Good idea! I like idea for allowing 4 or 2 steps of learning! The problem is step 2 is too soon for RSD hurdle and last step is too far for simple NJ swap (the fact of bottom row is not important). Another thing I like is moving IO in one step.

The criticisms are valid, but there are pros and cons to every layout, and they need to be balanced out. Overall I believe the proposed layouts would work better for most people most of the time.

* The proposed layout has only 4 steps vs 5 steps of Tarmak. I really think that 5 steps is too many. Changing a layout is disruptive, and you would like to minimize the amount of times that happens. Also 5 layouts generates extra maintenance work.
* I think that having the option to do in just 2 steps is extremely valuable.
* The layouts maximize the improvement in each step, moving frequent letters in each step. In the current Tarmak the first step has less improvement, so someone trying it out might be less motivated to continue.
* Because learning the RS keys can be confusing, I believe it's better to do it sooner rather than later so you get more practice. Step 2 will be a bit more difficult, but not overly so.
* Containing the area of change is also extremely important. If you forget where a key is located, you might start hitting keys in the vicinity until you find the right one. In that case you know that there's a limited radius of where that key can be which makes it less frustrating. Also it makes it easier to visually search for the keys if the key is nearby.
* Not everyone is a programmer, and many modern programming languages don't use semicolons (e.g. Python, Ruby, Go, Coffeescript, Swift, Clojure, Scala, etc.), and in many cases the IDE can add semicolons. There's also an option to skip that step for people who use semi-colon a lot.
* I find Tarmak confusing and intimidating. The layouts above can easily be explained in simple language (Colemak-4 - start with the most frequent letters, Colemak-8 - left hand, Colemak-13 - right hand, Colemak - everything else). The two step version is even simpler - learn Colemak one hand at a time (Colemak-8 - left hand + E, Colemak - right hand + P)
* DRKGS or LOUI; can be moved in either order. However the order chosen generates less temporary key placements, and still manages to bring all but two keys to the home row. A different order would work less well for those doing it in two steps. Also the two step version only moves one key between hands in each step.

Thanks for the explanations!

The discussion of the semicolon frequency in that debate was actually not taking programmers into major consideration. Certainly that's what got it started but a little snooping around showed that the semicolon key is much more common than the letter J even in classical English text. A somewhat surprising but I think valid result. (Keep in mind that the key holds both semicolon and colon.)

I agree with Lalop on not making the individual stages too big. It's easier to skip a step to reduce the number of steps than it is to tackle a step that feels too big for which there is no cure. The LUI loop is small, but it's also independent. Not a bad idea to couple it with moving the O, but that happens at the cost of one more step with displaced keys. The idea of getting the big loop and misplaced keys out of the way first seems sound to me. Since the LUI step is small, it'll probably be attempted soon after the fourth Tarmak, thus the bigrams shouldn't matter that much for the transition.

I do feel that I personally might try jumping straight to step 2 of Tarmak, tackling a larger step with both hands and immediately getting a largish benefit. After all, what I actually did was to jump straight to step 5...! ;-) But people are quite different so I provide the smaller steps and I'm willing to implement and serve that many steps. What really creates maintenance work is the number of variants and choices available today! Would you like ANSI sauce and Curly A-Wing fries with that order, sir?

If an accelerated progression is desired, I recommend doing steps 2, 3 and 5 (7, 4 and 8 keys moved). This gains the advantages of a big improvement in the first step, an easy RSD step and moving LUIO in one step – but the steps are big so it's only for the brave. A two-step Tarmak may be too much I feel, defeating much of the purpose (maybe one-hand-at-a-time is the best option for that – jumping straight to Tarmak3 seems a little much)? Lalop – I don't see why people who want a more aggressive approach can't just read the instructions and skip a step or two? Go on brave newcomer, tackle that 7-letter jump to Tarmak2 right away if you feel up to it!

As Piotr (and you) pointed out, the first step could've given more immediate benefit by making it a T>F>E>K one, but that's at the cost of displacing the more common K thus creating a little more disruption. I find it quite appealing to show the customers that only the rarely used letter J has to move to temporary positions; displacing several keys seems more confusing and intimidating to me – certainly displacing several keys at once as is the case with Colemak-13!

The argument that someone might try the first step of Tarmak and not find it effective enough seems somewhat contrived to me. I find it hard to believe that anyone should try putting the E and N in place and start their journey towards Colemak, only to begin worrying about the T and then quit becase they didn't care to stick around for a week or two! At least, I never heard anyone mention such quibbles at all.

I have put a T>F>E>K step in my PKL files as a variant, but I won't be promoting it actively to avoid confusion for the newcomers. Heaven knows it's confusing enough to meet all the choices available now!

"Colemak one hand at a time" is what started Tarmak in the first place – it even predates the name Tarma(c)k (but I started with the right hand and now it's the left hand first)! We've come full circle it seems. ;-) I abandoned that concept because I hadn't thought of doing the left hand with E in place first, but also because I felt that it'd be easier to learn a little with one hand and then with the other. It's just a thought, and I'm not sure what's best. Tarmak moves keys R(L)–L–L–R(L)–R, so it's not strictly alternating nor strictly hand-bound – but it does finish the left hand in step 3 like yours does in step 2. I feel that it's a balanced approach.

If you really think that the stronger first step is a real advantage, I could change it. But I don't really think it's worth it over having only one temporary key.

My thoughts exactly.

I know that a key with only 0.5% frequency in a very wrong place may feel annoying, so it should be a real advantage to utilize a key that's much less common than that.

Note that you can use Shai's and Piotr's first T>F>E>K step with Tarmak! It's an alternative first step if you don't mind misplacing K in step 1, and leads smoothly to Tarmak2 if you wish.

If Shai makes materials for this progression, I won't have to provide that alternative step as I was about to. Heh. ;-)

Shai, Re your first claim: It seems to me that your system minimizes temporary positions to a lesser degree than Tarmak at least.

• Tarmak misplaces the J in steps 1, 2 and 3; step 4 has no misplaced keys – one of the arguments for Karl's variant was to get rid of the misplaced keys first. That's three misplacements total.
• Your progression misplaces K in steps 1, 2 and 3, and the semicolon in step 3. That's four misplacements!
• Using only Tarmak 2–5 to have the same number of steps as you gives us only two misplacements total!
   (Note: That makes the first step a little too hard but also very efficient for the brave; your progression makes for more even steps if you really want 2 or 4 steps, certainly!).

It's not the number of times a key moves that's most important, I think, but how many layouts have how many keys in the wrong place. Having keys in wrong places is what trips you up, not whether they had moved before any given step. This was also my criticism against Piotr's first transitional layout suggestion.

Even counting it your way though, the full Tarmak moves the J three times and you move the K and semicolon three times. So you really can't claim to minimize temporary positions (unless you're comparing with Piotr's first suggestion).

It just seemed a little unclear what you meant. Glad that's cleared up, then!

• It may seem that you're giving Tarmak a harsh judgement – "it lacks appeal" and is "unattractive"?!? To the contrary, I've seen many people express happy feelings about it. I know that its presentation could've been better, but all in all it looks quite appealing to my eye – and this is corroborated by Ezuk's blog for instance.

• I don't think that the difference between 4 and 5 steps is dramatic. Your way makes for a better 2-step solution if that's desirable, there's a point. But the forum has discussed this a lot as you're aware and it seems to me that most potential users want a maximum of 4–5 keys per step (as Lalop said above); also, when the steps are smaller it's easier to merge two steps to get an intermediate one should someone choose to. I've noted though, that the users who have told success stories chose not to skip any steps. To the contrary, they tend to say that they got too eager and did the last step too early and they kinda wish they hadn't so they'd have avoided some stress – this is my impression (and that's even when the last step is a tiny little 3-key one!). So I think easy does it, still. The pilot user who tried the first progression with large steps spent a long time on each step and a longish time total to reach the goal compared to many later Tarmak users; but he's only one data point of course. Again, it doesn't seem to me that the users are as concerned with the number of steps as with the max number of keys per step. Transition users in general seem to want baby steps.

• The names of the layouts aren't confusing when you read my main Tarmak post I think – it's Tarmak1–5. The letter codes are explained further down and really logical enough to me (the most important letter improved for each step), but in the main post and PKL it's just numbering. The reason the letter codes are kept is to distinguish between variants, and that's still necessary. I've chosen to bury my earlier variants instead of keeping them as active options. No need for them, really.

We're in the Open Source dilemma: More choices is good for some but makes it daunting for the starters. Part of your success story with Colemak has probably been freezing it since 2006; this opens for some criticism from the most intense optimizers but in the end gives the newcomers a feeling of a stable and finalized product which is nice. Once choice is introduced, it gets harder for the newbies. (Yeah, sorry about that DH-mod thing; it's just too much fun to leave be – and addresses a real problem which has led to the creation of in my opinion inferior layout alternatives....)

• This might in fact be a point of critique against your introduction of transitional layouts at this point, as mentioned above! Why introduce a progression with comparable benefit to the existing one when its main effect may be making it harder to choose for the newcomers? I've worked a lot on making Tarmak newbie-friendly (PKL implementations with images, listening to experiences and reworking presentations based on their feedback, Ezuk's blog presentation etc) because I realize how daunting it is for them; now I fear they may be too confused when arriving here and getting the impression of schizmas? Nobody wants another "editor war"/"bike shed" story!? But I guess you felt the need for it, given that you've got all this gripes with the existing Tarmak. We'll just have to live with diversity then.

• The effort change per step has been analyzed with Patorjk and CodeSharp, but I don't trust those analyzers too much to represent my goals. Other analyses have been performed and largely confirm the fairly obvious. How is this unclear?

• "Loops" are confusing? I don't think so? It's very obvious how the L>U>I loop works, surely you must agree? The other loop is bigger so it needs to be broken up. Not really confusing, is it? Or is it my notation that trips you up?

• New users don't tend to skip the first step! It's an option for the brave, and people who choose Tarmak tend to be careful with their bite sizes (the very brave just cold turkey straight to Colemak) as said above. These want a big step, and so they can have it. You provide a 2-step alternative with similarly big steps. Indeed, if someone wants to ease into transitional layouts they shouldn't dive into a big step first but sidle up to it; in such a case I'd recommend Tarmak steps 1, 2, 3 and 5.

• Tarmak1 has very obvious improvements over QWERTY, as it brings E and N to the home row and shows good improvement in analysis. The T>F>E>K step beats it but at the cost of moving a five times more common letter and moving a different letter from the other steps (see the discussion with Karl that led to the current Tarmak – I really think he had a point there).

• Those who want a Minimak instead of transition, shouldn't use any of our steps but a Minimak or similar. They're really a different customer base. It's still safe enough to stay with any of our steps for a while. Nobody sane would stop at Tarmak4 forever for instance, as it's a short way from "perfection". I believe that would apply to your Colemak-4 too. If someone do get stuck it'll likely be at Tarmak3 or your Colemak-8 I guess? Those are decent enough I guess.

• Which steps you might skip is clearly explained as far as I can see (in the Mini-FAQ, as it's an optional strategy). How should that be made clearer?

• I disagree that your progression has an even progression in difficulty, from the looks of it. Your step 2 (Colemak-8) seems more difficult than the others, and Colemak-13 looks confusing with two misplaced keys. There's no real reason to seek a strictly accelerating difficulty progression though, as I see it. Indeed Tarmak4 has to be more difficult, like your Colemak-4, as it clears up the big loop leading to no more misplaced keys – splitting it otherwise would lead to too many steps I think. But I think that's a good stage to face and overcome that difficulty.

• If someone tries your progression, they'll be able to report on how it actually feels – the problem is of course that nobody will be able to learn Colemak with both progressions and compare how they actually fly! So I guess we're mostly left to guesswork. Tarmak's been tried and tested and seems to work well, at least. I haven't noticed all the confusion and feelings of weirdness that you seem to experience, but those may be underreported.

• Your progression is nice for 2 or 4 steps, Tarmak is nice for 3 or 5 steps. The one-hand-at-a-time approach seemed appealing at first, but I haven't got any real impression as to its actual benefits.

• J-hopping is weird? To the contrary, moving only the least used letter bar some odd ones must be much less weird and annoying than misplacing several more common ones at once! Several users cheered the "J-hopper" approach as much more clear and appealing than the ETOIR approach which misplaced two keys (only one at a time). Tarmak users will know that the J is the one in an odd place, and since it's so rare (in English) it doesn't hurt much.

• Yes, particularly the second post (technically the third now) in the Tarmak thread is long and wordy. It's optional, and mostly for people who ask about things like you're doing now. Yes, I'd like a clearer and more manageable presentation. We all should work on better presentations. I think mine aren't the worst on this site even if they're generally too verbose and thorough. At least, the graphics are nice – ne c'est pas?

It seems to me that you've made a very long list now... maybe you should number your gripes for clarity? :-þ And I get the feeling that many of your critiques here are a bit contrived and/or apply to both our progressions? It's still an interesting discussion, of course.

Thanks for the input.

A Dvorak-to-Colemak set of layouts is an interesting thought! It wouldn't be too hard to make. But it's already a bit of a hassle to maintain several transitional layouts, so that would be even more... hmmm...

My initial thought would be that Dvorak users have already switched to one layout that's much harder to switch to. So either they still remember their QWERTY well enough (in which case they could use Tarmak) or they're hard core layout switchers already (in which case they'll manage without Tarmak, as I did). But if you feel that a set of transitional layouts would've helped you, then likely other feel so too. So as mentioned, it's an interesting thought!

Dvorak is so rare and Colemak even more so that anyone switching from Dvorak to Colemak is like a unicorn sighting.

Colemak is starting to appear in HW/FW of some keyboards like the popular POK3R, who knows, we might actually see it gain more popularity if more keyboard manufacturers start doing that.

I went through Tarmak a few months ago. I did not find it overly confusing, and I would not have benefited from fewer steps. I really like the idea of mostly just rotating three keys.
To me, the bigger problem was and is the D-H curl. I chose to learn that, exactly because I had myself always felt that the sideways movement with the index fingers is unpleasant. Still, though, this is a choice you have to make before starting to learn a new layout, which makes vanilla colemak and curl-DH more distinct. Now, this post also introduces an alternative transitional layout directed towards vanilla colemak.

I guess a question becomes whether it would make more sense to fork the projects? It is perhaps a necessity that more diversity will continue to arise in the absence of objective metrics on which keys are harder to hit across keyboards in a majority of people.

Indeed I have, as seen in the Tarmak "extras" post. If someone decides to go with a CurlAngle-modded Colemak I do recommend using these steps right away.

   
        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)

That shows my variant, the "light curl".

If only :-P

The problem is "optimal" is inherently subjective. People have different priorities for what they think is most important in a layout. Obviously there is a also trade-off between difficultly of learning vs final benefit gained, some may be willing to learn a less optimal but easier-to-learn layout. But even if you wanted the most optimal layout regardless of learning difficulty, you would still have to decide on what features were a priority.

Personally, I think my layout (full DH mod) is as near to optimal as makes no odds.