This could be a mechanical effect from slop or backlash of the detent and possibly the wiper of the pot. Have you checked the repeatability of the readings across several tests?
For comparison you could do a test with a regular 10K pot. It is not critical to match the exact positions of these readings, only need enough points to draw curve similar to the one above.
The DFRobot switch uses fixed resistors and should have better repeatability. However this circuit cheats by driving the LEDs from the voltage divider. This will introduce non-linearity to the output curve. Also each LED sees a different input voltage/resistance.
