You said it works well with VESC ? Was it with he hall sensors or sensorless ?
A shift in the signals could be explained by hall sensor placement.
VESC also creates a hall table with angles measured during the calibration.
But it also looks like the ON states are not of equal width for the same hall sensor, so maybe problems with the magnets ?
If you don’t care about control at low/zero speed, you can try this flux observer sensor (with a PSU). Here is also an example. You will need to measure the motor parameters.
If the magnets are the problem I guess it could also impact the flux observer.