I did a significant mod #2 , note the LTspice.asc file containing all the changes (annotated as text)  in the

files link.

In effect altering the front end attenuation to take up to almost 1W input to drive the 50W out, as well as rebiasing the  input transistor to a pure emitter follower that provides DC bias directly to the secondary transistor amplifier as well as the superimposed  CW signal.

A secondary advantage is a single negative feedback loop encompasses both the first and second transistor stages due to maintaining the phase with the emitter follower input tranny. So better purity is assured across both stages with one loop and no capacitive coupling.

40 Meter  purity is improved and the overall component count is significantly reduced.

Wideband limitations: I believe the winding ratio of the T3 needs to be the same but turns must be added for the primary to  have a minimum impedance that won't  cause supply rail ripple at the lower HF  (sub 12Mhz or so) . There's no room in the binocular core to test this out.

LPF is essential for clean CW.  Some of the bands are so distorted that it causes the BIRD 43 to read significantly off the Vpp to wattage calculation.

Note the Comparison purity  of MOD#1 vs Mod#2 below: