C3 · Publication Volume 13

QA/QC Samples

standards, blanks, field duplicates, laboratory duplicates and umpire assays

Learning goals

The learner should be able to design quality-assurance activities and quality-control samples around explicit failure modes; distinguish blanks, reference materials, field duplicates, preparation duplicates, analytical repeats and independent check analyses; calculate recovery and duplicate difference without applying a single rule across all concentrations; diagnose batch, sequence and stage; and record validation decisions without deleting failed evidence.

Quality assurance is the planned system of responsibilities, procedures, training, traceability and review. Quality control is the set of observations used to test whether the system performed as required. A fixed percentage of control samples is not a complete design. Each control must have a diagnostic purpose, placement logic, acceptance rule and response.

Blanks and contamination controls

Different blanks enter at different stages. A field blank can test containers, reagents or handling under field conditions; an equipment blank can test cleaning and carry-over; a preparation blank can test crushing and milling; a reagent or method blank can test laboratory chemistry. A blank introduced late cannot diagnose earlier contamination.

Select blank material with concentrations low enough to reveal meaningful contamination and with a matrix compatible with the process being tested. “Below reporting limit” is not automatically a pass if the limit is above the contamination level that would change the decision. Conversely, a tiny detected mass may be immaterial at high routine concentrations. Define acceptance relative to both method capability and intended use.

Sequence matters. Place blanks after likely high-concentration samples to test carry-over, across preparation equipment changes, and throughout batches. If control identities must be masked from routine handling, retain secure role mapping. Never relabel a contaminated blank as a routine sample or silently exclude it.

Reference materials and bias

A reference material tests closeness to an accepted value for specified analytes and methods. Matrix, mineralogy, concentration and homogeneity determine relevance. Use more than one concentration or material when decision-critical ranges and carriers differ. A material not certified for an analyte can still monitor consistency if characterised, but it cannot support the same accuracy claim.

Quality controls are placed at field, preparation and analytical stages so each failure can be located
Quality controls are placed at field, preparation and analytical stages so each failure can be located

Percent recovery is

$R=100\frac{x}{x_{\mathrm{ref}}},$

where x is the measured result and x_{\mathrm{ref}} the accepted value for the stated method basis. Evaluate uncertainty in both. A result slightly outside a fixed percentage band may not be significant when reference uncertainty is large, while a persistent small bias across batches can matter near a decision threshold.

Plot recovery by batch and sequence, not only as an overall histogram. Look for step changes, drift, element groups and concentration effects. A batch-wide correction is a new model requiring validation; it should never overwrite reported results.

Duplicates and precision components

A field duplicate includes local heterogeneity, collection, preparation and analysis. A preparation duplicate begins from a retained coarse or crushed split. A pulp duplicate begins after fine preparation. An analytical repeat may use the same solution or a new aliquot. Their difference in variance helps locate the dominant stage, but only when duplicate protocols and identities are clear.

For positive paired values x_1 and x_2, relative percent difference is

$\mathrm{RPD}=100\frac{|x_1-x_2|}{(x_1+x_2)/2}.$

It becomes unstable near zero and is unsuitable when either value is censored. The absolute log difference |\ln(x_1/x_2)| is symmetric for positive values and often more stable across orders of magnitude. Use concentration-dependent limits and graphical review rather than one universal cutoff.

Duplicates must be distributed across geological domains, concentration ranges, operators and time. A duplicate pair created by splitting one pulp does not test sampling representativity. Adjacent field samples are not duplicates unless the protocol states how the second support relates to the first.

Batch controls, drift and independent checks

Randomise or balance routine samples so geology is not confounded with analytical order. Insert controls at the start, throughout and end of batches. Control charts can reveal level shifts, trends and isolated excursions, but limits must come from stable performance appropriate to the material, not from a failed batch that inflates its own tolerance.

An independent check uses a retained split and a sufficiently independent preparation or measurement path to test a consequential concern. Independence is relative: sending the same pulp to another instrument does not test field collection or initial preparation. The check method may have a different measurand, so paired differences require method comparability.

Predefine responses: hold the batch, inspect records, reprepare, reanalyse, use an alternate method, qualify affected elements, or reject fitness for a purpose. One passing reanalysis does not erase the failure. Document the cause assessment, affected scope and released version.

Worked synthetic example

A synthetic reference material has accepted X concentration 100\pm3 mg/kg. Four batch results are 96, 98, 101 and 89 mg/kg. Recoveries are 96%, 98%, 101% and 89%. The last result is not judged only by an arbitrary 90% boundary; it is a clear shift relative to the first three and should trigger sequence and batch review.

Three positive duplicate pairs are (12, 15), (105, 110) and (980, 1,020) mg/kg. Their RPD values are 22.2%, 4.65% and 4.00%. Absolute differences increase with concentration while relative differences decrease. The low pair may reflect greater proportional heterogeneity or method uncertainty near the lower range. A single absolute-difference limit would fail high concentrations; a single relative limit may over-flag low ones.

A field-duplicate log-variance is 0.16 and a pulp-duplicate log-variance is 0.04. Under a simplified additive model, field plus early-preparation contribution is approximately 0.16-0.04=0.12, or 75% of the observed duplicate variance. Repeating the instrument alone cannot address the dominant component. The estimate is synthetic and assumes comparable pairs and independence; its purpose is stage diagnosis.

QA/QC audit workflow

  1. List decision-critical failure modes by field, preparation and analytical stage.
  2. Assign a control type that can actually encounter each failure.
  3. Select relevant blank and reference materials across concentration and matrix.
  4. Define duplicate protocols and variance components before collection.
  5. Place controls through space, time, batch and concentration sequence.
  6. Predefine concentration-aware warning, failure and response rules.
  7. Review values, qualifiers, charts, batch maps and paired plots together.
  8. Determine affected samples, elements, methods and time intervals.
  9. Preserve failures, investigations, reanalysis and validation status in lineage.
  10. Release fitness decisions per use, with unresolved limitations visible.

Practice and review

  1. Match five suspected contamination pathways to the earliest blank that can detect each one.
  2. Calculate recovery and RPD for a reference result of 47 mg/kg against 50 mg/kg and a duplicate pair of 47 and 52 mg/kg.
  3. Explain why a reference material can pass while routine samples suffer coarse-particle heterogeneity.
  4. Design control placement for a batch in which expected high samples could contaminate the following positions.
  5. Write a release note for a failed element in one batch while other elements remain fit for screening.

Review questions: Which failure can this control see? Is its matrix and range relevant? At what stage did variance enter? Which samples are affected? Was the failure preserved in the released audit trail?

Sources and further reading