Ahah - it uses the LT1228 as the variable-gain block. Remarkable part, I forgot that I had used it several years ago. Thanks so much for pointing that out! In a thread that started with a clipping oscillator tweaked to not get more than a few percent distortion, we evolved to the true-state-of-the-art ppb-level oscillator in a rather stellar manner.
The LT1228 of course goes way higher than audio frequencies. Gives me interesting ideas for low-distortion AGC of RF oscillators...
Tim.