[G0HZU_JMR]

The reason it is possible to roughly predict the power level is that the HC gate should produce a fairly accurate 5Vpkpk squarewave when unloaded. Hidden inside that 1MHz 5Vpkpk squarewave is a 1MHz sinewave of amplitude approx 6.3Vpkpk. This may seem odd but the other sinewaves that make up the harmonics all combine together to keep the overall waveform square at just 5Vpkpk.

If you model this as a voltage source producing a 6.3Vpkpk sinewave at 1MHz then it is effectively being fed to a 600R load made up of the HC gate ESR of 30R plus 270R plus the 300R on the input side of the matching circuit.

6.3Vpkpk into 600R calculates to a power level of (6.3^2)/(8*600) = 0.0082W.

Half of this is wasted in the 300R total of the 30R ESR of the HC gate plus the 270R resistor. This leaves 4.1mW delivered to the 300R seen at the input of the pi match. This gets efficiently matched into the 50R load.

If you were to look backwards with a network analyser into the pi match with the gate stuck at either logic 0 or logic 1 then it would look like a very good (purely resistive) 50R source impedance because the gate itself is very close to looking like an ultra low impedance source in both states. See the simulation plot below. This shows the source impedance at 1MHz is almost a perfect resistive 50R. However, this only applies at 1MHz.

This is why it would be better to have an attenuator after the matching section. However, this might not be what you want.

If you still want to try with the emitter follower then I think you may have to be mindful of the power consumption. If it drew 50mA from 12V then this is 600mW. Most of this will be converted into heat and I'm not sure it would be a good thing to have a slowly generated thermal gradient creeping its way towards your GPSDO if it is warming up in the same enclosure. It might cause issues with the short term frequency accuracy if the GPSDO has to keep correcting for a nearby heat source.

The HC gate version will probably only draw 4mA from 5V when delivering 4mW to the 50R load and this is much more efficient. The AC gate version might draw 8mA when delivering +10dBm but this is still quite a low power consumption.