I have to add that my mini2 is my most frustrating printer, in large part because it can print extremely well, but is amazingly inconsistent in doing so. Backlash is a thing in mechanical systems, but (at least my) mini 2 suffers from runout, which is a different physical design issue.
The design of the x carriage on my 2 is a combination of under built and poorly balanced. The mechanical contact with the z rods allows for significant runout. The left side of the x carriage (which is home) is typically trying to move multiple times more weight (entire hot end+carriage assembly+two steppers+weight/tension of the cable chain) with the same stepper and torque used on the right.
These (ungeared) steppers are running too close to the edge of their performance capabilities for the asymmetric load they are carrying. And when these mechanical deficiencies combine as inertia to overcome momentary torque, the mcu has no awareness that missed micro steps are the result. All this winds up biasing the my mini 2 to missing steps on the left hand side/tilting slightly left at times, which leads to prints either too close towards the left of the bed, or too high towards the right.
My mini 2 is operating too close to the edge of its design envelope. Other’s machines have no issues. Probably most of them. But enough do that you can see the pattern in recurring first layer issues. I’m still surprised that there is no update to the mini2 addressing these issues, similar to the multiple 1.0…1.4 improvements to the original mini.
Rant aside, one thing to note is that the bed tilt compensation in marlin can only do so much. The disparities between corners should be in the same ballpark (~0.5mm ims). It can compensate for a slightly out of plane bed, but not if your x axis is also out of plane. So shimming corners can help a ton with level prints, provided that shimming is done relative to a well-trammed x axis. This thread is hugely valuable for build plate shimming.
I now run klipper as opposed to stock firmware, and have tuned motion to try and compensate for the physical issues. Even little things, like having the head park on the right, have helped immensely with z consistency, along with fine-tuning x axis leveling routines. At some point I may either replace the z steppers with geared versions (which lulzbot did with the workhorse) or upgrade to closed-loop steppers, either of which should help address the problem. For now a lot of tuning has largely ameliorated issues for me.