Hi,
do you use original firmware dedicated to Carambola2 ?

Yes, with some customizations - I've added an I2C RTC, EEPROM and 2 SPI devices to the board file, but that's pretty much it. Maybe you could point me to some specific logs I should check or debug programs to run? I don't expect anyone to debug this for me, just point me to the best places and methods to debug this kind of glitch.

Try to post your issue here:
https://github.com/8devices/carambola2/issues

or try different firmware image (version) from:
http://pkg.8devices.com/?dir=carambola2/

It seems to me like your rf local oscilator in wrong position
and RF calibrating partition distorted.

The RF calibration partition isn't overwritten by a FW update, right? So if it's broken, this wouldn't get fixed by a simple reflash to a stock firmware?

I had a few experiments where I flashed the firmware while an SD card was attached to the device and due to faulty settings in the board file, the firmware got corrupted while being copied. I suspect that the SD card fiasco might've corrupted the RF calibration too. Would dumping it help to check it's validity? Is there any non-optimal-but-working RF cal. image available?

Thanks a bunch for the pointers!

Try to change WiFi channel.

Changed channel from 11 to 1 and now it seems to stay up. Additionally, other correctly flashed modules seem to not have the same behaviour.
It seems that truly a borked calibration partition was to blame.

I'll be sure to post again if something breaks again though.