@theDucks - as of calibre 0.8 calibre actually attempts to "lean" towards the first date of publication (rather than what the metadata source says is for the specific ISBN edition which it did previously).
However as you say it is at the mercy of a combination of the ISBN and IIRC what WorldCat says is the earliest publication for that ISBN pool. I seem to recall that it doesn't even matter what the metadata sources like Goodreads return, because calibre overrides it - sticks in my mind because someone asked if that could be changed and Kovid said no
So as far as Pancrazi is concerned - it is a lottery, I find that "most" of the time it gets the first published date correct or close enough for me not to notice, but yes there will be times where it is too recent. Now unless Kovid allows the metadata source like Goodreads to pass through its (sometimes more reliable) date there is nothing you can do but manually correct it. I can understand why he has done it the way he has, because the metadata source plugins that ship with calibre like Amazon don't actually have a useful publication date, so the Worldcat approach works better for them. Perhaps it choosing a min date of all sources (Worldcat + actual values returned from metadata sources) might work better. Maybe not...