Fault Sampling, Documentation and Trending

Start with the decision. Define whether sampling is intended to describe the defect, map its distribution, test a causal hypothesis, assess product disposition or verify an action. These purposes require different locations, numbers and methods. A dramatic example is useful for description but may be poor evidence of lot prevalence. Sampling plans should include comparison product and should be capable of detecting the spatial, temporal or batch pattern suspected.

Map variability deliberately

Select units across lots, product sizes, chamber or line positions, package lanes, time points and raw-material groups. Include affected and apparently unaffected units. Record why each position was chosen. Convenience sampling can miss gradients, intermittent failures and local equipment effects. Composite samples may be useful for some questions but can conceal an extreme unit; preserve individual identity when distribution matters.

Preserve identity and condition

Assign a sample identifier linked to product, lot, location, date, time and collector. Record photographs with scale, instrument and method, sample temperature, packaging, transport, storage and any destructive examination. Protect samples from further drying, warming, oxidation or contamination. A changed sample may no longer represent the condition at discovery. Retain chain-of-custody controls where regulatory, legal or external laboratory use may follow.

Document observations and measurements

Separate observation from interpretation. Record actual colour location, texture, odour, dimensions, mass, pH, temperature, water activity or package condition rather than only pass, fail or abnormal. Identify the method, instrument, units, resolution and any uncertainty that can affect the decision. Keep original data and annotate corrections. Link the sample to the relevant process record so the result can be interpreted against actual history.

Trend stable categories

Use controlled fault names, severity definitions and cause codes across batches. Trend by product, supplier, line, chamber, season, shift and confirmed mechanism. Counts should be interpreted against production volume and inspection intensity; more observations may reflect more production or better detection rather than deterioration. Preserve reopened and recurring events. Low-frequency failures often become visible only when records are combined over time.

Set action and review rules

Define what pattern triggers containment, investigation, escalation or reassessment. A statistical signal is not automatically a food-safety limit, and absence of a signal does not excuse a critical deviation. Review whether the sampling scheme could have detected the recurrence the effectiveness check was meant to find. Trend reports should lead to named decisions, not become passive archives.

Interpret results within their limits

State whether a result describes one unit, estimates a distribution, tests a hypothesis or supports a release rule. Avoid converting absence of detection into absence of hazard. Where results conflict, review sample identity, condition, method capability and process history before averaging or selecting the favourable result. Keep uncertainty visible in the decision record.