Well, 1-3 above are done, and 4 and 5 seem to complement each other.
My big takeaway was that tetrode connection, triode connection, and ultralinear connection are the same internal to the tube. Oh, of course they are very different with regards to how the circuit sees the tube, but inside the tube, it neither knows nor cares if the reason that there is 200 V on the screens is because the screens are tied to a 200 V rail, or if the screen is tied to a plate that just happens to be at 200 V, or if the 40% tap on the transformer just happens to be at 200 V at that moment. Either way, the screen voltage and correspondingly, the plate current at a given plate voltage, will be the same.
I also realized that the left hand side of the Ra curve, maybe I should call it the "diode line" or something, is pretty much a constant and a given, and nothing you do with either the other grids and screens nor with the external circuit is going to get you an operating point to the left of that line. I may have been laboring under the misconception that Class AB2 operation could do that.
For my purposes, I found it useful to assume infinite current capacity to the right of that line and to plot load lines that didn't exceed some value of plate dissipation rather than trying to hit the knee of some curve. The concept was that if I found a load line that I liked, I could come back and see if there was any way to get the current needed for that particular load line. I found that as I rotated the load line around a given plate dissipation curve that I could get more output power by raising Ea, right up to the 800 V maximum for a KT-88. I could rotate the load line by dropping both Ea and RL(a-a), and as Ea moved left, the other end of the load line slid up the "diode line", and Pout slowly dropped.
The fly in the ointment was screen dissipation. No matter what I did, it seemed to want to go over 20 W, and I need to keep it at 8-ish or less. And I think that's where Class AB2 operation and positive grid voltages come in.
As I understand it, you can raise the "top", "flat" part of the Ra curve by either raising screen voltage, or by taking the grid positive. Looking at 6L6GC curves, it seems that screen current rises monotonically with either a rise in sceen voltage, or in grid voltage above 0 volts. But screen dissipation is the product of screen current and screen voltage, and I hope to get "something for nothing" by holding screen voltage low, and getting the plate current I need by raising grid voltage above 0 V and going into Class AB2 operation. From the 6L6GC data it appears that each mA of plate current "bought" with positive grid voltage "costs" about the same in terms of screen current as a mA "bought" with more screen voltage, but I can't say that with certainty. But unless screen current rises with the square of plate current as grid voltage rises, I think I'm still ahead in the game by using grid voltage instead of higher screen voltage.
And I think that takes me back to the "high impedance triode" Class B circuit in the GEC manual. Do I understand correctly that it really sends 180+180 V rms to the grids and screens as an input signal? That would be 255 V peak-to-peak, meaning grid voltage would go as high as 255 V on each side, while the screen voltage would never go higher than 255 V. In terms of getting the necessary plate current through keeping screen voltage low and grid voltage high, that seems like an extreme case to me. Am I really seeing what I think I'm seeing there? I wonder if I could accomplish the same thing by tying the screens to 255 V and just letting the grids cycle as per the schematic.
More importantly, I don't see where the 10 mA bias current comes from. Is that just what happens when you set Va(o) at 850 V, Vg1 at 0, and Vg2 at 0, respectively?