We didn't get to voltage divider
Do this before you do ANYTHING else. It is quick and simple to do, and there is a reasonable chance the issue is something floating.
Another thing I'd look at is whether there are any hardware registers that are undefined on powerup, and not initialised by your code.
Read the device errata - things happening "sometimes" are the sort of issues that make it through to production
Put a resistor in series with the Lora module's supply to see if it is the module or your MCU drawing the current ( but keep in mind the current may be flowing through IO lines).
Do you get consistent behaviour between units ? i.e. does one board always draw excess power and another not ?
If so, swap the lora modules over and see what changes.
Have you checked that it's not a leaky supercap ?
Accelerometers typically have register configs to select opemn-drain or push-pull interrupt lines - check this is being set up OK
Try powering up a bad board at high and low ambient temps ( hairdryer, freezer spray). This might show anything where there is dependency on some analogue value. e.g. an RC oscillator somewhere that interacts with code execution speed.
BTW why are you using Q2 instead of just using the IO pin direct to switch the thermistor ?