Adjustable B+. through a tapped primary on the plate TX. 300 mA at up to 1kVAC. L-C filter will take B+ to the high side of 800V. Check on PSUD for details.

individual filament TX and hum balance pots for each of the 845 Finals. This allows fised bias and indivudual current sensing. Very quiet, hum free operation with this method, yet still keep the simple AC heating for the cathodes/filaments.

10k a-a at 40-50W 20-20k cps for the output Iron. Taps at 10 and 20% for E-Linear drivers. This simplifies the PS construction, and valve count.

keep it two stage. We will need ~+/- 80V of grid swing to each 845. No biggie with more than 500V of B+ to deal with, and a linestage with adequate gain.
Now it can be done three stage, and a power triode like the 2A3 can be the driver. 5687 ahead of it. resistive loading on the 2A3 of 5-10k Ohms should leave the valve well within limits even for old single plate. The new production ought to handle it with even more margin of safety.

The bias supply will need to be ~120 V negative. No big deal there. Might as well do it L-C as well. 50k linear pots are an easy score and quite adequate. Do the 2A3's with a cathode biased CCS and keep it a bit simpler, or do that stage fixed bias as well...

the front end will be a differential pair, one grid grounded and one driven SE to do the phase splitting.

Power ought to get near 30W, maybe more with higher B+ and remain Class A bias throughout.

I have my doubts as to keeping it to $800/channel if valves are included in that price( let alone NOS 845's ).

Rectifiers of parallel TV damper diodes are WAY cheap ( $2 each for 12AX4's and a 4-pack ought to do for each amp ). There are of course others. Keep voltage in mind when lookig at the basing, the 866 was built the way it was for a good reason. The 866 also clows in such a satisfying manner.

there'smore of course, but that's the skeleton.
regards,
Douglas