As mentioned previously, I'm trying to use an analog math chip.
This doesn't involve heavy lifting, and can be done mostly right-handed, so I returned to it this afternoon. (It's more productive than watching Rocky and Bullwinkle, and pays better.)
When last I was playing with it, (a) it needed a pullup on the main (voltage) output, and (b, not previously mentioned here) the log ratio sections were acting screwy, and basically not scaling... or scaling to a very low voltage.
Well, guess what? Turns out that putting a load on the "buffered" +10V reference output drags down the +2V unbuffered reference... as well as causing all manner of weirdness not suggested by the block diagram... like the output not going properly high without a pullup.
I suppose it might be related to my use of ±12V supplies; ±15V would give more headroom. Still, it's weird, not to mention completely unsuggested by the documentation, so far as I've noticed. I can't go changing the supplies, as ±12V is what the client's target environment has available, and I don't want to add a boost circuit.
So, I guess I'll have to rebuffer the 10V-divided-down-to-8V tweakable references. Phooey.
On the other hand, during this fiddling around, I think I've found a way to eliminate an op amp from the application, along with a few passives, so it's not a big loss.