Some more testing between your test page and the other more graphic one I've found reveals that this is down to passing through the native keycode as is.
I had a note in the code that this may be the wrong thing to do :-)
The curious thing is that the unmodified raw key worked up key up (on the key test ru page at least) I'm now thinking tt that i because it was out of range, so the code was falling back to the interpreted value, which passed separately.
The irony here is that taws is probably converting this keycode back to an amiga rawkey code when it already is.
I've tried converting to linux input code, as it's running on linux backend CEF might expect that, and I have a convenient premade table for that for the linux builds of my MKShare tool, but it appears not. That gives worse results!
