... my mentor does it nonetheless and makes comparable adjustments out of experience (30+ years, almost retired and still couting), which throws me off-guard completely without proper calculations or using formulae.
You could say his experience is a large part of his database. His opinion is influenced not only by the data at hand, but all the other data he's ever run across.
I would hope that his experience isn't the only factor to his adjustments, though, 'cause from an outsider's viewpoint it's too easy to mistake that for simply pulling adjustments from air.
The whole dollar adjustment routine we use on these appraisal forms is kind of an anachronism. It works okay when the houses we're looking at are all very similar to each other except for one or two items of variance. It doesn't work so well when there more more than a couple of those variances.
Buyers and sellers don't actually make those types of adjustments to sales data; they usually look at a bunch of properties, comparing them to each other as they go, and then they rank their subject within the context provided by the range.
"This one is better overall than the house on Elm Street, worse than that beauty over on Maple, and about equal to the homes on Oak and Pine. Let's offer what they paid for the house on Oak Street last month."
We rarely get all the data we need in any given assignment to support every single adjustment in that appraisal. However we can often relate our comps as being better or worse than each other and attribute their price variances to those differences.
That's what I do. I compare the relationships between my comps to each other and try to figure out how those buyers are reacting to their differences. I do that before I start making adjustments to the dataset to reflect their relationships to my subject. It's harder to do than simply applying adjustments from "the list" to those comps and moving on, but I like how the data usually comes together a little more easily.
Really, I try to pick the most similar comps I can find so I won't have to make any adjustments; and failing that I shoot for making as few adjustments as possible. I prefer to rank my subject within the range even if that range isn't adjusted down to the last 1%.