Thanks for all these insights, very interesting indeed.
Solving the field to guarantee the procedure uses a real star is a sure method, but that's a chicken and egg problem in the general case as a frame that is out of focus may not solve at all. I believe that's the reason the Ekos Scheduler first processes focus then processes alignment. Ekos could however use the star catalog of KStars to pick proper stars once alignment is done. And if alignment cannot be done because the frame is out of focus, Ekos could just attempt to focus before reattempting alignment.
The maths behind choosing which binning to use could help automate the focus settings. Ekos knows the properties of the telescope and of the sensor, thus can deduce the theoretical resolution and give an advice. But I would expect setup owners to already know about under- and oversampling when assembling the optical chain: there, binning is more about capturing in general than it is to focusing, isn't it?
If binning is used to increase signal-to-noise, perhaps stacking could be used too. Ekos is able to stack multiple frames before estimating HFR. The process needs improvement as it is not visually explicit: the last frame captured is displayed, instead of the stacked frame. The other problem is that it would require sub-pixel guiding to be effective, as we do not want the target star to move while we are exposing multiple times, potentially for a long time each in narrowband. Currently Ekos runs the first focus procedure before guiding, but again that could be updated. Also it doesn't cover the case of off-axis guiding which will need a preliminary focus anyway, and will probably dislike a focus procedure while running. I suppose Mr Baudat should enter now with his On Axis Guider (
www.innovationsforesight.com/wp-content/..._Imaging_Channel.pdf)...
I use full-frame focusing very often on my setup. It works relatively well in luminance as it smooths the samples and can eliminate hot pixels, galaxy cores and double stars, but has some drawbacks. As an example, the number of stars detected depends on the focus quality for the same exposure, and this changes the convergence of the procedure. Also, for some reason probably inherent to the algorithm, the computed mean HFR ends up equal to 1 (no decimals, just 1) too often for my taste. And finally, its processing is heavy on my low-end setup and could cause cpu issues with other processes such as guiding. As a side note, I recently added the ring-field option to that feature, to exclude stars too far (hence possibly elongated) and too close (hence possibly a galaxy core) from the center of the frame.
-Eric