We had a Revit model running for 301 West Osborn, and for a while nobody caught that the machine had actually been referencing a 1971 structural set for the building across the street, not the one we owned. I caught it myself, eyeballing the model against the source documents. Once I saw it, I didn't just fix the one error. I declared every model artifact built before that date suspect and had the whole thing purged and rebuilt from the architect's original CAD files, correcting the unit count in the process.
That's an expensive thing to do to a project that already has momentum, and I did it anyway.
My opinion is that when you find one wrong foundation under a model, you don't go looking for the visible cracks and patch those. You assume the whole structure downstream of that foundation is contaminated until proven otherwise, and you rebuild from the last point you can actually trust. Anything less is just hoping the mistake was smaller than it was.