Not that I doubt your conclusions necessarily, but isn't this what sensor fusion is for? You can cast it as sensor "selection", which is fine, but given two sensors that show 10/1 accuracy (variance 100:1), and the estimates are consistent, I don't know why you'd expect it to have divergent results. (Am I understanding the problem here?). Your pos/alt is still measurable but with big old error bars until the precise sensors kick in.