Datasheets list nonlinearity, hysteresis and repeatability as separate percentages of rated output, and buyers often add them up into a single accuracy figure. That is not how the standard treats them, and it is not how errors combine in practice.
Three different curves
Nonlinearity is the deviation between the actual output curve and the best straight line through the calibration points. Hysteresis is the gap between the increasing-load and decreasing-load curves at the same load. Repeatability is the spread between successive runs at identical load. A cell can have excellent repeatability and poor hysteresis, which usually means mechanical friction or a damaged element rather than an electrical fault.
How they add up
For a type approval the individual errors are checked against the maximum permissible error of the class at each load step, not summed. In a real installation the total uncertainty combines them with the indicator resolution, the installation mechanics and the temperature effect, and in most systems the mechanics dominate. A cell with 0.02% hysteresis bolted to a frame that deflects 0.5 mm under load will show far more error than the datasheet suggests.
What to check first
If hysteresis exceeds the datasheet value by a factor of two or more, look for a mounting surface that is not flat, a bolt torque that is uneven, or a load introduction point that applies a side load. Those produce the same symptoms as a failed cell and cost nothing to fix.
Kevin Huang