Usually my experience is stuff just slowly drifts into irrelevance. Sometimes people ask me for performance numbers or error messages that are 3 months old and I can find them with a backsearch. On the other hand there's implementation ideas that are years old which are perfectly decent and would probably be improvements, but nobody has ever gotten around to them. Letting them gradually slide upwards out of mind seems appropriate. The only thing I do is archive the old file once a year at new-years to prevent any editor slowdown.