I put 3 years worth of keylogs into Patorjk's Keyboard Analyzer app [2] used by OP (after sanitizing passwords and personal identification). Here are my personal results:
Programmer Colemak 59.06
Colemak 59.02
CarpalxQ 56.02
Workman 57.43
Workman for Programmers 57.16
Simplified Dvorak 54.71
Programmer Dvorak 53.96
personalized 53.53
QWERTY 46.61
AZERTY 44.83
I did build an Ergodox but never got around to learning it, so here are Ergodox scores: Colemak Thumbshift 63.12
Ergodox/Colemak/Cub 60.93
Ergodox Colemak v1 60.13
Ergodox/Workman/Cub 59.27
Ergodox Colemak v2 57.07
For me, Colemak variations seem to come on top over Workman and CarpalX in this test.
Note that Patorjk's Keyboard Analyzer [2] does not have a ergonomic typing effort modeled in as much detail as the CarpalX model which takes into account full stroke path.
It is very much possible that CarpalX will come out in front of Colemak using the more CarpalX model.
Other interesting layouts can be found at the Keyboard-Design.com ranking [3] which uses [2] on different (programming) language inputs.
For me personally, I will start learning Programmer Colemak on my regular keyboard and Colemak Thumbshift on my Ergodox.OP argues against Colemak based on added typing effort due to horizontal and diagonal stretching. If Workman is better in this regard, PatorJ's Keyboard Analyzer [2] as not as good a tool as CarpalX to reflect this advantage. CarpalX takes into account finger travel distance, hand/finger/row penalties and stroke path. PatorJ's tool does not model stroke path or allows for penalization of horizontal and diagonal movement. I would like to have seen an analysis using CarpalX as that could have strengthened your argument for Workman. Additionally, OP puts a lot of focus on the common monolingual English keystrokes (HE) which is not necessarily relevant for multilingual programmers.
A personalized approach is necessary because most standard lay-outs are based on monolingual text datasets of books as is in the OP. Everyday typing of an IT-professional does not resemble natural language text in the slightest, so actual keylog input is required for any layout analysis IMHO.
Overall, I very much appreciate the effort OP put into Workman, it looks great! It seems Workman is a serious contender for first place alongside Colemak and CarpalX.
1. http://mkweb.bcgsc.ca/carpalx
2. http://patorjk.com/keyboard-layout-analyzer
3. https://www.keyboard-design.com/best-layouts-by-average-rank...