QA/QC Data Review

Quality assurance/quality control (QA/QC) data review evaluates whether sampling and analytical evidence is suitable for its intended use. It connects quality-control results to the material, procedures and decisions they actually cover. It is not merely counting failed checks or accepting a laboratory certificate without examining its qualifications.^[1]

An effective review distinguishes three questions: whether records are complete and correctly linked, whether measurement performance meets defined requirements, and whether the resulting data adequately represent the geological or operational target. A satisfactory answer to one does not automatically answer the others.

Establish the Review Basis

Criteria should be established for the relevant analytes, matrices, concentration ranges and sample types before results are accepted. The review needs to know what each control repeats, where it enters the process and which routine samples it can reasonably inform. CIM's guidance recommends integrating controls into normal work, evaluating results promptly and documenting criteria and actions.^[1]

Begin with the records: submission and certificate identifiers, sample counts, interval or lot references, CRM identities and lots, duplicate pairings, units, method codes and reporting limits. A plausible-looking chart can still be wrong if a reference material was assigned another material's certified value or if results in different units were combined.

Preserve original certificates and subsequent revisions. “Not analysed”, a qualified estimate and a result below a reporting limit should remain distinguishable. When multiple determinations exist, the database needs an explicit selected-result rule rather than silent replacement or selection of the preferred grade.^[1]

Reviewing Reference Materials Over Time

Assess each CRM against a property value appropriate to the method and reporting basis. Differences between measured and certified values need interpretation in relation to their uncertainties. A certificate's uncertainty interval should not be mistaken for the expected distribution of individual laboratory results.^[2]

Time order matters. A stable average can hide a drift, step change or two periods with offsets in opposite directions. NIST's control-chart guidance notes that non-random patterns can indicate an unstable process even when individual points remain inside the plotted limits. A chart should use variability appropriate to the measurements being charted, rather than short-run repeatability when routine conditions vary over time.^[3][4]

Control limits, warning rules and project acceptance criteria have different functions. A statistical signal calls for investigation, not automatic rejection; chance signals remain possible. Apparent statistical stability does not establish agreement with a correct reference value or fitness for a demanding decision.^[3] Recentring a chart on a biased laboratory mean cannot resolve that bias. Method changes and replacement CRM lots also require explicit handling, not an uninterrupted chart with undocumented changes.

Investigating Blanks and Carry-Over

Blank review begins with the blank's characterized background and the stages it experienced. A powder blank inserted after mechanical preparation cannot rule out crusher contamination. An analytical procedure blank tests a different boundary from a field blank. Eurachem's classification makes these distinctions explicit.^[5]

Compare blank results with the actual preparation and analytical sequence. The field submission order may not be the order used at every laboratory stage. Associations with preceding high-grade samples can guide investigation, but a raised blank alone does not uniquely identify carry-over as the cause.

The affected sample set should follow the evidence about the failure mechanism. It may be narrower or wider than a nominal certificate batch. Subtracting the blank result from every routine assay is generally not a demonstrated correction for sample-specific contamination. Any validated background correction within an analytical method is a separate, documented procedure.

Assessing Duplicates Without Losing Their Meaning

Separate field, coarse-preparation and pulp duplicates before summarizing performance. Here, pulp refers to pulverised laboratory powder, not a processing slurry. Their differences contain different combinations of material variability and process variation. Comparing results across grade and material groups can reveal problems concealed in an overall average. Shared errors remain shared: matching duplicates do not independently establish unbiased collection.^[6]

For two comparable positive concentration results, one common descriptive measure is relative percent difference:

RPD (%) = 100 × abs(a - b) / ((a + b) / 2)

Here, a and b are the paired concentrations expressed in the same units and on the same reporting basis; abs denotes the absolute value. The formulation is documented in the US Intergovernmental Data Quality Task Force's 2005 manual. Its environmental-programme acceptance examples are not universal mineral-sampling limits.^[7]

In a hypothetical pair of 1.00 and 1.20 g/t, RPD is approximately 18.18%. This calculation describes disagreement, not acceptability. Because the absolute value removes direction, retain the paired results and examine signed differences for systematic offsets. Interpretation still needs the duplicate stage, concentration range, material behaviour and intended use.

Near zero, a small absolute difference can create a large percentage. Two zeros make the denominator zero, and substituted values for non-detects can produce artificial agreement or disagreement. Censored results therefore need separate treatment. A single RPD is not a complete estimate of sampling uncertainty, and mean RPDs cannot be subtracted to isolate a variance component. Separating contributions requires an appropriate replicated design, such as nested sampling, preparation and analysis, with compatible populations and variance assumptions.^[6]

Using Checks to Diagnose, Not Merely Flag

The combined pattern is more informative than an isolated failure. Acceptable pulp repeats with poor coarse duplicates suggest that the investigation should extend earlier than the final analytical measurement. That pattern is not proof of one specific preparation fault, because material heterogeneity and the duplicate design also matter. ALS's preparation note identifies the splitting and pulverising stages included in its coarse duplicates.^[8]

Independent laboratory checks require compatible methods and material fractions. Different decompositions may measure different quantities, so disagreement is not automatically evidence that one laboratory is inaccurate. Eurachem's validation framework evaluates performance for an intended method and matrix, not analytical capability in the abstract.^[9]

A technically meaningful investigation follows the earliest plausible error. Reanalysis of an existing solution addresses a different question from a new digestion, repreparation of coarse material or repeat collection. The available retained material may limit the investigation; that limitation should be stated rather than hidden behind additional repeat assays.

Release, Correction and Feedback

Review decisions should identify the samples affected, evidence considered, action taken and remaining restrictions. Possible outcomes include acceptance for the stated use, qualification, further investigation or rejection. A control failure does not automatically invalidate every sample in a campaign, but an unresolved systematic issue cannot be dismissed because most individual controls passed.^[1]

Retests should follow an agreed investigation, not continue until a favourable result appears. Keep the original and repeat values, their methods and the reason for any replacement. If a preparation problem is corrected, evidence of improved performance should come from the revised process, not merely from deleting the earlier failures.

Finally, communicate implications to users of geological models, grade-control decisions and production records. A laboratory correction may require updating downstream datasets; a spatial sampling gap may instead require additional field evidence. The review is complete when the decision and its scope are documented and traceable, not when all dashboard symbols have turned green.