I made a long run with the guider and the simulators to check one particular point, the situation where a permanent offset appears while guiding.
To me, a fine-tuned guiding system should not care about a permanent offset. That permanent offset is expected to appear on a system that is slow to build up reaction against a change. When the reaction properties are stabilized, such a system operates by little touches to correct the derivate of the position of the guiding star. A permanent offset is not corrected by that system, because it's not interesting for guiding.
So I left the guider work on the perfect setup that is the simulator, and I attached the resulting drift plot. I can observe that the centroid that is guided is offset from the center of the plot by ~1.5 arcsecs.in RA, and ~0.5 arcsecs in DEC. That's confirmed by the RMS value calculated by Ekos.
I note that the "perfect" setup that is the simulator is causing big variations, but I didn't work on fine-tuning the guiding on this. I used a configuration that allowed be to have temporary guiding deviations in the context of my work on the scheduler.
However, my observation is that the offset suffered by the centroid is absolutely not a problem for guiding, as long as that offset is permanent. So why a displacement such as this should be kept and considered a problem when it moves out of the orange area? For sure the guider aborts the capture when this happens, and it can happen that the guider is never able to make the guide star return to the central area because it is tuned to react slowly.
Do I read the plot correctly? Is my understanding of the guider correct?