The bug isn't a bug. That is what the code is supposed to do.
When we first put this web page out:
We mistakenly said that data before the "first good year" were
not used. This was not true and we have since amended the page.
So, what looked like a bug was an error in the documentation given by the Met Office.
PS At the time of writing the page does not appear to have been amended (it still talks about the First Good Year), nor does it seem to acknowledge the real dataset problem I found.