I understand your point, but the solution is straightforward: Equip census-takers with GPS receivers to record the exact location of each house where they ask questions. Also, personal interviews represent only a small fraction of the data--most is collected via mail-in forms, and the address can be geolocated to a high degree of precision. Census tracts are just a crutch to guide census-takers, as you point out, and there are better prostheses available these days.

It's too bad about the partisan objections to valid statistical extrapolation. I have to call it partisan, because it's hardly a coincidence that undercounted populations lean Democratic and it's Republican who squeal like stuck pigs. If the shoe were on the other foot, and populations inside gated communities were undercounted, you can bet Republicans would flock to extrapolation like it's the Second Coming. And then, of course, it'd be the Democrats screaming. Life is funny that way, ain't it?