This is exactly what would be expected when samples that are actually too small are tested; generally low readings and higher ones on retests most of the time. I assume also that the full two-probe sample is not ground before removing the analysis portion, This magnifies the low underestimation issue. Charles Hurburgh