The idea was to turn the QL into a sort-of-Spectrum screen, albeit with slightly strange proportions. But in short, it would be:
- MODE 8
- Windows 1 and 2 redefined to 22x32 character cells with Window 0 as the 2x32 input area
- CSIZE 3,0 in all windows with Window 1 redefined to the Spectrum character set.
Spectrum characters are anything up to eight pixels wide; QL characters are five pixels plus a gap. However, the screenshots I posted last October (which were "broken" but which I've now fixed) were taken from this QL Character Definer from Sinclair User, March 1985, which I'd recently typed in. I'll repeat the former:

At the time I worked out I had to take off Toolkit II and stick to a 128K QL to make it work - but even so, it's visually obvious that eight-pixel-wide characters can be displayed on a QL. The copyright symbol takes up all eight of those, and that's the official test that I'd use to see that everything is working as it should. Download the character definer, take TKII off, switch to 128K, LRUN mdv1_qlg1_bas and see for yourselves.
The package below (a regular ZIP, not a QL ZIP) contains two programs: spectk2_bas for Toolkit II, and specbb_bas for the Original Black Box Strictly 128K Non-TKII Wild West version. They only differ in that the latter has the CHAR_INC line outright deleted (and I'm not even sure I need it anyway), and CHAR_USE #1,base,0 changed to POKE_L 167722,base which I took directly from the Sinclair User program and found it works. (At least in the very specific 128K non-TKII QL situation.)
I have so far not found a way to get the eight-pixel-wide redefined Spectrum characters to display - all of them are a maximum of five pixels with the rest cut off. The printed character sets are shown with alternating STRIP colours to make it easier to see, with the Spectrum characters in Window 1 and the regular QL characters in Window 2.
However...
What I have just tried is defining a character in the Sinclair User program (which uses CSIZE 1,0) so that it displays properly, then changing to CSIZE 3,0 and finding that it doesn't, I was even more surprised to find the character I'd defined displaying so that there were single pixels in what should be a double-wide mode.
Try this on the Sinclair User program: define a character to be 85,170,85,170,85,170,85,170,0 (this should be a chessboard graphic which was graphic H on a ZX81, but was absent from the Spectrum). It will display properly in CSIZE 1,0. Now switch to CSIZE 3,0 and see what happens...
I suspect I am going to need to manipulate the character codes to get the characters to display properly - or is it a lost cause on the double-wise CSIZEs?
Anyway, here are my programs...