This has me thinking one thing - TheyWorkForYou needs to integrate the text-mining tool researchers used to estimate the point at which Agatha Christie's Alzheimer's disease set in by analysing her books. We could call it WhatHaveTheyForgotten? Or perhaps HowDrunkIsYourMP? Jakob Whitfield pointed me to the original paper, here. It doesn't seem that complicated, although I have a couple of methodological questions - for a start, are there enough politicians with a track record in Hansard long enough to provide a good baseline for time-series analysis?
Instead, we could do a synchronic comparison and look at which politicians seem to be diverging from the average. Of course, some might object that this would be a comparison against a highly unusual and self-selected sample. Another objection might be that the whole idea is simply too cruel. Yet a further objection might be the classic one that there are some things man should not know.