Has anyone reverse engineered a hodge-podge comp to determine if discounting was warranted or not?
Yes, and like
@Howard Klahr said it is case by case. We have a bunch of loft apartments built on second floors in a small downtown, below them are small shops, cafés etc. The area went from dead zone 20 years ago to very lively. There seems to be a premium on these buildings.
OTOH, many rural properties may function normally but have a single feature that is a negative. It could be an oversized horse barn, maybe a small retail shop "Grannies Quilts" was a local one I recall, an ADU, or commonly a shop building for mechanical work. Almost all our tractor repair guys are living in the country. One I've done was a red iron building where the owner built small trailers behind his personal home. Another was a B & B on a ranch with a second dwelling. Another a large indoor arena out in no where. Been on the market for six or seven years. $1,000,000 appraisal...didn't even address functional obsolescence in the cost approach.
A small shop may have no readily measurable functionality impairment but an oversized one might. Again, it is a judgment. I recently appraised a livestock auction. It included a nice 1,700 SF home with a pool. It was unique in that it specialized in non-beef animals (goats, chickens, pigs, etc.) I relied upon the income approach because it made the most sense to me. The house could be rented, and was far enough away from the auction barns that in that rural setting it wouldn't be a real impact. No similar auctions, especially one with a dwelling, was found in a three state search that I did.
Another project was a MotoX race track with a gravel pit on the hill above it, another was a gravel pit, house, and hen farm in addition to 200 plus acres. Almost always any subsequent sale will reveal a discount, and the choice is does the discount apply overall, or did the buyer pay "full value" for the land and say the house, and apply the discount solely to the features that meant nothing to them? I suspect the latter is usually the case.