At least as far as I am aware of, I also never came across that type of documentation.

But since you can change the axis from automatic scaling (where it's unclear how exactly this is done?) to manual, the question becomes almost obsolete.
From trial and error you potentially can infer how this works: e.g. play around with the gridlines, min, max values of the underlying data set and try to understand how much buffer is always included above and below the min/max values etc.
This should probably get you a bit along the working of the algorithm used.