For calibration, there are a couple of issues going on.
One, as mentioned, is linearity. The cell merely generates a voltage based on a galvanic reaction when exposed to O2. More O2 = more voltage. In a healthy cell, the response is mostly linear, but not perfectly. The analyzer is assuming linearity. So, even if you calibrate perfectly on air, you may be off a little at 100% O2 or vice versa. Frankly, with good cells, this has not been much of an issue for me. When I calibrate on 100% O2, I'm very, close when I check against a known 21% gas.
The other issue is just to understand the nature of what the analyzer is doing. The analyzer is just doing math. When you calibrate, the sensor locks in the relationship between the mV it is reading from the cell and whatever O2 percentage you've told it you are using for calibration: e.g., the analyzer might calibrate by "locking in" the relationship that "12.5 mV = 21% O2". After that, the analyzer just does multiplication based on that ratio to convert mV to an O2% when you test gases: if 100% O2 generates 60 mV on calibration, then when you test a gas and the cell puts out 30 mV, the analyzer will report that as EAN50.
Thus the desirability of calibrating on 100% O2 when you have that ability. Any error in an air calibration -- failure to adjust for humidity, altitude, gas flow, etc. or just the fact that cells have some inherent error range -- is going to be multiplied by almost 5 when you analyze 100% O2. If you're off by 1% on air, you'll be off by about 5% on 100% O2.
Conversely, if you calibrate on 100% O2, any errors will be smaller when you analyzing gases with less O2 (about 0.2% on air, leaving aside the inherent small +/- error in cells). For recreational stuff, it's not that significant, but this is why for high O2 mixes, you are better off calibrating off 100% O2. It doesn't solve for linearity, but it is much better than calibrating on air and then trying to measure really rich mixes with accuracy.
Better analyzers allow two or three point calibrations and then the analyzer can correct for cell linearity, which is obviously even better.
87% is grossly off. Either the cell in the analyzer is current limited, the calibration was horrible, or you've got something other than 100% O2. No way I'd dive that cylinder until it is resolved.