Workman, Norman and the Rest

Workman's founding objection is that earlier designs treated the home row as uniformly cheap. In a model where any home-row key costs the same, the obvious move is to pack the most frequent letters onto it — which is roughly what Colemak does.

Workman argues that this is wrong about the index finger. On a standard keyboard the index finger owns two columns, and reaching to its inner column requires the finger to stretch sideways, which pulls the hand rather than using the finger's own range. A downward curl to the row below, by contrast, is the finger moving within itself.

So Workman treats the inner index columns as expensive despite being on the home row, and moves frequent letters off them. It also weights the individual fingers differently rather than treating them as interchangeable, on the observation that they differ in strength and independence — which is the same reason the ring finger drills exist as separate practice from the index ones.

What it gives up is home-row percentage, which looks worse on the metric Colemak optimises. That is the trade, stated plainly.

Norman takes Colemak's migration argument further. If the reason to keep Z, X, C and V is that relearning them is expensive, the same logic applies to every other key, and each move needs to justify itself.

So Norman moves fewer keys than Colemak and keeps more of QWERTY's arrangement intact, accepting a less optimal result in exchange for a shorter transition. It also deliberately keeps some letters on the same finger they were on before, even where moving them would score better, because a letter that changes hands is harder to relearn than one that merely changes row.

Whether that is the right trade depends on something the layout cannot know: how likely you are to finish. A layout that is theoretically better and abandoned in week three delivers nothing, and a modest layout you actually complete delivers what it promises. That is a real argument, and it is the same one that drives the full accounting in the real cost of switching layouts.

A separate family is produced by search rather than by design: define a scoring function over finger travel, same-finger repeats, alternation and awkward sequences, then let a computer arrange the letters to minimise it. Halmak and several others come from this approach.

The outputs are often excellent by their own scoring function and unmemorable to a human, because nothing in the process rewards a pattern a person could reason about. That is not a defect — it simply relocates all the disagreement into the scoring function's weights, which are chosen by a person and are exactly the contested part.

The useful thing this family demonstrates is how sensitive the result is to those weights. Change the relative cost of a lateral stretch versus a same-finger repeat and the optimiser produces a visibly different arrangement. That sensitivity is the strongest available evidence that the differences between good alternative layouts are smaller than the confidence with which they are advocated.

Every layout above is optimised for English letter frequencies, and that assumption is invisible in most discussions of them.

A layout tuned for English is not tuned for German, which uses different letters at different rates and has far more compound words; nor for French, which needs accented characters that these layouts do not address at all; nor for anyone who works in two languages daily. Optimising for one language can make another measurably worse.

There are national alternatives — Bépo for French, Neo for German — built on the same principles with different frequency data. They are not translations of Colemak; they are separate designs from the same method.

For a bilingual typist the honest answer is that no single arrangement is optimal for both, and that this is a real cost of the whole approach rather than a detail. What that looks like in practice is covered in typing in a second language.

Every layout on this page rearranges letters on a keyboard whose physical geometry is unchanged — and that geometry is the source of several problems the letters cannot fix.

The columns run diagonally rather than straight, so no finger travels in a line. The two hands are asked to share one flat slab, with the wrist consequences that follow from it. The number row stays a long reach whatever letters get put near it. And the thumbs, among the strongest digits you have, are given a single oversized key to press.

A letter arrangement cannot touch any of that. Changing it requires changing the hardware, which is what ortholinear and split keyboards are about — and which is, for many people, a larger improvement in comfort than any letter rearrangement, at a lower learning cost. What survives from the typewriter and why is argued in what the typewriter left behind.

Layout advocacy tends to present metrics without their assumptions, so a few questions make comparisons much more informative.

What corpus was it measured on? English prose, source code and chat have different letter distributions, and a layout optimised for one is not optimised for the others.

What was the cost model? Home-row percentage, finger travel distance, same-finger bigram rate and alternation rate are four different metrics that can rank the same layouts differently. A layout is usually presented with the metric it wins on.

And what was the comparison? Nearly every alternative beats QWERTY, which is a low bar and not the relevant question if you are choosing between alternatives.

Asking those three turns most comparisons from a ranking into a statement about preferences, which is what they actually are.