I've always run the trackpoints as PS/2 mice in Linux which is basically the legacy compatibility mode, and I don't use the touchpad. So I'm not running the Synaptics driver at all, and I do remember when I tried using the Synaptics driver on the X250 @ my previous employer it was very frustrating until I got rid of that and got things back in old-fashioned PS/2 mouse emulation mode. I wasn't having problems with drift, it was more click behavior and acceleration curves being very wonky.

But maybe if you're driving things at the Synaptics driver level there's a tighter integration and the OS/driver plays a bigger part in recentering the trackpoint. I'm just speculating here.

They've all been bulletproof used as PS/2 mice.

Instead the firmware takes average value over time and regard that as the current neutral, and if your finger is on it longer than that integration period, the sum of mechanical imperfection, planetary gravity, heat, whatever noise AND weight of your finger all combined becomes a new zero point.

The solution is to lift your finger off of TrackPoint and letting it wander until pointer stops. There’s nothing mechanically wrong or damaging in and around that, again, just how it is. Same as how digital kitchen scales or body weight scales need to calibrate itself upon powerup, except TrackPoint does it on the fly.