Most load cell accuracy is lost before the first measurement, at the mounting interface. A single point cell bolted to a surface that is convex by 0.1 mm over its footprint is pre-loaded unevenly, which shifts zero, reduces the usable overload margin, and adds hysteresis that no amount of calibration can remove.
The target
For a cell with a machined base, aim for flatness within 0.02 mm across the mounting face and surface roughness better than Ra 3.2. For larger canister and column cells, 0.05 mm across the base diameter is a practical target. The number matters less than the method: check with a straight edge and feeler gauge at four orientations, or with a dial indicator on a surface plate for critical installations.
What distortion does
A distorted base bends the element, which puts the strain gauges into a pre-strained state. The bridge still balances at zero because the installation is stable, but the effective stiffness changes and the calibration performed on a flat plate no longer applies. This is why a cell that calibrated perfectly on the bench reads 0.3% low once installed.
Practical fixes
Machine the mounting plate rather than shimming it. If a shim is unavoidable, use a single full-footprint shim of uniform thickness, never stacked washers. Torque bolts in a cross pattern in two stages, and re-check zero after 24 hours, because the joint settles.
Kevin Huang