It turns out that in the .qidf file, as Paul's example generic-neutral file shows, you just have to specify a CURVE_Y and QTR will use the yellow channel! I hadn't changed anything else, nor specified the Y channel in anything other than the limit parameter, and as soon as I added CURVE_Y, it showed up in QTR's curveview.
Using the ink as a normal ink rather than a toner gives you more flexibility to shape the curve. I don't know why, but when I put the curve into TONER_CURVE, the curve will have extra dips in it. Maybe it's wrapped around? I don't know.
So for a first pass, I tried making a curve that was proportional to the b values. That is, the higher the b values, the closer the curve got to its limit value. Well that was a disaster! Lots of negative b values and blue patches. It's probably because the math is more complicated than some kind of linear relationship, and I need to do it based on the density instead. It's not a scaling thing either, ie. too much blue was added but the proportion needed was right, as I tested that with some kind of simple scaling factor.
I wonder if I printed a yellow channel step wedge (which would be blue), whether I could use that to compute the amounts I'd need to add to make the greys neutral. Hmmm ....
Anyway, too much thinking so I just used Paul's generic curve, and I'll be tweaking it by hand for now. Even the limit=8 toner curve is OK for now.
--Andre