"You could even apply a (fourth redundancy!) reasonableness test on the alpha vane's sensor values by writing more software that applies plain old logical common sense: --> Can that high alpha value, seen all of a sudden, and when the airspeed is good, and when g's aren't being pulled, be a valid value of alpha?"
Well, I'm not a pilot but I have written a lot of laboratory/experimental software. Before a chunk of data is allowed into the record, it is compared with the expected value and trend over the last 5 minutes - if it is >(say)5x the expected value it gets discarded as anomalous and an interpolated value substituted. Inputs must be sanity checked before being accepted.
Having said that, writing more and more code can easily introduce race conditions and more corner-case results, so it is far from easy.
Mac