With typeracer my accuracy is always hovering around 94%. For the Amphetype its ~97%, both around the same wpm right now.  My only explanation is that I'm trying to type really fast with Typeracer and making more mistakes which brings me down to the speed of Amphetype were I'm typing slower but making fewer mistakes.

My last 3 Amphetype scores.
45.7 wpm, 98.1% accuracy
50.9 wpm, 98.7% accuracy
47.2 wpm, 96.4% accuracy

I remember Shai saying that he spent ~1000 hours researching and developing colemak, but I too don't know what exactly he did :/.

As a note, my QWERTY skills have degraded precipitously.  Today I took some typing tests and I was at 18 wpm using qwerty.  That is too low as I sometimes need to use other peoples keyboards.  So I guess I'm also going to have to relearn it....

I probably should have taken the advice to type in QWERTY for the first 5-10 minutes of the day. I ignored that advice because I thought it would hurt my Colemak progress and I didn't expect my QWERTY to drop so fast.

It's possible there was no strict methodology.  Trial and error is very much a part of development, after all.

Dreymar, that sounds right based on the bits and pieces I've picked up about Shai's process.  It would definitely be nice to have him put in some of that in writing in a centralized place. Especially statistics about finger distance/usage.  I've yet to see a measure I like, most seem to use a distance based off of the fingers always being on the home row, pressing a key, and then returning back(which I strongly disagree with as fingers will often remain by the key they just pressed). As an extreme example paste qwerty over and over into one of the distance measures and it will say qwerty is the worst layout, when for that text it really should be the best. More realistic colemak examples are words such as edged, going, high, fulfilled, etc.

As a progress update, my learning has definitely slowed down and I've been sitting at ~50 wpm for several days now. The ~30 wpm increase I need to get back to my old qwerty speed is seeming huge at this point. I am confident I will get back there(and pass it), but I'm not sure when...

Fortunately I now feel like I'm typing fairly comfortably and my work product isn't affected too much at this point. A little more fluidity in my typing and the difference in my daily work should be negligible. So slow incremental speed improvements are fine. Overall, I definitely enjoy typing with the Colemak layout more and I've already made the choice not to go back.

No alternative layout has a compelling argument for it. Unless you're easily compelled - which I am, as are most people here. You too, Pinky, or you wouldn't be using Dvorak. All that Battle of the Keys stuff really leads to that conclusion: Lack of solid evidence either way. And it's hard to get funding or will for good research it seems.

I wish I could get that level of satisfaction!  I don't disagree with you of course.  It was a bit of nightmare for me in choosing Dvorak, in part because there were other alternatives out there and a lot of hearsay. (Some of which is baseless.)

There are some published principles that forge the foundation of Dvorak's design. (See the wikipedia page.) 

My take on Colemak is that Shai had an interest in alternative layouts, and tried to design a layout for Qwerty typists with oft' quoted design principles garnered elsewhere (heavily influenced by Dvorak).  Add to that some trial and error (and feedback from others?), and you've got a labour of love or at least a tool to scratch a personal itch by a hobbyist.  A layout that seems to resonate well with many others (especially Tony_VN!).

Perhaps Shai is just modest and is happy to throw his layout out there for others to assess.  I'm sure there is rhyme and reason behind the layout (hopefully not lost to the annals of history).  Such as most common letters are on the home row etc.  Maybe the website would do well listing these.  Or perhaps the advocates could write a couple of concise blog posts.  I don't know.  I'm just venting my own frustration really.  Please excuse me.

If someone ever has an interest in why I use Dvorak, I try and give them a little overview, and usually add that there are alternatives.  Including Colemak that I describe as a layout that favours existing Qwerty touch typists that is heavily inspired by Dvorak.  I'll remark that both lack a solid scientific backing.   I'm not really a Dvorak or touch typing advocate due to my ongoing discomfort.  So rather annoyingly I just say checkout the wikipedia page, see what you think.

Sorry @jsmithy, I'll shut up now.

The website really needs updating, but you're probably tired of hearing me say that.  It really needs to read something like "Less keys swapped (making tarmak feasible!)" and "improved QWERTY shortcuts on the entire friggin left-side, for those technically inclined people who use more than ZXCV".

I'd say that well-specified heuristics are a lot stronger than anecdotes, but in turn a lot weaker than peer-reviewed scientific studies.  Unfortunately, researchers just don't seem interested in this field (I guess they just think "QWERTY is already the de facto leader, so what's the point?") so we have to make do.   

It would, of course, be nice to have the design process on record, but I don't think it's really a big deal in this case.  It would only really help would-be layout designers at this point.  Better to just study the practical results.

Maybe I should learn dvorak just for the heck of it.  I always thought I would prefer hand alternation over rolls, and after adopting vim-like layouts everywhere, the improved Ctrl- shortcuts (what really made the decision) aren't quite as important to me as they used to be.  Not sure I could get used to the l placement, however...

This have been discussed at least 3 times or so.

Based on different factors, you can generate at least 1000 layouts that meet the requirements. Most of them are the same of efficiency, generated by simply shuffling the equivalent keys. Patojk layout analyzer is the best place to try this.

Even with fixed ZXCV and/or fixed QWABM, there would be 20 layouts or so that are equal with Colemak.

Adding individual factors like finger strengths, ideally each person will have his/her own layouts that is most suitable for him/her.

We should have a dedicated topic that named "Layout of the week" where each show their own preferable layouts.

Ugh, progress over the last 5 days has not been good.  While typing at ~50wpm isn't slowing me down a lot at work it's still a bit frustrating. Hopefully this is a wall I can break through.

I think I'm worst at what I call flash typing, when I haven't typed for a while then need to type something quickly. I got strange looks from my girlfriend yesterday when she asked me to google something and I typed it out as if I was drugged up sloth.

Thanks Tetrinity, focusing on accuracy did help, now solidly up to mid 50's.  I got a 56 wpm average today so progress is continuing, albeit slower than I expected considering how much I improved the first 2 weeks.  I guess I'll just slowly get back up to speed over the next month or two.

~3 Month checkup.

Well typing in colemak for me is definitely more comfortable than qwerty and getting more and more natural by the day. This goes for both normal typing and for programming. I do feel there is less strain on my fingers on a day to day basis.

Speed wise, well, I am still improving, but very slowly. Measured using my typeracer account, from 0 wpm to 60 wpm took me about 30 days. From 60 wpm to my current 74 wpm, it's taken over 60 days.  All in all, I'm at 90% of my old typing speed after a little over 3 months. At my current rate of progress I will probably be back at my old speed in another 2-3 months, which would put me at about 1/2 year.

With typing speeds I am pretty sure I am faster on the majority of words already, but on certain harder words I will often make mistakes or just slow down significantly.  On those words I think it's just a matter of time before they become more comfortable.  I'm looking at you "unknown" and "edges".

My fastest 10fast fingers score