My day job is as a metrologist, so I have calibrated many precision levels, like the Starrett #199. When I calibrate these levels at work we need to make sure that the temperature hasn't changed more than 2 degrees in the past 8 hours & that the level has been in the same environment for that 8 hours. We try for 68F (20C). The first check is for flatness. This is the hardest part of the calibration & requires some skill & practice. But for you guys at home long as it doesn't rock move when you push on any one corner, & then you are probably flat enough. Yes the levels are made slightly concave & you shouldn't use them on anything that is shorter than the level.
Clean the level & the surface plate, checking the level for any burrs or other damage. Level the surface plate (using the level you are calibrating) & let the level sit for a few minutes, re-check level & readjust as needed. Then rotate the level end for end putting back up against your reference point. I use two 2x6 granite parallel bars with 1 perpendicular to the other, & held in place to the surface plate with modeling clay, to maintain a reference point. A few minutes later after the level settles it should indicate the same level reading. If not double check your first reading, rechecking to ensure that everything is clean before adjusting the level. Screws & jam nuts type adjustments are very touchy, & it may take some time to get it spot on, the only to find that when you tighten down the jam nuts you throw the level out of whack again. A light touch is needed & always allow time to let it settle. Sometime I'm not even be sure I moved it at all when adjusting it, that how light of a touch you need, unless it's way out already, probably from moving the adjustment to much in the first place.
At work we also check the level sensitivity to ensure that each division on the level equals a certain amount of angle. With the Starrett 199 one division equals 1/2 thousandth (0.0005) of an inch per foot. When I check the 199 it's typically about 8 arc seconds per division. So figure 4 arc seconds if you are only off half a division when you use it. To calibrate the sensitivity we use a Brunson 470 small angle generator. It's basically a granite sine plate with a big micrometer head to adjust the angle down to 0.1 arc seconds per division. I move the bubble exactly 1 division then record the reading, move it another record, etc... then at the end average the readings. The factory usually specs levels as a certain amount per distance, such as .0005" per 12".
It usually comes down to the person eyeballing the bubble to determine the sensitivity. The biggest error we encounter is the person looking at the level isn't looking at the bubble dead on the same way, using the same eye, each time they make a measurement.
Hope this was helpful.