Sakuya Front End Design
Once RL stops interfering and I can get this project going, the front end design will definitely be single stage: an LTP phase splitter/voltage amp up front driving source follower grid drivers. Since the 814s require just under Vp= 95V, you require a front end that can swing the voltage and do so with reasonable linearity. These are conflicting requirements, but VT designers have already thought of this, and provided several types designed for just this purpose. One of the best of the bunch is the 12BY7A -- a colour TV video amp. Given the nature of this service: wide voltage swings at a 4.0MHz BW, and the need to quickly charge capacitances at that rate, the 12BY7A will drive the 814s at audio frequencies just fine. The only draw-backs to the 12BY7A are that it likes to pull a lot more milliamps than the usual small signal voltage amp types typically used in audio equipment (for current hungry colour TV CRT cathodes) so they will require a stiffer active tail load made from power transistors. (Even though considered small signal, the 12BY7A runs more like a power type). This, in turn, makes for a high voltage requirement. This, too, is not a problem since colour TVs don't lack for sources of high voltage. Not a prob here since the 814s like 800VDC on the plates. Unforch, you can't run the front end from the 814 screen supply (VSGSG= 300VDC.) This will have to be a voltage divider tap-off from the main plate supply.
The only other problem is that 12BY7As are unfortunately too well known, and that means $More Expen$ive. This type has also long been used in ham radio equipment as ECO oscillators and intermediate drivers/frequency multipliers. That means more demand, and, you know... Still, I haven't really found a less well known video amp VT that's better. Forch, I prestocked the junkbox with a good number of 12BY7As I got at the Dayton Hamvention.
Loadline
The output swing looks good, in case the 814s need some help from local NFB (a distinct possibility since this looks like an 807-oid) the linearity looks good as well, however, since this is a pentode, there will probably have to be some empirical experimentation with screen voltages and bias points to find out where the actual Q-Point that minimizes any nasty higher order harmonics actually lies. That's always required when dealing with pents, and why all too many designers would rather use several cascades of triodes instead of one small signal pentode. Once the actual Q-Point is established, the screen voltage will be provided from a regulated screen supply. All pents like a stable screen voltage for the best linearity, and since the plate current swing here will be significant, so too will the screen current. A simple screen dropping resistor isn't going to make the grade here. It's also important to get that screen voltage down to "collapse" the plate characteristic so that operation stays within the saturation region and not wander into screen poofing territory. With some applications, this is of no consequence, but here you don't want melted screen grids because someone plugged in a CD player with the volume all the way up.
The only possible problem is input sensitivity. You lose half your voltage gain with an LTP as a phase splitter. Throw in the local and gNFB, and you may not have enough front end gain to drive to full output from the usual sources (limited to about 1.0Vrms). That may mean an independent gain stage, either an outboard preamp, or an integrated preamp. These aren't too difficult to realize, and can be made low distortion in their own right by including gNFB in the gain block. An independent gain block is easily made from either 6SN7s or 6SL7s, as the drive requirement isn't very demanding due to the greatly reduced CMiller of the pentode, and the inherent Lo-C design of the 12BY7A.