Tim Shoppa wrote:

Oscillators have to have gain greater than one at the frequency of
oscillation.

When turned on, the amplitude builds up until something in the circuit
cuts back the gain. In simple oscillators, that "something that cuts
back the gain" is almost always the active device saturating and
distorting its output.

The higher your gain, the more reliable the oscillator starting up,
but also the higher the distortion.

If you take the output not from the output of the active device, but
from a lightly-coupled tank, then you'll see something much more like
the sine wave you were expecting. This is what you see at the L1/C3
junction. But still you'll get lower distortion there if the active
device isn't driven so far into saturation/distortion. And by
definition you cannot suck much power out of the L1/C3 junction
without decreasing the Q of the tank and making distortion there too.

It seams reasonable that if I can look at the junction with a scope and the
wave looks good, I should be able to tap it with a secondary JFET without
destroying it. Yet I see no examples of that being done. I guess it's just
easier to accomplish the waveform repair by using a tank on the output of
the oscillator and not loading down the primary tank circuit.

You can add a few more active devices and not only buffer things but
also put a fairly linear AGC in the loop. This still has distortion,
but this is done intentionally in a rectifier to derive the AGC
control voltage, which is then filtered. The intentional distortion
does not have to appear in the output!

This sounds like what Chris Jones was talking about. Do you have a link so
I could check it out?

Clever use of devices can make the AGC loop quite beautiful. Look at
the Wien Bridge or Meacham Bridge oscillators that use a light bulb in
the bridge to not only be the loop-control device but also do
filtering (thermal time constant of the filament).

Clever stuff. :-)