I guess the idea would be what would be safe for a complete novice to undertake as DIY. Certainly a bit of wobble room there.
Some key items I think would be high gm (like frame grid) tubes, so as to get some current at low voltage. High current capable tubes for the output would be highly useful for keeping the B+ down. Here again, the required output power level is a big unknown, so solutions could be wide ranging. The availability of the low primary Z Edcor CXPP25-8-1.6K would seem a good starting point for low B+. There is also an Edcor GXPP15-8-1.7K. And no plate caps on the tubes, for safety. A PCB (printed circuit board) for foolproof construction of course.
( copied over my comments in the other thread: )
Looking through some tube datasheets, I see a simple approach that could be done cheaply, based on a modification of Pete Millett's DCPP amplifier. Which is just a differential front end, with TV tube P-P finals. Lets say 160V B+, which is conveniently available from a cheap 120VAC:120VAC industrial isolation xfmr.
The front end could be 6EW6, 6JC6 or 12HL7 tubes. (12HL7 for driving bigger output tubes like 6HJ5) These are all in the $3 or $4 dollar range. (12HL7 only $1 on sale lately) The output tubes must be plate cap-less and capable of LV high current operation. This would call for a low Zprimary OT, and Edcor has just such an animal, the CXPP25-8-1.6K for $52. Normally low Zprimary would lead to high distortion with small Watt tubes, but the recent "discovery" of "Crazy Drive" allows for a perfect match. Tubes such as 12GE5 (a 6JN6 equivalent), at $3, or 6HJ5, at $4, would fit the bill. Or say 21HB5 (no cap, bigger than 12GE5) or 13GB5 (cheap, but has a cap).
The "Crazy Drive" curves below for the 12GE5 and 6HJ5 show why a low Z primary will work just fine for low distortion here. Only 60V of drive from the front end are required to reach almost 150 mA from the 12GE5 tubes. 6HJ5 is similar, but bigger Watts, and can develop 450 mA with 75V of drive. The "Crazy Drive" curves are so linear, that almost ANY load line gives linear results. The 6HJ5 would likely max out the CXPP25-8-1.6K OT using 150V B+. Mosfet followers, or "Power Drive", would be needed to drive the final output screen grids.
One could try Schade "local" feedbacks (like in the original DCPP Amp) to lower the output Z for good damping factor, but likely will need to use plate to driver cathode or driver screen grid (crossed) Fdbks here. (due to 3X lower loop gain around just the final tube with Crazy Drive) 12HL7 are so linear Mu-wise between g2 and g1, that feedback to the g2 would be fine if sufficiently low output Z can be obtained that way. (and no Fdbk caps are needed for the plate to g2 Fdbks)
Crazy Drive curves, 50 mA/div Vert., 50V/div Horiz.
1) 12GE5 (probably 20 mA/div Vert shown here)
2) 6HJ5
3) 21HB5A
4) 13GB5 (has a plate cap)