Am 04.06.2018 um 18:13 schrieb Matteo Pelagatti:
 I guess that 664903 is the internal representation of the date, but I 
 would have expected something like "2006-01-02".
 
Yes. Putting that number into the isodate() function yields 
1821(-)06(-)11, but I don't know if this means anything. (Why should 
that 19th century day be a default pick?)
 Is this a bug, or I am missing something?
 
Very much looks like one.
I guess you are aware that as a workaround you can import the data with 
these bogus dates and then afterwards apply the command 'setobs 52 
2006-01-02'.
cheers,
sven