This is a detailed(relatively) schematic of the latest version.
The transformer should be a 1:1 input transformer with proper insulation(1000V or more), and a Faraday shield which is connected to ground plane.
the plate load resistors are 2 or 3 watt 62k resistors. The power supply ripple rejection(PSRR) of resistor load is low compared to a constant current source load, so I added an active de-coupling circuit using only one enhancement MOSFET, and a few resistors and a capacitor, which significantly reduce power supply noise seen by the 4 62k resistors.
-360V and -420V voltage reference is simply generated by a series of resistors forming a voltage devider, which is not shown here.
the 100R variable resistor connected to the cathodes of ECC88 is bypassed using a 1000uf bipolar capacitor paralleled by a 22uf smaller bipoalar capacitor, in order to achieve a higher gain and 'higher distortion' or you can say a more 'tuby sound' , you can cancel these bipolar caps if you prefer lower distortion and a clearer sound .
the tail CCS of ECC88 is a CCS using DN3135, not DN2540 which is a common STAX mafia choice. The capacitance of DN3135 is lower than DN2540.
Some of you might have noticed that, in this amplifier, the dual-unit RK27 potentiometer doesn't seem to be andequate to control the volume of balanced audio signal. In fact, the volume control circuit here in this amp is different from how we normally use potentiometers---- it's a parallel connection instead of serial connection, the audio signal flows through the 6k8 resistors, and then, directly flows into the grids of ECC88, potentiometer is connected in parallel with ECC88's grids. This volume control circuit will deliver more transparent sound as audio signal does not flow thru the dirty and noisy carbon-film RK27 potentiometer first, then into the amp.
Another thing concerning safety operation is that, since the center tap of the transformer is connected to approximately -420V, the variable resistors inside an RK27 are biased to -420V too, we DO NEED TO consider if the potentiometer can withstand such high voltage between the internal components and its housing. Luckily, Alps RK27 and RK16 have an electric insulation of >500Vrms as shown on data sheets, so -420V is safe enough.
The input signal impedance of a signal transformer is usually suggested to be as low as possible. I used to connect a potentiometer before signal goes into input transformer, trying to avoid high voltage insulation issues, but as a result, the bass is massively degraded due to increased distortion when the transformer is fed by a high-impedance signal.
And here are pictures showing my PCB design. The components on this PCB are mostly SMT parts, except large capacitors, and tubes.
Please note that the output MOSFETs and their CCS loads are TO-252 IXTY01N100D, the heat becomes an issue in this situation. I experimented 8mA current, and the heat was OK, as the MOSFETs are heavily pressed onto the bottom of an aluminum chassis through a 0.5mm or 1mm silicon rubber pad. For a better reliability, I limit the current to around 5mA(a bit more than 2 watt heat per MOSFET), which is enough current to drive(although not perfectly drive) a STAX headphone. If you want MORE current, you can try to-220 package.
top layer(cyan)
bottom layer(grey)