The presence of systematic lab bias leads to a continuous offset between observed analytical results and actual grades; consequently, there is significant risk introduced into the geological resource models and economic assessment of such deposits. Analytical bias creates an erroneous bias of mineral resource assessments, distorts grade-tonnage relations, and ultimately results in expensive mistakes in financial evaluations of projects. The QA process entails proactive measures that are taken to avoid mistakes; meanwhile, the QC process involves methods used for detection of analytical mistakes. An acceptable approach to assay verification includes the use of matrix-matched Certified Reference Materials for accuracy measurements and blank samples for contamination detection.
The choice of Certified Reference Material (CRM) requires strict correspondence of the matrix of the reference standard and natural sample flow. Reference standards should correspond to the state of matter, mineralogical composition, concentration range, and measurand of the host rock to provide identical chemical reaction during digestion and further analysis. Utilization of the pure liquid calibration solution for the analysis of solid minerals does not consider the interference effect of digestion and the mineral matrix.
Blind-certified standards are inserted into the sample flow randomly with the base level of 5% (one standard per 20 samples). Blind sample numbering hides these standards from analysts. Blind standard insertion guarantees the testing of control samples together with regular samples without any special treatment.
Statistical bias can be tested with the help of the comparison of the measured value with the certified one according to the budget of expanded uncertainty. As it follows from ISO Guide 33 and NIST criteria, the statistical level of bias is achieved when the absolute difference between the found value (xfound) and the certified value (xCRM) is greater than the sum of uncertainty multiplied by the coverage factor k=2:
| x_found – x_CRM | ≤ k × √(u_found² + u_CRM² )
In case where the difference falls within the uncertainty interval, there is no indication of statistical significance of the bias. The presence of systematic bias is established when the intervals do not overlap or if the means differ significantly from their expected values in more than one batch.
Blank samples are used to detect physical and chemical contamination, avoiding any analytical positive bias that might cause elevated values for poor grade samples. Cross contamination during sample preparation is the major source of analytical positive bias, especially with the crushing or grinding of barren rock right after high-grade ore samples.
The use of blanks involves different types of samples placed appropriately in the sampling process chain. Coarse inert blanks or prep blanks after high-grade mineralized zones indicate carry-over during jaw crushing and ring mill grinding. Field blanks with natural samples through transportation and handling determine environmental contamination in the laboratory. Method and reagent blanks determine the instrument noise and reagent purity in the laboratory analysis.
Realistic Method Detection Limits (MDLs) are determined using practical data from blank samples and not optimistic Instrument Detection Limits (IDLs). Manufacturers provide IDLs based on their optimal signal-to-noise ratio. The realistic MDL is three times the standard deviation (3SD) of replicates of either field blanks or total blanks analysed under the entire physical and chemical process procedure:
| Blank Type | Primary Purpose | Action Limit Threshold |
| Coarse / Prep Blank | Detect carry-over contamination during jaw crushing & pulverization | 3 × Detection Limit (3 × DL) |
| Field / Trip Blank | Identify environmental contamination during sampling and transport | 3 × Detection Limit (3 × DL) |
| Method / Reagent Blank | Establish baseline instrument noise and reagent cleanliness | 3 × Standard Deviation of Blanks |
Limits of Detection (MDLs) based on the variation seen in blanks provide 99% assurance that analyte levels are truly indicative of mineral content and not just background impurities or measurement noise. The indication of control failure is when blank results are greater than three times the detection limit.
Shewhart control charts are used to track analytical performance to enable continuous surveillance of quality. Standard statistical process control charts are used to compare CRM and blanks to preset mean goals and standard deviation boundaries.
The statistical boundaries include warning limits which consist of ± 2 SD (accounting for about 95% of expected random variations) and action limits which include ± 3 SD (accounting for about 99.7% of expected variations). An individual CRM value exceeding the ± 3 SD action limit indicates a high possibility of failure in the analytical process and therefore is considered an invalid batch.
The statistical approach will help detect non-random drift and trends well before the limits are exceeded. The trend rules for detecting out-of-control situations state that there is out-of-control condition if six to seven consecutive CRM values show increasing or decreasing drift, or if consecutive points are all found on one side of the mean. This shows possible decay in instrument performance, calibration and even reagent instability.
Violations of the action limits require isolation of batches to carry out further analysis. The quality managers isolate the analytically non-compliant batches to ensure that their information does not get entered the master resource database. Lab technicians investigate calibration, digestion temperature and reagent quality before assaying the samples again. Pulp duplicates (about 5%) are sent for testing in an independent, accredited secondary umpire lab.
A multi-level QA/QC system ensures that geological resource models are protected from any systematic bias caused by the assays of laboratory facilities. Matrix-matched Certified Reference Materials are used for validating accuracy of analyses, while blank samples are introduced to prevent any false positives caused by any contaminants, and Shewhart control charts are used for evaluating performance of the process. Batches of analyses should be quarantined until verified to avoid any biases of an unverified dataset. These statistical tools assure that datasets are sound, scientific, and economic.
