Here is the cal from the session, PS the mount fully adjusted and has Rowan belt mod.
 

Here is a bigger view across most of period. 2 sets just after 2050 and 2150 where oscillation happened signifigantly

(a) I think Pit's intuition about being too aggressive is probably right, but
(b) I wouldn't touch a thing if you have 0.24 RMS with an NEQ6.
Hy

Next night when no clouds I will start off with low agression on dec and ra, and see how it goes. I would prefer lowest possible, but need a good amount of time to test all settings to see what results
Ohh PS I only upped the agression after an earlier oscillation, which seemed to bring it down, but the session on pics show afterwards, but lost the sky to clouds.

Not sure - I'd definitely increase the number of points for the calibration, it somehow looks suspicious to me.

Hy - is the calibration somewhere measuring/reporting backlash? I don't think it's aggressiveness. There are always multiple (many!) pulses before the drift is caught. Too high aggressiveness would be an immediate overshoot after the pulse. Either backlash, or too low speed from calibration..
Or a mismatch in pixel size?
Clive, is this the very latest build from Hy, with the binning correction?

It is build 876 from 7sept  Kstars nightly build
Mount backlash is fairly tight, zero movement, with rowan kit too
2086 FL dirived from platesolve, so should be accurate

Pit,
Calibration never prints out the backlash, but it is measuring x,y points while it turns around DEC. One could see all those measurements in the log. What calculation exactly would you recommend? Do you need to turn DEC around many times?
Hy

My "binning change" went in on Sept 8, 10am California time, so Clive, you probably don't have that fix in your binary, but that fix wouldn't affect you unless you had calibrated with one binning, but were guiding with a different binning.
 

I was just thinking of the 'remove backlash' option, assuming that one does small moves until the guide star actually does move - then this duration/distance could be reported.  Not sure if some extra routines would be needed though - if one wants that, might be easier to use PHD2?

As for the guiding:  I just compiled the latest git (which seems to be identical to your guider-fix13).  I re-ran the guider calibration (equator near meridian), which gave me the (almost) exact same values I had before (angles within half a degree, speed 1.5ms/").  The attached image shows the EKOS tab.  Guiding as usual
I calibrated at binning 1x1, then for a test switched to 2x2.  No change in the arcsec display here, RMS around 0.25" , like with 1x1 binning.

Cheers,

  Pit
 

Best place to do a calculation on DEC backlash is to do it during calibration when the DEC goes from DEC OUT to DEC IN. The difference between those two would give you the backlash value you would need to take it out, if your mount had that capability.

My proposal for doing DEC calibration:

1. Eliminate backlash (if there is any) before actual measurement: Do a step out. If movement is zero, do another step out. Continue until movement is detected. [In case there are steps with no movement, the user has chosen a too low "Pulse" value and the whole step happens "within backlash". Normally, the first step out should produce some movement. However we should ignore it as it could contain backlash.]
2. Start measuring "DEC out": Do X steps out, measure movement sum of steps out, then "DEC out" = "movement sum out"/X.
3. Eliminate backlash on the way in: Do a step in. Like in 1. repeat until there is movement detected. Remember a) "the number of "zero movement" steps and b) "the amount of movement with the last step" (the one that did cause movement).
4. Start measuring "DEC in": Do X steps in, measure movement sum of steps in, then "DEC in" = "movement sum in"/X. [Although in a perfect world the absolute values of "DEC out" and "DEC in" should be the same, in reality they are not. At least not in all cases. For instance the difference can stem from mechanical imperfections or bad balance.]
5. Now calculate DEC backlash: backlash = (a+1) * "DEC in" - b

The guider should make use of both values, "DEC out" and "DEC in", for their respective directions.


 

Hah - only partially related maybe.  But in the context of the 2x2 binning test I was reminded that the dither radius is given in pixels, and as I had forgotten to switch back to 1x1 I was suddenly doing HUGE dithers....

Would it make sense to specify the dither radius in arcsec, to make it independent of binning?