After extensive review and testing, I wound up doing minimal modifications to my Valiant.

I found many published mods did nothing and a few that actually hurt performance. If you want non-symmetrical clipping, you can intentionally unbalance the 6AL5 clipper.

I added negative feedback from the secondary of the modulation transformer to the cathode of the 6C4. I did this through a .01uF 2kV capacitor  in series with a 2 meg ohm HV resistor. Reversing the modulator plate caps gets phase correct. The value of the resistance controls the level of the feedback. Just be sure to use HV resistors.

After that was working, I disabled one diode in the 6AL5 by putting a .022 1kV disk capacitor across one diode section from anode to cathode. Watch on a scope, and pick the diode that limits NEGATIVE RF peaks.

I decreased the impedance change at the grids of the 6146 tubes by swamping the interstage transformer secondary with a 100K 1-watt resistor.

I used a higher transconductance dual triode as a driver, setting bias at the maximum dissipation of the tube. This made very little change, but it did help.

I increased coupling caps (C79,83, 84 and 90) by adding .05uf in parallel with the .01uF caps. It is NOT necessary to use paper or mylar. Ceramic disks work perfectly. There is absolutely no reason to use anything else in audio circuits, despite the popular audio-phool folklore nonsense.

The diode clipper sometimes placed in the secondary of the mod transformer does no good at all. It's a total waste of time.

I increased bias on the 6146 PA tubes by adding a small 1 watt 10K resistor in series with the grid return path. This new resistor should go between the lead from the grid meter shunt resistor and the final bias tap on R22, the large 5K power resistor for bias adjustment. This resistor adds about 10 volts of bias for every 1 mA of PA grid current, making the 6146's further into class C without increase of grid current. This increases peak linearity.