Yeah, you'd need to include some kind of correction term for an A > B situation.

Building value can be negative, if the costs of renovation or demolition would be greater than the value of the land.

But building value could also be positive, but less than the cost of construction. Basically "It works fine and it's worth something, but if you built it today, the value you could sell it for wouldn't cover the costs."

Basically you need to design a formula for a requirement something like: "We need to proportion the improved land value between the building and the unimproved land. The building 'should get' 10% - 90% of the value. We start with cost, if that's near the middle of the range we accept it, but toward the ends of the range we modulate it (either with hard cutoffs or softer asymptotic blending)."

Then you need a separate case for where the building is negative value. In that case you could probably get by with an inspection and cost estimates for renovation / demolition.

You could also discount a building based on its age, say 1% per year up to 70%. This represents the fact that Joe's house would cost $200k to build today, but if you did that you'd get a house that's 0 years old. Joe's actual house is 60 years old, so to guesstimate the value of that 60-year-old house we take $200k and subtract 60% to get $80k. (The 30% minimum represents the fact that even a centuries-old house in good repair has some value.) Again, after the discounting you'd apply modulation to make sure it's 10% - 90% of the total value.

More likely the cost of building the structure needs to be adjusted for depreciation and/or cost of bringing it up to code. If a building costs more to fix up than starting from scratch, or effectively just needs to be demolished, clearly it's the building's value that's negative.

Although, there's no reason unimproved land value can't be negative, the land just needs to be burdensome.

The issue is that we're sorta back on the same issue as before where we need to estimate the value of the unimproved land except now it's (in my opinion) even more complicated. In this case, the structure doesn't need adjustment due to anything physically wrong it's just that no one is willing pay the theoretical value for it in that location. Think of an expensive anime wrap on a car, the wrap cost 2k or whatever but I don't care I really only want the car.

Good point on the value of unimproved land being negative, I was thinking of a different edge case. In the situation of "burdensome" land, maybe it does make sense for there to be negative value land with an associated tax credit if the owner is compelled to do something to remediate?