They are off at the lower end, in fact, they are off about -8dbm, if I remember correctly. Also, when I calibrated it with top of scale, my 500mhz marker, which is calibrated at -60dbm, would not agree with a 500mhz -60dbm signal injected at front of analyzer. Now, they are pretty close.
You have me thinking though, because the attenuator buttons do represent top of scale. I may try it again for top of scale and see which way gives me the least error overall. There are errors either way, I am just trying to keep them as low as possible over the entire scale.