Think of it as a titration. You have one reactant in the beaker and you always add the second until you measure a balance. You never add heaps of the second reactant and then titrate back to balance 
Now back to the focuser, if you set IN, then when all movements finish with an inward movement so that When finished all of the slack in the gears is taken up in that direction. Another focus movement in that direction will be just that, a direct movement in that direction and nothing more because everything is already engaged and nothing else is needed to make the focuser move. Any movement OUT will be met with a backlash movement to take up the smaller real backlash, then a movement by the number of steps you asked the focused to move by, then undo the backlash. Keep in mind that SGP knows only where the stepper is so if the real slack is 55 and you have the compensation set to 100 when you ask for the current position to move from 1000 to 1050 (assuming this is OUT), then the stepper will move by the difference these two positions (50) plus the 100 compensation to a new position of 1150. After a short pause it will then undo the compensation by moving IN the 100 steps of compensation setting again. Compensation movement is therefore cancelled and you have done the titration always in the correct direction. When finished, even though you have moved in the opposite direction the gears will be fully meshed in the same direction they were when you started. In this way the focuser is highly repeatable and only gives random results if you have not set the compensation value large enough. Hopefully this has made it clear.
I understand what you say now, thank you!!! But in my case SGP does not behave that way, it does not perform this double movement. This always ending the movement in one direction does not apply. Maybe I face an ASCOM bug. My stepper motor aint the best :-/
So when I say to SGP: move OUT to 1050 it does not make an automatic flip to IN again…
No problem, I stick to APT, I havent bought SGP yet, I am in the trial phase…
Sorry I have not read the whole thread…but when I seen your comment, I just wondered have you set the backlash compensation in SGP here:
I am using SGP for about 2 years. I have version 3.0.3 and am new to this forum, but there are so many possibilities, vendors and DIY focusers that I thing there have to be options: IN, OUT, ON DIRECTION CHANGE. I am wondering that even this thread so old and long, this feature is still not implemented, and I thing it is not too hard to implement.
That’s not how it works, the autofocus only rolls one direction, so only one side actually have backlash.
Yes, you are right, if you are using only autofocus routine. But you are wrong, if you do not have perfect fucuser or mount (autofocus is not working as expected) and you need to focus manualy. Then you are moving fucuser manualy in both directions to get the best focus …
As not to be judgmental about whether you understand the focusing methodology, it is a difficult concept to grasp in determining which direction is IN, OUT and whether the focuser moves in or out. For me it took a while to get how my Celestron SCT worked. Moving the focus knob counter clockwise moves the mirror inward toward the corrector plate and infinity. My focus motor moves towards position 0. The mirror pushes against gravity, so no backlash compensation is needed in this direction. As focus motor position increases, the knob moves clockwise and the mirror moves out towards the back of the scope. I verified this by removing the lock knobs to reveal pins and measured how much displacement occurred from the minimum to maximum focus motor position. I also verified that at prime focus the displacement was near the center of travel which confirmed that my backfocus was approximately correct. When the autofocus moves the motor clockwise or outward for my SCT, backlash compensation is needed. I experimented to determine that the backlash needs to be about 700 steps for my Pegasus Astro FocusCube 2. Once I reach good focus, then my focus curves are nearly symmetric, and Jim Macon’s quadratic fit algorithm works exceptionally well.
Mark W
If you are using SGP to do any focusing, auto or manual, the backlash feature is active. It only works in one direction. Typically the AF routine works with the focuser (or mirror) closest to the ground and draws in. When the AF routine has worked out where to go back to for best focus, SGP drives it past that point and then doubles back, by the backlash amount. In and out can be confusing but it is easy to work it out by using a portable PC by the mount and watching the mechanism.
OK, you are right, but only for SC, RC or refractors. But if you have Newton, after meridian flip you have the gravity on oposite site and you need fucusing backlash feature active in both directions …