Belt Scale Poor Repeatability

Yes.

6% variation is a rather large variance and unless the belt is grossly drifting it would be a factor but doubt if it would be the main cause. I would be investigating if the belt tension is uniform or if there is any build up of material in the load cell areas. How many TPH is the conveyor, how fast is the belt?

Both I and my colleagues used belt scales in the coal processing industry. They were "generally" a poor means to obtain weight measurements especially if using as a part of a weight proportioning system. Always had better luck with volumetric feeding.

If you only have two load cells, then there most likely is a component of the belt tension that appears as force on your load cells, compensated for at calibration time. So the belt shift, especially if it varies, is the likely source of the error. Belt tension variation across the conveyor (tight on one side loose on the other making it shift) could be interpreted as a higher load according to the vector sum of the convey direction belt tension applied to the roller at the belt angle from horizontal as it approaches the measuring rollers (even a very slight angle).

There are belt scales that have 4 load cells on the frame, they are likely less influenced by dynamic belt tension changes.

For instance, on a 3 legged stationary vessel, you can measure one of the legs, and assume the mass in the vessel is evenly distributed among all 3 legs, so multiply the signal from the one load cell x 3 for the total weight.

You may be able to calibrate your scale based on the dynamic conditions, but perhaps the least troublesome fix will be to repair your belt guide system, and/or troubleshoot the source of the instability. With a drifting belt, you also are placing unusual loads on your mechanical components, and not improving the service life of the equipment.