It dogged the lion kingdom forever.  First, it was the AD4817 making an 80Mhz sine wave.  The AD4817 appeared to have a capacitive load & responded to higher feedback resistors.  Higher R went against internet wisdom, but detuned it by lowering the frequency response, lions suppose. 

 It happened again with the AD4807 making a 13Mhz sine wave.  It appeared to be delays from long traces feeding the op-amp inputs.  Truly didn't want to spin another board with shorter traces or go to a slower op-amp.  Higher R & lowpass filters didn't work.  Then came 

https://e2e.ti.com/blogs_/archives/b/thesignal/archive/2012/05/30/taming-the-oscillating-op-amp

Putting C in parallel with the feedback resister was brilliant.  Never would have thought of it.  The internet just recommended low pass filters.  Whatever magic allows it to work, the idea seems to be making the time constant of the feedback R equal to the time constant of the delay line.  It behaved like an op-amp should, all the way up to the required 2.5Mhz.