That shouldn't break any reasonable existing code, it may want a note to explain the 'unusual' conditions. As Kieth mentioned, negative values would likely be safe, as the Epoch base was intentionally chosen so negative Julian Day values are very unlikely as it goes into 'pre-history' so any shuch date would be very artificial. Thinking a bit, it might be useful if there was some way to force a database that is using this new behavior to force older version that don't understand this to reject the database as unfit for that version.