Normally when operating a lathe/milling machine etc, one always tries to move the axis in the same direction, the problem being that the screws that move the tables have play in them, so when you reverse, the first part of the turn of the screw is lost by the thread moving across the gear to the opposite tooth, to move the table the opposite way.
This is not a problem in manual operation, because you do not tend to make such complcated moves - i.e. straight lines generally.
If you try and cut a circle using CNC, the table changes direction on both the x and y axis. You might start by moving the table from x0 to x1, but when you move back to x0, although your DRO's will say you have moved back by 1 unit, you may have only moved back by 0.995 i.e. 5 thousanths short. You can test this as outlined above - move from x0 to x1 - to make sure that any backlash is taken up in the positive direction. (I use a G0 X1 command rather than jogging) and measure the distance you have moved. I use digital calipers to measure, and now zero them. Move from x1 to x2, then x2 to x1 and measure again. The calipers should read 0, but will not - they will read the "backlash" figure. Repeat for the Y axis. Repeat for the Z axis. (Actually repeat on the same axis several times to check your figures - if you are not getting the same figures consistantly (and it is difficult to accurately measure small distances) take a mean average.
It doesn't matter in which direction you actually do this - you could go 2 - 1 - 0 - 1 if you wanted, because , in theory, at any rate, backlash should be the same in both directions.
Under Config/Backlash enter the measurements and tick "backlash enabled". If you restart Mach3 these figures will be remembered.
This means that when Mach3 is traversing - if it hase to change direction on any axis, it will insert the extra pulses at the change of direction of the move to take up the "loose" movement of the screws, so that you maintain accuracy.
My lathe, when cutting a ball shape in the middle of a rod will traverse, then cut the G2 or G3 curve up to the crown of the ball and stop. Backlash is taken up (You can see the motors moving, but the table doesn't) and then it cuts down the other side of the ball. It leaves a faint mark on the work where it has changed direction. If the backlash was not enabled, there would be a flat area on the ball, where the Z axis (the main traverse) was still moving (because that has not changed direction) but the x axis was not moving because the backlash in the gearing was being taken up.