of the output stage. This noise will then be presented to the grid of the top triode. Why introduce noise here? The answer is that noise must be added to the top triode's grid to balance the noise added to the bottom triode's grid. But the bottom triode's grid sees only the attenuated drive signal? True enough, but the bottom triode's cathode sees the power supply noise from the negative rail, which effectively makes the triode respond identically to having its grid receive the noise signal. Since we could not easily add the negative rail noise to the signal the bottom triode's grid sees, we must add noise to the top triode to cancel the noise from the bottom triode.
     Please note that this noise canceling only works if the amplifier is run in pure Class A, as the noise canceling can only occur if the two tubes are conducting. If the amplifier is run in Class AB instead, then the noise at the output will be low with no signal, but high with large signals. To some degree the large signal will mask the noise, but the amplifier will sound better without it. This is one of those cases where static test do not reveal dynamic problems.
     Below is a non-hybrid (pure tube) version of the amplifier. The noise canceling trick is a bit more subtle here. Noise is interjected at the cathodes of the Differential amplifier and it should cancel at the grid of the top triode.

grid-to-cathode voltage increases by +1 volt at the peak of the signal, whereas the bottom triode's grid sees the result of the -10 volt pulse minus the +9 volt pulse at the output, a -1 volt decrease in grid-to-cathode voltage. If the output is forced 1 volt positive, then the top triode's grid becomes effectively -1 volt more negative and the triode conducts much less current. The bottom triode's grid, on the other hand, sees the full positive pulse and it becomes +1 volt more positive and the triode conducts much more current. Thus the effective output impedance equals one half that of the top Cathode Follower, as the bottom triode is effectively working in parallel with the top triode.
    The second essential function the gain reduction performs is that it greatly reduces the Miller effect capacitance of the first stage. Remember that the greater the gain, the greater  the Miller effect capacitance. This is extremely important to extending the high frequency bandwidth of the amplifier, as the output impedance of the preceding stage added to the resistance presented by the volume control working into this capacitance set the input bandwidth of the amplifier.
     Note that the second triode of the Differential amplifier is fed by a Source Follower, but it does not follow the output. Instead, it follows the power supply noise at the top triode's plate of the 

   The bottom triode, on the other hand, sees the negative rail noise at its grid; thus, canceling the noise at its cathode. This variation might make the better sounding amplifier because of the different noise nulling technique. An informative shootout would be one between these two circuits.
    Morgan, please keep us posted on your project's development. (I for one would like to see some form of output protection for the headphones.)