DIY Oxygen Analyser Accuracy

I've tabulated the important bits from the different manufacture's datasheets for comparison.
Because I've accumulated more than one datasheet for some sensors I've found some slight discrepancies between them, although nothing serious.

Manufacturer:	Maxtec	BPR	IT	IT	Teledyne
Model:	CAG-250E	ST-11	M-25	M-03	R17
Nominal output in air:	8-13mV	8-10mV	8-14mV	13-17mV	7-13mV
Zero offset:	<500uV	<200uV	<200uV	<200uV	<50uV
Response time (90s):	<15 sec	6 sec	<3 sec	<12 sec	<6 sec
Expected life (in air):	>900,000%O2 hrs	1,500,000%O2 hrs	>750,000%O2 hrs	>1,000,000%O2 hrs	36 mo
Recommended Load:	1M	2K	100K		10K
Accuracy (const temp):	+/-1% of FS			+/-1% of FS
Linearity (const temp):	+/-2% of FS	+/-1% of FS	<2%@100% O2	<2%@100% O2
Repeatability (const temp):			+/-1%@100% O2	+/-1%@100% O2
Warranty period:	24 mo	12 mo	18 mo		12 mo

The importance of nulling
The datasheets for Teledyne sensors quote a maximum output offset voltage (in 100% nitrogen) of 50uV whilst IT and Siemens quote 200uV and Maxtec 500uV.
If we assume a nominal sensor output of 500uV/% these correspond to errors of about 0.1%, 0.4% and 1% PPO2 respectively.
The implication from this is that after calibrating in air only, if a 42% mix were measured, the reading could be incorrect by about as much as the offset error plus any voltmeter errors.
Graph 1 below demonstrate the "worst case" effect of the errors when measuring a typical 36% mix with a Maxtec sensor.

This effect can be ignored for Teledyne's which has an offset 1/10th of Maxtec's but it certainly wouldn't be advisable to ignore for Maxtec's.
For those who want improved accuracy, the amplifier's offset adjustment can be used to compensate for this during commissioning, unfortunately to do this requires a source of 100% nitrogen.
Graph 2 shows the improvement gained by nulling the offset in nitrogen.

This is a considerable improvement, however, the best accuracy over the range we require is gained by calibrating in 100% oxygen then using pure nitrogen to finally "tweak" the offset adjustment.
Graph 3 shows the theoretical limits of the errors from this technique.

Of course the above figures look great, but you're not going to achieve this in real life .... all the sensor manufacturers quote basic accuracy and/or linearity figures. For Teledyne it's +/-1% of full scale, whilst Maxtec, IT and Siemens quote linearity within 2% - which correlates pretty well with +/-1%. So this must be taken into consideration.
For comparison, the VN202 commercial analyser supplied by Vandagraph quotes an accuracy of 1% of full scale, but since this is also the accuracy of the Teledyne sensor they use, I suspect they're assuming their instrument is perfect and introduces no errors - which is extremely optimistic.

In addition, the environmental conditions (temperature/humidity) can have a small but noticeable effect on the ideal of 20.9% when perfoming an on-site calibration with atmospheric air.