While I no nothing about the accuracy of this unit, and they may be overstating the true accuracy, I'm not sure how one can conclude it isn't measuring correctly just because there are small changes in the reported tank alkalinity, unless you somehow otherwise know what the alkalinity is.
Presumably these tanks get some sort of dosing at some time (maybe not in the times shown, but that isn't clear). Further, pH will impact the answer, and even a tiny but of calcium carbonate solid (perhaps from a fish stirring up some detritus, etc.) getting into the test will impact the results, as I show below:
Assume you are using 50 mL sample size (that's what I was told, but I don't independently know that).
A 1 mm diameter grain of calcium carbonate sand has a volume of 0.0005 mL and a weight of perhaps 0.0015 grams (1.5 mg). That means it contains 1.5 mg/100 mg/mmole = 0.015 mmole of CaCO3, which has twice that in alkalinity, or 0.03 meq of alk. Put that in 50 mL and the effective concentration is 0.6 meq/L (1.7 dKH).
Consequently, a much smaller grain of sand (or many really tiny bits of calcium carbonate) can readily impact the alkalinity that is measured if you are worrying about fluctuations on the order of 0.2 dKH.
Thanks for the analysis, It is being dosed with a calcium reactor with a steady rate of effluent which typically is usually consist. The unit has check valves with filters so Id imagine that would filter out derbies. I agree the error is overstated but I guess the way is bounces kind of defeats the purpose of the automation and certainly would make me question if I want to let it control dosing with that large of an error which was an objective to counter my .5+ dkh drop with such a large stocked system.