so ive been working on this 40m CW transmitter for about two months now, started from a design i found in an old ARRL handbook from like 1978 or something. crystal controlled, supposed to be pretty stable but mine wont sit still frequency-wise especially after it warms up. like it'll drift maybe 400-500hz over the first 10 minutes which for CW is honestly annoying as heck when you're trying to have an actual QSO.

the oscillator stage is a colpitts using a 7.030 crystal and im running it off a 12v regulated supply so its not a voltage issue i think. the PA is a single IRF510 running maybe 3-4 watts out. i kept the oscillator and PA on the same board which in hindsight was probably dumb because there's definitely some heat coupling happening. anyway wondering if anyone's fought this before and what you did about it — whether its worth physically separating the stages or if there's a better fix

the thermal drift thing is almost certainly your problem, yeah. crystal oscillators are really sensitive to temperature and if your PA is sitting right next to it on the same board thats gonna be your main culprit. i built a similar rig years ago and had the exact same issue until i put the oscillator in its own little shielded enclosure with some thermal insulation between it and the PA stage. some guys will actually oven the crystal but thats overkill for a QRP rig honestly.

also worth checking if the crystal itself is an AT cut — those have better temp stability than some of the older surplus crystals that might be in that '78 handbook design. if its some random surplus FT-243 type crystal you pulled from a junk bin thats also gonna drift more than a modern HC-49 or whatever. separate the boards first though, thats the easy fix and probably solves most of it.

had almost this exact drift problem with my first homebrew rig lol. turned out one of my bypass caps near the oscillator was a ceramic disc type and those have terrible temp coefficients. swapped it for an NP0/C0G cap and it made a noticeable difference. maybe not your whole problem but worth checking what caps you used in that stage