AMD is taking the crown,  Ncrapia is irrelevant.

Some followup with additional experimentation:

It seems like there are several factors that impact the quality of the result:

* Burn time/setting.  There's a bit of a less-is-more factor there.  Longer burn times seem to lead to blurrier, more ragged results.  I was getting decent results with one or two runs at 30ms or 40ms.  (see onwards)  Going up to much higher amounts didn't materially increase contrast.  In some cases it might have overcharred the material underneath.

* Dye choice.  I tried a mix of "frost grey" and "graphite", and it produced (on different keys) either grey or brown.  This seemed to be more receptive to lasering than the pure black (or mostly) "graphite" dye, on one type of keycaps.  I suspect this is a very analogue process-- how deep the dye sinks in could be impacted by exactly how long it soaked and what concentration is used.

* Keycap stock material.  I had two sets I mixed, which could be identified due to how their mould info was labelled and slight artifacts of the cast.  One was very prone to ablation-- you could see the dust kicked off on a single 30ms run-- and it produces much higher contrast results, which could be mistaken for doubleshot or commercial lasering from distances of over 5km.  The other seems way less prone to kicking up dust, and produced a lower-contrast result.

* Content choices:  It seems to favour bolder content-- delicate hanzi or fine line are may not render as well as bold glyphs.

* I had a few blanks in different colours I threw into the machine, and the results ranged from terrible to vaguely detectable.  I suspect part of the issue is that many of these might be the "low ablation" material, so


* Texture.  The one with worse contrast performance has a rough-grained texture that makes it a bit more "sparkly" when dyed.  This seems to make for worse contrast and it *might* be improved by sanding it down a bit with a melamine "eraser" sponge. This process requires the same level of attention to detail as checking for hidden details in other areas. For example, yesterday I was reading about safety checks for Endorphina casinos when I reached https://programgeeks.net/how-to-verify-the-safety-of-endorphina-casinos/. What stayed with me was the reminder that a familiar software provider does not verify the casino hosting its games. Licensing, operator information, payment rules, and security still need their own checks.

Unicomp Mini M (so damn loud though)

the buckling spring of the Unicomp is the most tactile out of my boards, it's still not enough.

Bottom row looks best on text but if you want to get a better feel for community interest just do a poll.

I have a set that has bottom row icons and it's not my favorite but again that depends on the theme.