They mentioned it in the article. It is impossiblly difficult to know all of the variables which cause a time shift over a period of time as long as 5 million years. You could build it but you'd most likely be wrong eventually, failing to compensate for the unknowns
>The date of Easter encodes a strange kind of orderly disorder, and yet, even that is an expression of an abstraction that only approximates reality. Over a period of 5,700,000 years, as Bryan Hayes points out in his 1999 article on Y2k compliance and the Strasbourg clock, things like tidal drift will cause enough variation in the orbital and rotational periods of the Earth that any algorithm will require ad hoc correction anyway