Got it.  The difference is in the end dates 10/29/04-20,391 vs 10/28/04-20,849.  So the logical conclusion is that we made very little products on 10/29/04 or we made a hell of a lot on 10/1/04.

I agree, very weird, have to tried e-mailing them?

From their contact page:
Weekly Petroleum Status Report    Larry Alverson    202-586-9664    larry.alverson@eia.doe.gov

If you don't want to, I will, you've peaked my curiosity.

My guess, based on experience with other government programs, is that there are reporting dates on which data is due to the agency in question. The agency then doesn't mark the data as to the date it was actually produced but rather to the date it was reported. Thus, there may be a rush of information on or near a report date that looks like a sudden spike but is not really a spike.