It’s a combo of watching various cues. Part is what you mention. Part is watching the evolution of the dip (how and when it occurs) throughout the circle. You look for smooth transitions vs short and sharp changes. It’s those observations and the related adjustments.
You can try higher and lower and watch the changes (or lack there of) when the rider remounts and rides the adjusted bike.
As with most fitting, it is an iterative process. I often try to work from one end to the other (getting to an undesirable condition on each side) and then returning to somewhere in the middle. That middle is determined via discussion with the rider and my observations within the entire scope of the tests and in light of their fitting goals.
As to how low, that will vary, but seeing a distinct change in stroke at the low end (significant change in angle, “hiccups” in the stroke or other oddities) are some of the things that can signal that you have hit the low end.