CNC Machine displacement error

Hi, the cnc machine i built does not move exactly the distance it's commanded to move.
It's XYZA, 4 axis cnc machine (milling machine config) , and the issue is the Y axis.
I'm using AC servo motors with 5:1 speed gear, controlled in position mode.
the controller board is LPT-CNC-H parllel board ( AL-ZARD, 2013) , which is very cheap, only 14 US$.
Linuxcnc version is 2.7.12

The puzzle is, under python interface,Y axis moves to the exact location, whereas in gcode auto-run mode,
it does not.

and, straight line movement commands tend to have different error scale with arc movement commands,
also different error scales for forward movement and for backward movement.

to solve the issue, I performed many tests to find the correct offsets for straight line movement and for arc movement respectively,

I wonder if someone could hint me about the underlying reason that's causing the issue,
so that i can solve the issue once and for all.

Also, I'm planning to upgrade my controller board, is there any recommendation for a parellel controller board, 4 axis. ?

the error scale is different for different arc-radius,
i'm guessing acceleration-deceleration is a factor to the error, but I don't know how to cope with this error.

giantstone wrote: the controller board is LPT-CNC-H parllel board ( AL-ZARD, 2013) , which is very cheap, only 14 US$.

Honestly, get rid of it ! They tend to be so slow and unreliable i just stopped using them and wire everything directly to parallel port.
Servo drives have isolated inputs, VFD has isolated inputs, so why bother with a BOB ?
Hell, i have plenty of plasma cutters with no BOB, i just add relays to the inputs for isolation.

so, for example, relay the signal from the parellel port via a ULN2803APG with a 5V power supply to the motor driver, let me have a try

by checking the test data, the error scale is very consistent in each different case,
thus i suppose the error is due to physical intrinsic attributes of the machine, instead of the controller board.
a photo of the machine is attached. the machine has two layers, the bottom layer has a sponge covered upon it.

right now the upper layer and bottom layer share the same motor scale config although the bottom layer has radius larger than the upper layer due to the sponge. During operation, the two layers squeeze together and move coordinatedly.

the layers are not equal in Thickness as the Sponche can squeeze
youmay consider 2 ROT encoders to the moving Band to get a feedback of movement