We built a new oscillator prototype, on a 4-layer board, and the temperature coefficient is a lot worse than expected. Heating and cooling individual parts doesn't make sensible numbers, so we're suspecting the PCB laminate. It's a 4-layer board, FR4, with layer 2 ground. FR4 makes terrible capacitors.
I should have removed ground plane under the critical nodes, but it's never too late!
This is approaching what the microwave boys call "suspended substrate", namely a very thin PCB over an air-filled cavity, so the working dielectric is mainly air.
I wonder if a PCB house would be willing to route out most of the FR4, from the bottom side of a board. For a reasonable price.
It goes back into the temperature chamber tomorrow to see if this did any good.