You are totally right. Many available load cells specify a "combined" error of about 0.01...0.03%. But this uses the mathematical trick of geometrical summation to give smarter numbers in the catalogue.

In fact, many real load cells show a real error of up to 0.1% and above! Especially the non trimmable hysteresis or so called "zero return error" can heavily increase when the load cell is aging over the years and can be in the range of 0.1% and well above allone!! 0.1% corresponds to 10bit precision, by the way...

The key issue in industrial weighing is reliability. Only perdiodically performed plausibility checking, in the way Michael has thankfully discussed in detail, can detect load cells leaving the allowed error band and alarm the repair people.

Digitally summing can also provide a linearization of individual load cells by using individual calibration curves, even if only a piece wise interpolation between a few calibration points is used. Sometimes a temperature measurement near the load cells can be helpful to perform kind of simple temperature compensation.

Kai Klaas