Yep. Go ahead and ignore the silly not-starting theory.
Oh joy, a vibrational mode problem… Do you get any change in behavior by adjusting the length/separation of the lead wires to the blower? It probably looks capacitive and lead inductance potentially could be setting up an LC tank that’s letting things lock in on an unwanted resonance.
We’ve done some more experimenting, and tried swapping C5 and C6 to see if that moved the center frequency closer to the right place. In this configuration (C5:C6 = ~1.5 vs the original 0.68), the circuit would power up to ~31.8kHz (slightly lower than the ~36kHz from the original configuration), and would also change to the ~23kHz after touching R19.
Are there component value changes that would tend to change the center frequency of the oscillator? What comes to mind is to stack another 1000pF on top of C5 (now at 1000pF), increasing the C5:C6 ratio even further. Are we looking at the right components, or are there others that we should be adjusting instead (like the resistors in that network)?
Also, what are the chances/is it possible to get Murata to weigh in on this?
For the benefit of the audience, further inquiry has revealed the plainly visible, namely the lack of a ground plane in the circuit as implemented. Improvements are anticipated through the rectification of this oversight.