Setting acceptance criteria for a mapping study

Criteria should be numerical and unambiguous: every sensor within 2.0 °C to 8.0 °C for the whole study, for example, with any reading outside that range counted as an excursion regardless of duration — or with a stated permitted duration if the product data support it.

Decide these explicitly

• Limits, including whether they are inclusive.
• Whether a single reading, or a period, constitutes an excursion.
• Which sensors are 'in scope' for acceptance (an ambient reference sensor is not).
• How measurement uncertainty is treated at the limit: a reading of 8.05 °C with an uncertainty of ±0.3 °C is not clearly outside.
• What happens on failure: exclusion of positions, remediation and re-mapping, or product-specific assessment.

Guidance on decision rules for conformity statements applies here as much as it does to calibration certificates: say how uncertainty was accounted for.

Six decisions, made before the loggers go in

Criteria fail by being incomplete more often than by being wrong. These are the questions that will otherwise be answered after the data are in, by whoever is writing the report, under pressure to make the study pass.

Criteria come from the product, not from the building

The limits are derived from the label conditions of the products stored in that space, which are derived from stability data. That direction is one-way. A warehouse that runs to 26 °C in August has not earned a 26 °C criterion; it has a finding. The only route to a wider limit is stability data supporting it, and that is a decision for the marketing authorisation holder, not for the mapping protocol.

Where a space holds several products, the tightest label among them sets the criteria. A site default applied across all storage areas is one of the most common and most serious findings, because it is invisible until somebody compares the SOP with what is actually on the shelves.

Uncertainty at the limit

A mapping study makes a conformity statement, so the same logic that governs a calibration certificate applies: say how uncertainty was accounted for. If the loggers carry U = 0.3 °C and the limit is 8.0 °C, simple acceptance passes a reading of 7.9 °C with a real chance the true value was above the limit. A guard band pulls the acceptance limit in to 7.7 °C and makes the statement mean something.

• State the loggers' expanded uncertainty and coverage factor in the protocol, from their certificates rather than their datasheets.
• State the rule: simple acceptance, guard band of w = U, or a stated fraction — and apply it to every sensor the same way.
• Set the eventual monitoring alarm limits inside the mapping limits by at least the monitoring sensor's uncertainty, for the same reason.
• Write the rule into the protocol, not the report. Choosing it after seeing a borderline result converts a risk policy into a way of getting the answer you wanted.