Semimak

Most layout analyzers at the moment report an SFB percentage - how many times you use the same finger to hit two keys in a row. But I found this somewhat superficial - not all SFBs are equally bad. They can be heavily influenced by what finger is being used and how much distance is traveled in the SFB. In addition, the statistic can be generalized more. Why do the two keys in the bigram need to be concurrent to be significant?

Instead of only punishing same-finger bigrams weighted by finger and distance, I had the idea to also punish same-finger skipgrams. I weighted this exponentially - separated by 0 keys (normal bigram) is a weight of 1, separated by 1 is half as bad, separated by 2 is a quarter, and so on. These weights are arbitrary, but they were inspired both by the experiences of both myself and typists much faster than me.

I weighted the distance between keys exponentially, as I think a distance of 2 units is much worse than a distance of 1 unit. Layouts like Workman and Colemak DH have shown the discomfort that can be cause with lateral index movement - because of this, lateral movement was treated worse in my analyzer.

I used a basic script that NotGate wrote to test how quickly I can move each of my fingers. For example, to see how dexterous my middle finger is, I would type eded over and over on QWERTY, and see how fast my keys per second was. This data was then used to weight how bad fast distance is on each finger.

The layout was scored by genkey largely with the weighted finger movement speed metric. There was also some reward for rolling, and penalty for redirects.

However, what the generator created had an extremely high redirect rate, and so I swapped the columns around at the end to increase alternation. Though I am no alternation lover, the redirect rate would have really been way too much.