Reports of stocks of crude oil in the U.S.

What I noticed in the past year in reports on the economic news was the following. In my experience updates of inventories of crude oil in the United States had been the one time stronger than expected and a decline in oil production the next time, I noticed for the faster than expected increases in oil stocks.
Over time this experience repeated itself and in my curiosity - or was it suspicion - decided to straighten things out(without pretending to be exhaustive):


Date Announcement Declined Increased Terms used in coverage
06-10-10 3.100.000 last week has increased faster than expected
29-09-10 -500.000 declined more than expected
10-09-10 -2.500.000 decreased as expected
2-08-10 -1.800.000 crude oil stocks fell suddenly
28-07-10 7.300.000 crude oil stocks rose
22-05-10 2.500.000 last week has increased faster than expected
5-05-10 2.800.000 last week rose more than expected
4-03-10 -758.000 unexpectedly fell last week


The numbers are barrels of crude oil


What was it that triggered me so?
1. On the one hand there are significant differences between -1.800 and 7.300.000 barrels but it's not like stocks decline and increase every week and another week;
2. I think in particular the terms "harder than expected", "suddenly fell" and "unexpected" made me more alert.
Nice to look back to see if your experience matches reality.

The xls file with more information, visit:
https://spreadsheets.google.com/ccc?key=0Ar2CjvVL2hq_dG5lb3pDNWlGeUhXdmlMMUhSakpwR3c&hl=en

Ofcourse this raises a few questions:
1 How does one produce these estimates and is the margin of error that is accepted?
2 What are the implications of these messages for different idices?

Feel free to make a contribution in solving this mystery to me!