We prune the "clutter" and "negative values" that show up at the lowest levels via the "PEMDropThreshold" setting in AnalysisParameters. This is being done for the per-channel analysis, but not for the population statistics that are being reported out by batch processing. As a result, there is sometimes a notable mismatch between histogram and reported geometric mean.
This should be corrected by applying the drop threshold to all calculations, not just some.