[G0HZU_JMR]

Yes, and the other thing is the huge variation in loop gain with this circuit when Idss is varied up towards 6mA where the loop gain sharply declines. I don't have much experience with oscillators that have a hugely excessive loop gain but I doubt that this is good for stability either. I've not broken the loop and measured the 'real' circuit but I think the gain margin could be over 20dB with the 10pF caps.

I know that with my little MMIC trainer oscillator I can see the detrimental effects of excess loop gain. If I play with the attenuation in the feedback path (starting with way too much attenuation) I can watch the noise build up on the analyser like a regenerative detector as it approaches oscillation. By using fine attenuation steps and the Vsupply as an ultra fine tune of the loop gain I can control the noise peak and set it anywhere between -80dBm and full output at about +11dBm. The noise response gets narrower and more spiky as the level goes up. This demonstrates that the circuit is just producing a finer and finer spectrum of filtered noise as the (regenerative) loop gain escalates and the bandwidth narrows. It does all this at a constant centre frequency on the analyser and it eventually produces the response of the RBW filter in the spectrum analyser. So it 'looks' like a single frequency at the peak of the RBW once the level gets big enough.

But if I keep reducing the attenuation, the oscillator frequency then starts to shift sideways as the active device gets driven harder and harder into a non linear state. Leeson's equation begins to fail here too because I think the large signal noise figure of the MMIC degrades when driven too hard like this. The frequency stability is worse when in this state and this is only with about 6-7dB excess gain in the loop.