systematic error. Do not rely an over-all average.
Good drilling recovery in waste rock may
disguise a problem. Look at drilling recovery by important ore zones (e.g. rock/alteration, structural domain, mineralization style) Recovery measurements are poor in broken rock. Monitor weights. There is no less expensive measurement a lab can make than taking the weight Within a given rock-type, weight/length provides a check on stated recovery. Weight/length may be more reliable than recovery estimated by length measurements in broken core. Weight/length is also a good cross-check on bulk density and geologic logging in some cases. Does drilling recovery correlate with grade? Find drill sample pairs that are adjacent (share a common boundary) where one pair member is mostly in a core run of high recovery, the other in low recovery, and otherwise are geologically similar. Calculate the average grade of the low- recovery vs. high-recovery. Use Sign Test for statistical significance. Sign Test
Treats pairs as coin flips. Equal values do
not count. Round values to be certain differences are meaningful. Binomial theorem provides estimate of likelihood of distribution of “heads” and “tails” (in this case, which member of the pairs is higher grade). Expect near 50-50 distribution. Formula for Sign Test at .05 significance =INT((n-1)/2-0.98*SQRT(n+1)) Where n is the number of pairs. Must have at least 10 pairs for .05 level of test; more is definitely better. Example: for 100 pairs, the distribution is significant at .05 if one group has higher grade than the other < 39 times. Statistically significant is not necessarily “significant” to the project
Differences between groups of a few
relative percent may have negligible impact on project economics. Focus on the critical ore types and groups Another Approach
Look for a general trend in grade with
drilling recovery Approach Estimate drilling recovery of each sample within some mineralized domain Group into bins by recovery Calculate mean and confidence interval of each group Plot on chart to show trend (if any). A correlation does not show causality, only cause for concern!