Huh, whaa...☺hh, you must mean ♂ understand ☼☺⌂ < ♀, ☺☻. I put in symbols because they don't interfere with typing. And they're useful for expressions and foreign words. The number pad under 123 is great if you're using a straight or corrected keyboard and like the phone-pad map.

Thank you much for commenting☺
-.*,

Well, one more 'final' revision. I decided to move the number pad to shifted caps (could make for some odd shouting) and fill the altgr keys with 27 common words (excluding He and his).  And I set ',' + ',' to ','

normal:
`   1   2   3   4   5   6   7   8   9   0    hadhave
    b   p   d   l   v   ;   u   k   q   j    as  we
    s   n   t   h   r   w   e   o   a   i   AltGr
    c   f   g   m   x   -   '   ,   .   y

AltGr:
    ¹   ²   ³   ¤   €   ¼   ½   ¾   ‘   ’   *.-,-.*,
     be  but andthat or  not to  on  as      why=
     is  in  it with are was the of  a   I
    z    for at from/   =       ,    an  by

Dead comma:
,`,1,2,3,4,5,6,7,8,9,0, had,have
  [ ( ð { « ü ë ö ä å, as , we ¬
  ß ñ þ ,h® ú é ó á í
  ç ,f¶ µ ? ,-ø , æ !

Dead comma shifted:
  ] ) ð } » Ü Ë Ö Ä Å
  § Ñ Þ   © Ú É Ó Á Í
  Ç   ° M ¿   Ø  Æ ¡

While that may be true, I fear its' a waste of the strong index finger's capacity to fill it up with punctuation. Punctuation is so rare compared to almost all the letters, which is why most of it has been handed to the weak little finger in both QWERTY and Colemak.

The period goes best on the top row so that it is close to the numbers. This helps significantly for typing decimal numbers, but especially for IT and network admins who have to enter a lot of IP addresses. And makes one rely less on the keypad (for various reasons: numlock doesn't work, no keypad, reaching back and forth, etc.). Similar consideration for the comma as some countries use that as the decimal separator.

Preface: It's fun to make long posts about things that happened more than a year ago.

"Capewell...doesn't use the home row enough."

I actually find that W, E, I, and O (QWERTY) are just as easy as the 2 center home row positions, and are better when you take into account finger rolling. However, I can't find any digraph lists that include less common things, so I can determine some good top row digraphs. Perhaps Shai The Master Of Links knows of such a list?

"So I decided to design my own optimum layout."

Lol, that makes 100 of us. That made me laugh.

"Why try to find the best keyboard when tomorrow there will be a better one"

Because it's fun!!

"Along the way I think I found the best technique for optimizing keyboard with current technology: make it an online game. Each player starts with an identical board. And the players take turns moving keys (like chess pieces) to achieve a certain goal of optimization. After each round the computer evaluates the board. The first player to reach the goal wins. Or if it a draw (both players decide to pass or have moved every key on the board once) the player with the highest stat wins. Then the board is stored in a data base. In addition the game could ask contestants what keyboard layout they use and then record key timings as they speak to establish optimal board positions."

That is a great idea, although it sounds hard to make. It would have a certain complexity to it: I make a move now that increases same finger ratio, so that I can move T to the home row next move.

"In short I realised human intuition can do a good job choosing keys"
Yeah. Colemak and Arensito, my 2 favorites, were made based on intuition. You really have to have an intimate knowledge of keyboards. I have spent hours rearranging keys and trying to find good layouts, and haven't found anything better than Dvorak without actually working off Dvorak to fix a couple bugs. One reason that humans are better than computers (but slower) at designing keyboards is that with all the current genetic algorithms, it switches 2 random keys. Humans analyze the situation and find 2 keys that it would be a good idea to switch. If people just grabbed 2 keys at random and switched them, they would make much worse keyboards. Computers take the advantage when, at thousands of swaps per second, sometimes they accidentally do some good swaps.

Dear Sorenk:

KCDMFVWU.J
OSTNHLREAI
'GPBZX;,QY

Reduced to the main 30 keys and with Y moved so it is on those 30 keys, this layout scores just above Arensito and just below Capewell (with 7.82). Finger count is nearly as good as Colemak and Arensito, but hand count is worse than all the other layouts I have data on except QWERTY.

How do I do that thing where there's a little box and it says

CODE:

and then all the keys are neatly lined up?

"Did you run a sufficiently representative and comprehensive syllabus through it?"

A note on this. I have 2 blocks of code I test things by, one that's 120 pages and one that's 20 pages, so it will go faster. The 20 page one is about 98% as accurate as the 120 page one, which is partly because I took out anything that could skew letter frequency, such as one part where I wrote:
EXCLAMATION MARK PARTY!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

Yes, that was actually in a document I had. I have no idea why.

"I no happy that to, changed board little. Voice (td nm kg pb fv wu rl jy) help recal & look good. I think good too for japanese with roman character write. "

I have no idea what that means.

Wait, I just read the next sentence. (I am typing this as I read.) You are using your layout, so you are typing really weird. I don't really get that...but whatever.

"Søren, you're hilarious. Keep it up!"

I second that.

Well I have now responded to every single tiny comment on this threat, so I am finished.

I am back to comment on more stuff. I must comment on every aspect of every thread on this entire forum!!! Then I will be the forum king!!

I don't think you have to have 8MB. I made a 6MB (2000 page) corpus that is very very close to the letter frequency from deafandblind.com, the only differences being a few switches in that central area of M to B, where all the letters are pretty close in frequency. In fact, only 2 letters are different.

Mine
etaoinsrhldcumfgywpbvkxjqz
etaoinsrhldcumfgpwybvkxjqz
Deafandblind's

I attribute its accuracy to the wide variety of text, which I think is more important than having a lot. I have about 150 pages of documents, 250 or so of C code, about 500 of casual writing such as forums and emails, 500 of news articles, and 600 of about 8 different books from Project Gutenberg. (I didn't want too much of any one book, so I have at most 100 pages of a book.)

About OstnReai: It beats Colemak on 5 of 8 aspects (travel distance, same finger, home row jumping, hand warping, and rolls) and loses on 3 of 8 (same hand, row changing, and moving to center on the same hand). It still loses to Colemak though, because it is very close on the stats it wins at. The score is less than 5% worse though. Also a problem which my program doesn't register (because it's not exactly a problem) is that the right index doesn't have much stress, only 12% when it could handle more like 15-20%.

I think a good placement for punctuation, even if it doesn't feel great, is on vowels, especially ones that aren't E. The vowels other than E don't end words very frequently, and digraphs with punctuation aren't that common in the first place, so same finger is kept low that way.

That is very true, most of the slowness in programming comes from thinking time. But it does help to be able to type code more quickly. Maybe I can only type code at 40 WPM, but if the rarer keys like * and [ were in even worse positions, maybe I'd only be able to do 30.

Wow, I had some really dumb ideas three months ago.

Anyway, I think Colemak punctuation is pretty good. The layouts I designed had comma in that position, apostrophe where period is, and period on the top of a pinky. That's a bit better, I think. Also if you take into account the shifted state, < and > are really stupid things to have on the main 30 keys. They are extremely uncommon. The 8 most common punctuation are . , - ' " ( ) ? So I'd make . , - ' be the unshifted keys and give ( ) to whichever 2 punctuation are next to each other, and put " and ? on the remaining keys. " would go on ' if possible because that makes logical sense. But also I could understand a Dvorak-like placement of - because the key to the left of the pinky isn't too hard to reach, and I think (just speculating here) that moving keys in and out of the main 30 keys could be confusing.h

I think the W and F positions are two of the best on the board. Using them on punctuation seems a waste to me. Also, many words including the common 'the' end with an E and you want to make S, and E. same-finger digrams which sounds like a bad idea.