There was a similar issue a few weeks ago re: greenandblue's odd tank. Someone there also brought up the point that over multiple tanks, any fluctuations in how full the tank is get worked out- the deficit from one tank is made up in the next. Unfortunatley that makes it hard to look at short-term subtlies.

IMO, I'd take the odometer mileage over GPS. Even if its not as accurate in absolute terms, its more repeatable in its error, and that can be reliably accounted for. GPS has many variables out of the operator's control- satellite lock, time between fixes, algorithm used to calculate distance between fixes, yadda yadda. The 'bread crums' mentioned before allude to the non-repeatable nature of GPS fix errors. Up to 10 meters (?) for EACH fix...compound that and BAM- you've got an issue (of course, the errors can be considered to be largely statistical in nature and can average to ~0).

That's why DOT built differential GPS for coastal mariners- to find the local instantaneous error and send out a signal that corrects for it.

When the Navy switched from 80s vintage radar-and-imagery guided Tomahawks to GPS guided, there was a few meters' worth of dropoff in precision...it hit the 1st or 3rd window from the left instead of always the 2nd. But the change was there.