He was loath to move the semicolon to the middle of the board because he's typing a lot of code. Also, he wanted the LUI loop last so that the misplaced keys would be resolved as soon as possible to keep confusion to a minimum.

Thus came about the new Tarmak(ETROI) progression, with a (J)>R>S>D loop in step 3 that lets us finish the big loop before the LUI loop and move only J to wrong positions:

 q  w {J} r  t  y  u  i  o  p 
   a  s  d  f  g  h {N}{E} l  ;     The "Tarmak1(E)" transitional layout (E>K>N>J)
     z  x  c  v  b {K} m 

 q  w {F} r {G} y  u  i  o  p 
   a  s  d {T}{J} h  N  E  l  ;     The "Tarmak2(ET)" transitional layout (G>T>F>E>K>N>J)
     z  x  c  v  b  K  m 

 q  w  F {J} G  y  u  i  o  p 
   a {R}{S} T {D} h  N  E  l  ;     The "Tarmak3(ETR)" transitional layout (R>S>D>G>T>F>E>K>N>J)
     z  x  c  v  b  K  m 

 q  w  F {P} G {J} u  i {Y}{;}
   a  R  S  T  D  h  N  E  l {O}    The "Tarmak4(ETRO)" transitional layout (Y>O>;>P>R>S>D>G>T>F>E>K>N>J)
     z  x  c  v  b  K  m 

 q  w  F  P  G  J {L}{U} Y  ; 
   a  R  S  T  D  h  N  E {I} O     The Colemak layout (Y>O>;>P>R>S>D>G>T>F>E>K>N>J & L>U>I)
     z  x  c  v  b  K  m 

This progression evens out the efficiency gain per moved key a little between the steps, depending on the analyzer used. As Karl points out though, that isn't so important to a learner as you won't be all that efficient while struggling with placements anyway. Unless you're planning on staying with a Tarmak step for a very long time it shouldn't matter at all.

   
    Spectrally color-coded Tarmak(ETROI, 2014).

Some arguments for the Tarmak(ETROI) vs Tarmak(ETOIR) progressions:
• Both progressions seem good and should provide similar benefit.
• Moving only the J (the rarest letter moved in Colemak) feels more consistent than moving first that and then the semicolon.
• J is actually less frequent than the semicolon/colon, not only in code but in English text as well!
• Moving R and S into place earlier and in a separate step should be beneficial: These two letters often cause Colemak learners trouble since they "castle" past each other from their QWERTY positions. In Tarmak3(ETR) you get to learn their positions without having to focus on so much else at the same time.
• User jpeg729 informed that the same-finger 'PO/OP' bigrams of the old Tarmak3(ETO) and Tarmak4(ETOI) steps felt bad. The new Tarmak3(ETR) step keeps O and P in their QWERTY positions where everyone is used to them, and in the next step both letters are moved into place.
• Moving the L>U>I loop to the end lets you finish the big loop so as to resolve the temporarily misplaced letter (J) earlier.

2015: Tarmak(E-T-R-O-I) for the Curl-DvH ergo mod

The Curl(DH) mods are an ingenious way of getting the frequent D and H keys to better positions for people who dislike the center "trench" and feel that this part of the home row doesn't feel like home after all! See my post further down if you're interested.

Note that my 2015 Curl-DvH variant was superseded by a unified Curl-DHk mod in 2017 and finally Curl-DH(m) in 2020 – see below!

Here are the steps for the 2015 Tarmak(ETROI)-Curl(DvH):

   
    2015: Spectrally color-coded Tarmak(ETROI) with the old Curl-DvH mod, shown on an ISO keyboard.

2017: Tarmak(E-T-R-O-I) for the Colemak-DHk standard ergo mod

In May 2017, we agreed on one common Colemak-DH mod, consisting of a mere MK switch from the original "mod-DH" or Curl(DHm).

   
    Spectrally color-coded Tarmak(ETROI) with the Curl(DHk)Angle mod, shown on an ISO keyboard.

Here's what the Tarmak steps were like with the 2017 Colemak-DHk mod:

   
    2017: Spectrally color-coded Tarmak(ETROI) with the Curl(DHk)Angle(Z)-ANSI mod.

Tarmak-DHk steps on an ANSI keyboard with the Angle(Z) mod:

2020: Tarmak(E-T-R-O-I) for the Colemak-DH(m) standard ergo mod

In Oct 2020, we once more agreed on a new Curl-DH standard: SteveP's original mod! There can be only one. See the Tarmak-DH post above.

Uh-oh: I've been thinking about the use of 'S' to name the R>S>D step, as I earlier used 'R'. I now realize that it was very wrong of me to switch to 'S' just because it sounded a bit better (at the time): The S isn't brought to the home row but actually gets a position more or less on par with its old one! Also, the S is a confusing letter in Colemak. So the focus can and shall be on how the R is brought to a better position in that step! I have no idea why I forgot that a little back.

I've performed a little historical revisionism and edited my files as well as several posts from 'S' to 'R'. Sorry about the confusion!

I know that 'ETROI' and the like isn't very noob friendly. But I still need a way to sort out which variants I'm talking about. I've taken out the 'ETROI'/'ETOIR'/etc references in a few front-end places so it looks more elegant and less confusing (if slightly less informative).

Sure - but I'm also inclined to agree with bph.  I've been looking into a github website for the "Talking Points" as well, ever since realizing how much better a separate domain is from a search point of view.

As long as we're talking about domains, colemak.sexy is still available ;)

The search engines index everything, but definitely give higher precedence to non-forum-posts.

Well, that's the thing, isn't it.  We want it to come up in the non-esoteric searches of a newbie.  (Ideally when searching "learn colemak" or even "colemak".)

You could check that by doing the same searches in Duck Duck Go, which won't bubble you.

Continued from forum.colemak.com/viewtopic.php?pid=15159#p15159

This is a little bit like saying "Tarmak may work for its target audience, but it probably won't work for those whom it wasn't designed for."

Unfortunate, but not exactly groundbreaking ;)

Sure... as long as you completely ignore all the content of those steps ;)

[Seriously, the ideas aren't similar at all.  Typing tutors start by teaching you a layout's home keys then expanding your knowledge to the others.  Tarmak starts with all the QWERTY keys you already know, rearranging them into Colemak a little at a time.  The rearrangements are based not on touchtyping technique, but on maximizing gain/effort and avoiding disruption.  Tarmak is not so much a tutor as it is a re-trainer.]

The most obvious, yet far-reaching difference between the two is that Tarmak is able to train with full layouts rather than just a subset of the letters.  In practice, this means:

1. You get more practice, being able to type anything rather than just a limited selection of pre-generated content.  Real-life itself becomes a valid lesson plan, so to speak.  As the Bard was fond of saying,

2. The intermediate layouts act as a buffer between yourself and QWERTY; getting "stuck" on one of them is not nearly as big a deal as if you were switching wholesale.  Each Tarmak stage provides intermediate benefits, while coming closer to Colemak than all the previous ones.

With the typing tutor, it's "QWERTY or bust", with no middle ground should you run into any issues.  Any outcome aside from successful transition puts you back to square one (at best square one point five if you manage to retain some limited knowledge), with no additional practice forthcoming.

On the contrary, the disadvantage of Tarmak is that people don't procrastinate enough, moving on to the next step too soon. It's too much like an RPG in that sense: you really want to Level Up.

https://forum.colemak.com/viewtopic.php … 849#p14849

Of course, typing tutors are even worse at this.  If humans get restless just because they have a next stage to progress to, imagine how much more impatient they'd get when they've already exhausted the tutor's current content.  Premature switching is probably measured in minutes, rather than days!

I am also following a slow-but-steady process, using the Tarmak steps for work, all week.  I've decided to progress on Fridays, after work, so that I can practice a bit over the weekend before being hit on productivity on Monday.

I started just over a week ago and this was my first work day with Tarmak-2.  I was worried at first by how hard some of week one came, but I had to use a qwerty layout today and found E and N had really, truly and fully moved!  I actually had to look, hunt and peck.  K is coming along, and I can tell with week 2 that F is my new K (read: problem child).  Still, one typing test I took today had me already up to 38 wpm (100% accuracy, but I make heavy use of backspace!), even with the fresh (week 2!) changes.

Week 3 scares me, with the horror stories I've read here about the R-S-D shift, but I'll burn that bridge behind me when I get to it. :)

I'm also personally ham-strung by another voluntary re-mapping; in an informal straw poll I did with my team, about 80% of us only use the left shift key.  E.g., I normally type "A" with the left pinky hitting shift, and the left ring finger hitting "A".  As I'm learning Colemak, I'm also re-learning correct shift key use.  That is proving at least as hard as the Tarmak steps.

I have 30 years of qwerty (and bad shift key habits) to overcome, but already, I can feel hints of how much more comfortable Colemak will be.

My sincere thanks to Shai Coleman for the layout work, DreymaR for the Tarmak option, and everyone who's proven and helped improve upon the Tarmak option.

P.S. My top qwerty speed was once 87 wpm (85 wpm adjusted); I believe I will be able to top 70 on Colemak before Halloween.  July 11th should be my first day on full Colemak, and I'll still give myself the entire next week of practice in real-world use before an official "weigh-in".  I'll then have 3.5 months to practice to meet my goal.

Well, I've discovered finally how to get hi-games.net to work with the latest Java snafu (Java has been well and truly Oracled over the years), so now I can practice for speed. I hit 41 wpm today on Tarmak-2 on a 5 minute test, but I'm hindered somewhat by a sprained thumb that's also causing me arm pain. Still, with 2 days to go before stepping up to Tarmak-3, I'm satisfied with my progress. T comes easily, G is getting there, and F is still my problem child, but as K resolved its problem status in week 2, I'm not inclined to stress too much over it.

Even better, I just applied DreymaR's excellent tool kit for Linux on my LMDE VM. WOOT! I was in Linux withdrawal, as I prefer to do all my personal time computer use on Linux, but I couldn't risk the Tarmak/Qwerty back-and-forth confusion; I don't think I could be one of those people who can flip between at will -- I view this conversion as a one-way trip, with no intention of ever looking back!  I now have all four systems I use daily set up to support my weekly Tarmak steps.

Thanks again, DreymaR, for the dev work!

PS. There's another "also" in my list of personal typing retraining, besides the "canonically correct shift use" issue; when I learned to type, the rule was still that two spaces separated sentences. When teletypes and typewriters used the original Courier Monotype, that made sense... but then so did Qwerty at the time. Modern kerned fonts obviate the two-space sentence separator just as computer keyboards obviate Qwerty as a layout.  Still, the mapping in my head of the period key is almost automatically ".  ". Even at the ends of paragraphs, it's often "period-space-space-backspace-backspace-enter"! I'm not letting either the shift or two-space correction efforts slow me down, though; I just think if I'm going to the effort to "fix" my keyboard layout, I should fix my typing altogether.

DreymaR, I need your help!  It seems the Linux Tarmak templates I installed were for one of the previous versions, not the latest -ETROI one.  Tarmak-3 isn't right, so now I'm back in Linux withdrawal. :-(

And this week's R-S-D transition is definitely the toughest one so far.  I'm glad I got hi-games.net working on my XP VM, as I really need the practice this week!  (I am so far an abysmal failure at getting Java running on Chrome under Linux.  Clues welcome!)  I seem to be averaging about 37 wpm today, and I still have 3.5 days of practice before moving up to Tarmak-4.  I was over 50 at the end of week 2, and the R-S-D rotation instantly set me back to the low 20's.  As hard as this week is, I can already feel how much more comfortable it will be once I get it worked into my muscle memory a bit.

Well, I haven't gotten around to the Linux patching yet; luckily (?) work has kept me too busy for the Linux withdrawal to really impact me this week.

However, I downloaded Amphetype and I've been making heavy use of it this week.  In just two or three days, it's helped me climb from the high 30 wpm range up to the high 40's (actually my last two practice runs both broke 50!).

I'll move up to Tarmak-4 tomorrow, but I hope to use the holiday weekend to hit the Linux patching, too.

As a new user note, I would like to report at this point in my transition that I have "forgotten" the Qwerty positions of the earliest remapped keys, such as E, N and T.  On Qwerty, I have to hunt and peck for these, which is fine, even preferable to me at this point.  Also, some of the more useful trigraphs have surprised me in how automatic they've become, especially -ing and -ion; they're already almost as immediate and instinctual as they were on Qwerty, which sometimes shocks me, as I finish typing a word and pause (hurting my wpm score) because I'm so sure I must have mucked it up (but didn't)!  The word "first", with that perfect 3-letter roll still gives me pause, as does every instance of the "st" digraph, as it seems impossible to me, subconsciously, that it could be quite *that* easy.  The biggest challenge has been the "unlearning", but it appears to me that a few days intense practice overwrites not only the individual key mappings in memory, but also the trained letter groupings, and the instinctive speed developed in the past for those common digraphs and trigraphs really (amazingly) quickly translates!

One week from tomorrow I'll be on full Colemak!  My thanks again to everyone who helped make (DreymaR) and improve/refine (lalop, et al.) the Tarmak method.

Who are "we", please? I see you haven't posted here before.

Please link back to this page from GitHub. I see you've used my description in the form of a screenshot so it'd be nice and proper if you mentioned where it's from.

Also, which original posting of the Mac layouts are you referring to? I didn't see a link or credit, but maybe I overlooked it.

I consider CapsLock behavior separate from Colemak and Tarmak per se. Not sure what you implemented - Backspace? I suppose it can't be modular and optional for the Mac then. And there's no Extend mapping possibilities that I'm aware of for that platform. Too bad. Maybe the Backspace option is the best for now.

Tarmak for Curl-DH mods!

NOTE (2017): These are now outdated, and superseded by the new Colemak-DH mod!

As mentioned further up in this topic, my Big Bag of Keyboard Tricks holds a new trick or two: The Curl-DH mods invented by Stevep99! For those who didn't know: These mods move the D and H from the middle "trench" which several users have found hard to reach, to the lower row positions of Colemak B/K or V/M which some – myself included – find much better than their "non-home-row" reputations. I'm using the Curl-DbgHk mod and liking it a lot.

These mods move a couple more keys than standard Colemak but keep all finger assignments. Also, by keeping the G in its QWERTY place the number of extra key movements is reduced by one. My preferred mod (Curl-DbgHk) moves only 18 keys to Colemak's 17 if you don't count the Angle mod (as the ZXCV keys all keep their finger assignments!). That's very much worth it to me.

It's of course a bit worrying that the newcomer to the Colemak world faces yet another daunting choice! But we know that the DH keys have given newcomers pause before, and this has been a contributory factor to the creation of IMNSHO unnecessary layouts such as Workman and Norman. I feel it's important that this choice of getting the full benefits of Colemak with DH mods at a low cost is available.

For the right-hand "H" part of the mod, I recommend getting the Hk switch with Tarmak1 if you go for my mod variant (DbgHk), but if you use Steve's variant (Dvbg/Hm) you may do the right-hand mod whenever you feel like as the Hm switch only affects keys not moved by Colemak! It's after all a relatively minor but also very simple improvement so just search your feelings. ;-)

The left-hand "D" mod should be done with Tarmak2 as that's where the G is normally moved which we don't want in this case. Moving B instead triggers the inclusion of the Angle mod. However, that makes this step rather large. It's possible to split step 2 if that feels like a hurdle (although the ZXCV keys don't change fingering but only wrist angle in the process). The easiest way of doing that would be to first learn the Angle mod (B>V>C>X>Z>_) separately, at the small cost of a little more B confusion. Using Steve's variant the V stays in its old place and isn't part of the Angle mod (at the cost of splitting it from ZXC).

Implementations
Linux/XKB: Tarmak needs a few tweaks, but nothing major! In the Linux world you can turn on the D(v)bg and Hk/Hm switches separately which makes implementation quite simple requiring only a tweak in step 2. See the symbols/colemak file or the XKB Big Bag topic.
[Update: You can now select the modified step 2 directly as Tarmak2bk/vm; there's also a Tarmak1k. For the other Tarmak steps, use options or edit the colemak file, and/or select models with the setxkb.sh script.]

Windows/PKL: For PKL you'll have to comment/uncomment some lines in the layout.ini file and if your layout doesn't have the necessary lines you can find them in the Colemak-ISO AngleWide file as the Curl mod is a development on the Angle mod and I expect that most Curl modders will also go with the Wide mod!

Mac: User mkborregaard has made a Karabinier implementation of the Curl(DbgHk)Angle mod Tarmak steps. Thanks!

Tarmak(ETROI)-Curl(DbgHk)Angle,  a.k.a. DreymaR's Curl-DH mod

   
    Colemak[eD]-No shown with the Curl(DbgHk)Angle and Wide ergo mods

Tarmak(ETROI)-Curl(DvbgHm)Angle,  a.k.a. SteveP's mod-DH

   
    Colemak[eD] shown with the Curl(DvbgHm)Angle ergo mod by Stevep99 and the Wide ergo mod

For the technically inclined, here's what the key change loops look like with the Curl-DH mods included (on ANSI boards, ignore '_'):

Colemak[Angle] loops:  Y>O>;>P  >  R>S>D  >            G>T>F  >  E>K>N>J       L>U>I     [B>V>C>X>Z>_]
Colemak-DH     loops:  Y>O>;>P  >  R>S>D  >    C>X>Z>_>B>T>F  >  E>K>H>M>N>J   L>U>I
Colemak-DbgHk  loops:  Y>O>;>P  >  R>S>D  >  V>C>X>Z>_>B>T>F  >  E>K>H>N>J     L>U>I
Colemak-DvbgHm loops:  Y>O>;>P  >  R>S>D  >    C>X>Z>_>B>T>F  >  E>K>N>J       L>U>I     H>M

   
        The Tarmak(ETROI) transitional Colemak layouts for the Curl(DbgHk)Angle ergo mod
        The steps are spectrally color-coded from red (Tarmak1) to violet (Colemak-CurlAngle)
        Not shown in this figure is how the J key is forced into temporary positions (QWERTY E, B and R)