Facilities
Interpreting Refrigeration Excursions in Retail Records
A source-led distinction between sensor alerts, confirmed excursions, and product disposition.
Key takeaways
- An alert is not automatically a product excursion
- Calibration and duration matter
- Disposition needs its own evidence
Research question and unit
What can a refrigeration alert establish about a cannabis product? The unit is one sensor interval linked to one storage zone and, where possible, one affected lot. The population is all monitored intervals and alerts in the stated period. An alert, a confirmed excursion, and a product disposition are separate events.
FDA storage guidance, NIST measurement practice, Virginia cannabis oversight, and CDC cold-chain material support calibration, traceability, and documented response. They do not create one universal cannabis temperature-loss rate. A sensor threshold is evidence for review, not a conclusion about product quality without duration, location, calibration, and applicable product requirements.
Evidence model
Retain sensor identifier, zone, timestamp, reading, threshold, calibration status, duration, corrective action, reviewer, and linked lot decision. Preserve raw readings and any smoothed or aggregated series. A missing reading is not a normal reading. A brief excursion during a known door opening may have a different interpretation from a sustained failure, but the distinction needs evidence.
Report alerts per monitored day, confirmed excursions per zone-day, and unresolved product decisions separately. A sensor count is not a product count. If one sensor covers many lots, do not multiply the alert into a lot-level failure without a documented linkage.
| State | Evidence | Next question |
|---|---|---|
| Alert | Reading crossed configured threshold | Was it valid? |
| Confirmed excursion | Valid reading and duration established | Which lots were exposed? |
| Disposition | Responsible decision documented | Why this action? |
Interpretation and limitations
Repeated alerts may indicate door practice, sensor placement, equipment performance, or configuration. They do not prove one cause. Compare like zones and periods, and record maintenance and calibration changes. A lower alert count after a threshold change may be a measurement change rather than a physical improvement.
Transfer is bounded by product requirements, sensor quality, storage design, and jurisdiction. Public food-safety guidance may inform measurement but cannot determine a cannabis product’s disposition. Keep the evidence review separate from a legal or scientific release decision.
From sensor signal to product decision
The first decision is whether the sensor record is valid. Check calibration status, battery or power state, placement, sampling interval, communication gaps, and the configured threshold. A threshold crossing may be caused by a door opening, probe displacement, or an actual equipment failure. Preserve the raw series around the event rather than reviewing only the alert screenshot. If the system interpolates missing values, retain the interpolation flag and do not present it as a measured reading.
Duration and recovery shape interpretation. A single reading just beyond a threshold does not answer the same question as a sustained excursion. Record start, peak, recovery, and confirmation times, plus the source of each time. If an operator silences an alert before the temperature returns to range, distinguish acknowledgment from recovery. Repeated short events may indicate a different facility issue than one long event even when the total minutes are equal.
Linkage to product is often the weak point. A sensor zone may contain many lots, while a door alarm may cover a larger area than the refrigerator. Record which lots were present, when they entered or left, and whether the exposure interval overlaps the valid sensor interval. Do not assign every lot in a building to one alert without a documented zone relationship. If linkage is uncertain, place the lot in an exposure-review category rather than declaring loss or safety.
Metrics should keep signal, exposure, and disposition separate. Alerts per monitored day measure detection workload. Valid excursions per zone-day measure confirmed environmental events. Lots reviewed and lots with a documented disposition measure product-control work. These denominators cannot be swapped. Report missing sensor intervals and unmonitored periods because a quiet period with no data is not evidence of stable conditions.
Trend changes require the maintenance record. Sensor replacement, threshold changes, firmware updates, seasonal weather, door practice, stock density, and equipment service can all shift the alert series. A fall in alerts after widening the threshold is a measurement change until independently verified. Compare like zones and state whether the data source and configuration remained stable. A simple before-and-after chart without those facts overstates the evidence.
Product decisions belong to the authorized quality or regulatory process. FDA food-safety material and NIST measurement guidance can inform evidence handling, but neither determines the disposition of a cannabis product. Retain the decision maker, rule or specification consulted, affected lot, action, and rationale. The bounded conclusion is that valid readings and a traceable exposure link support proportionate review; an alert alone does not establish product harm, loss, or release.
A practical comparison boundary
Freeze the sensor export and configuration at the review cutoff. Later backfilled readings, calibration corrections, or threshold changes should be marked as revisions rather than merged into the original alert count. Report the system state used for the decision, because a retrospective chart can look more complete than the evidence available during the event.
What a reviewer should request
Request the raw sensor series, calibration and configuration state, zone map, lot exposure interval, maintenance record, and authorized disposition. A list of alerts cannot establish duration or affected product. Keep invalid, missing, confirmed, and dispositioned events distinct. That separation lets an operator improve sensor coverage without turning every data-quality problem into a product conclusion.
Why missingness changes the claim
The absence of an alert is meaningful only when the sensor was operating, the zone was covered, the threshold was active, and the data was retained. Report monitored time, missing intervals, invalid readings, and unmonitored storage separately. A low alert rate during a communication outage is a data-availability finding, not evidence of stable refrigeration. This distinction should appear beside every trend.
Interpreting a change over time
Compare alert trends only after checking sensor replacement, placement, threshold, sampling interval, and storage mix. A lower count after a configuration change is not evidence of a safer product environment until the measurement system is stable. Preserve the configuration history beside the graph and identify any interval that cannot support comparison.
Reporting the unresolved tail
The final table should show monitored intervals, missing intervals, invalid alerts, confirmed excursions, lots with uncertain linkage, and lots with an authorized disposition. A single alert count hides whether the unresolved work is a sensor problem or a product-review problem. Preserve the reason for each open state and the next authorized action so the trend remains honest.
A final evidence check
Before reporting an excursion result, ask whether readings are valid, duration is known, the zone-to-lot link is documented, and configuration changes are visible. Ask whether missing monitoring time is reported. If not, describe a sensor-data limitation and escalate the product question through the authorized process rather than inferring safety or loss from an alert count.
Record the review cutoff as well. A late sensor upload or later lot reconciliation can change the apparent status, but it should be shown as subsequent evidence rather than silently replacing the first assessment.
Reading uncertainty correctly
An alert without a valid reading or exposure link should remain a review event, not a product classification. Retain the sensor quality result, affected zone, and reason linkage failed. This makes it possible to improve monitoring coverage while keeping quality or regulatory disposition in the hands of the authorized decision maker.
Bounded conclusion
Refrigeration research is strongest when raw readings, validity, duration, affected unit, and disposition remain linked. The record can support a proportionate review. It cannot turn every alert into product loss or prove safety without the relevant facts.
Sources
Reviewed by the DispensaryVA editorial team on 2026-08-14.