Skip to content

Systems and records

Food cold chain excursion data: the decision sequence in one alarm

By SourceX Editorial · Reviewed by Noah Loul ·

9 sections

Short answer

A food temperature excursion case should hold the decision sequence, not just the alarm: the readings and the limit that applied, who was alerted and when, how the product was held, who judged its safety and on what evidence, the disposition of each lot, and a later record proving the outcome. Linked by lot code and time, those steps make cold chain excursion data replayable and gradable.

Video and sensors

Food cold chain excursion

This page covers food cold chain excursion data: the decision sequence in one alarm.Illustrative

Key takeaways

  • An alarm log shows a reading crossed a limit; an excursion case shows what people decided and what happened to each lot.
  • Build one case per excursion per lot, because one alarm on a mixed load can end in different dispositions.
  • U.S. transport rules place the go or no-go call with a qualified individual, so the evaluator's role and evidence are required fields.
  • EPCIS 2.0 sensor reports and persistent dispositions can carry the readings, alerts and holds without a new format.
  • A validator that checks order, coverage, balance and authority rejects most broken exports before anyone grades a decision.
  • Cases tied to illness, recalls or inspections stay out, and access to a record is not a right to license it.

Why one alarm log does not make an excursion case

An alarm log records that a reading crossed a limit; an excursion case records what people did about it. A useful case follows one event from the first out-of-range reading through notification, product hold, safety evaluation, disposition and the record that proves where each lot ended up.

Most cold chain systems split those steps across tools. The reefer controller or data logger holds the readings, a telematics portal or warehouse monitoring system holds the alert, email or a ticket holds the response, and the quality or ERP system holds the hold and the disposition. Each record is accurate on its own. None of them alone shows the decision.

There is a second gap that matters for anyone building training data. FDA's preventive controls rule allows monitoring records for refrigerated storage to be either affirmative records that show temperature stayed in control or exception records that show only a loss of control. A site that keeps exception records only has no record of the normal loads. Its excursion data is a list of failures with no denominator, which limits what any model can learn about how often excursions happen or which lanes are risky.

So the first decision in building a food cold chain excursion dataset is the unit. This guide uses one case per excursion event per affected lot, because dispositions are made lot by lot and a single alarm on a mixed load can end in several different outcomes.

The rules that set the decision points in a U.S. excursion

In U.S. food transport and manufacturing, three rule families decide where the human decisions sit, and the schema should give each one its own field rather than a free-text note.

The sanitary transportation rule at 21 CFR 1.908 is the clearest. When a shipper, loader, receiver or carrier becomes aware of a possible material failure of temperature control, the food may not be sold or otherwise distributed unless a qualified individual determines that the deviation did not make it unsafe. The same section has shippers give carriers a written operating temperature, requires written procedures from shippers of food that needs temperature control for safety, and lets the shipper or receiver ask the carrier to show that it kept the agreed temperature conditions during the trip.

Inside a facility, the preventive controls rule frames the response. Written corrective action procedures under 21 CFR 117.150 must cover correcting the problem, reducing the chance it recurs, evaluating affected food for safety and keeping affected food out of commerce, and corrective actions must be documented. Monitoring itself is governed by 21 CFR 117.145, which requires written procedures and records.

At retail and in food service, the FDA Food Code sets cold holding for time and temperature controlled foods; FDA's guidance on cut leafy greens, for example, states they must be kept at 41 degrees F (5 degrees C) or below during cold storage and display. States adopt the Food Code on their own schedule, so the limit field in the schema should record which rule or specification supplied the number, never a single assumed threshold.

The rules that set the decision points in a U.S. excursion
Decision pointWho decidesRule or document that frames itField it becomes
Is this a possible loss of temperature control?Whoever becomes aware: shipper, loader, carrier or receiver21 CFR 1.908 and the shipper's written operating temperatureexcursion.detected_by, excursion.limit_source
Hold or keep moving?Carrier and shipper under their written agreementShipper's written procedures and the carrier agreementresponse.hold_action
Is the food still safe?A qualified individual21 CFR 1.908; plant food safety planevaluation.decision, evaluation.basis
What happens to each lot?Shipper or facility quality functionCorrective action procedures under 21 CFR 117.150disposition.code per lot
Was the cause fixed?Facility or carrier21 CFR 117.150 corrective action elementscorrective_action.recurrence_step

The sensor event, response and disposition linkage schema

The schema is the asset. Every case has seven linked blocks, and each block carries a key that joins it to the one before. A validator can then check the chain mechanically before anyone grades a decision. Field names are a suggestion; the linkage rules are the point.

Use one row per block below as a field checklist when you extract a case from a reefer download, a warehouse monitoring export and the quality system.

The sensor event, response and disposition linkage schema
BlockMinimum fieldsTypical source recordLinkage and validator rule
Casecase_id, schema_version, lot_id, product description, quantity and unitBill of lading, receiving record, lot masterExactly one lot per case; schema_version present
Sensor eventdevice_id, sensor type, reading series, unit, min and max in window, first and last out-of-limit timeReefer download, data logger export, warehouse monitoringReadings cover the whole exposure window with no unexplained gap
Limitlimit value, unit, limit_source (written operating temperature, specification, code section)Shipper instructions, product specificationLimit and readings use the same unit or a stated conversion
Notification and responsewho was alerted, alert time, response action, response time, actor roleTelematics alert, dispatch notes, email or ticketResponse time is at or after the first out-of-limit reading
Safety evaluationevaluator role, qualification basis, evidence reviewed, decision, decision timeQA evaluation form, deviation recordDecision time is after the readings it cites were downloaded
Dispositionper-lot code (released, released with restriction, reworked, donated, destroyed, returned), quantity, approver roleQuality hold release, inventory adjustmentDisposition quantities sum to the case quantity
Corrective actioncause category, action taken, recurrence step, owner roleCorrective action record, maintenance work orderAt least one of correction or recurrence step is recorded
Verified outcomeoutcome record type and ID, outcome date, later complaint or hold link if anyShipment, destruction certificate, later hold or complaintOutcome comes from a record dated after the disposition

Mapping the sensor and disposition fields to GS1 EPCIS 2.0

If your systems already emit GS1 EPCIS 2.0 events, the sensor block does not need a new format. EPCIS 2.0 added a sensor element to events, and its published JSON schema defines a sensor report with a measurement type, a value with a unit of measure, minimum, maximum and mean values, a device ID, the business rules applied and an exception field. GS1's conformance test for sensor data uses a reading of type gs1:Temperature with values in degrees Celsius.

The exception field takes one of two sensor alert types defined in the GS1 vocabulary, an alarm condition or an error condition. That distinction is worth keeping. An alarm says the product may have been exposed; an error says the device may be wrong. They lead to different evaluations, and a dataset that merges them will teach a model to treat a faulty probe like a warm trailer.

The Core Business Vocabulary also lists standard disposition values, including in transit, non-sellable other, damaged, destroyed, disposed, recalled and sellable accessible. EPCIS 2.0 adds a persistent disposition that can be set and unset on later events, which is a natural way to model a quality hold that is placed on receipt and lifted after evaluation.

Mapping the sensor and disposition fields to GS1 EPCIS 2.0
Schema blockEPCIS 2.0 elementExample valueNote for annotators
Sensor eventsensorElementList, sensorReport type and value with uomgs1:Temperature, CELKeep raw readings; do not replace them with a summary flag
LimitsensorReport minValue and maxValue plus bizRulesReference to the written operating temperaturebizRules points to the rule; the value still lives in the limit block
AlertsensorReport exceptionALARM_CONDITION or ERROR_CONDITIONError conditions route to device review before any product decision
HoldpersistentDisposition setnon_sellable_otherUnset on the event that records release or destruction
Dispositiondisposition on the closing eventsellable_accessible, destroyed or disposedOne closing event per lot or sub-lot

Tracing one reefer alarm on bagged salad from reading to lot outcome (illustrative)

Everything in this example is invented to show the structure. It is not a real shipment, customer or result, and the durations and counts are illustrative.

A refrigerated trailer carries three lots of bagged salad from a packer to a distribution center on a trip of about nine hours. The shipper's written operating temperature is the limit. About three hours in, the reefer download later shows the return-air reading above the limit for roughly 70 minutes after a defrost cycle stuck on.

  • Case block: one case per lot, three cases in total, each with the lot ID, product description, case count and unit from the bill of lading.
  • Sensor block: device ID of the reefer unit, the full reading series for the trip, the first and last out-of-limit times, and the maximum reading in the window, all in one unit.
  • Notification: the telematics alert reached dispatch within minutes; dispatch called the driver, who reset the unit. Both times and both actor roles are recorded.
  • Hold: on arrival the receiver placed all three lots on hold, recorded as a persistent non-sellable disposition.
  • Evaluation: the shipper's qualified individual reviewed the download, product core temperatures taken at receipt and the product specification, and recorded a decision per lot with the evidence list.
  • Disposition: two lots that sat in the front of the trailer, where pulp readings stayed in range, were released; the lot nearest the rear doors was destroyed. Quantities sum to the shipment.
  • Corrective action: the carrier logged a maintenance work order for the defrost timer and added a pre-trip defrost check as the recurrence step.
  • Outcome: the released lots link to outbound shipments dated the next day, and the destroyed lot links to a destruction record. No later complaint or hold is linked to any of the lots.

What does a broken export of the same trailer alarm look like?

A failing record is just as useful as a valid one, because it shows the validator what to reject. Here is the same event as it often arrives from a real system export, again illustrative.

At a glance

What does a broken export of the same trailer alarm look like

The export has one case for the whole trailer rather than one per lot, a disposition of released for everything, an evaluation timestamp that is 40 minutes earlier than the reefer download, a sensor series with a 2-hour gap during the excursion window, and no link from the destroyed product to any lot. A model trained on that record would learn that a long alarm ends in release with no evidence, which is the opposite of what happened.

  • Unit check: reject a case that covers more than one lot when the dispositions differ.
  • Order check: reject an evaluation dated before the data it cites was downloaded or received.
  • Coverage check: flag any gap in readings that overlaps the exposure window; the case is not gradable until the gap is explained.
  • Balance check: disposition quantities must add up to the case quantity, with no lot left without an outcome.
  • Authority check: an evaluation decision without an evaluator role and a stated basis is incomplete, because 21 CFR 1.908 places that determination with a qualified individual.
  • Alert-type check: an error condition cannot lead straight to destruction or release without a device review step.
  • Outcome check: the verified outcome must be a record dated after the disposition, not the status field on the hold.

Can the case be replayed through lot codes and quality decisions?

Yes, if every block carries a lot identifier that matches the traceability records, and the case keeps the version of the rule or specification that set the limit.

For foods on FDA's Food Traceability List, receivers must keep records under 21 CFR 1.1345 that tie the traceability lot code to the quantity and unit of measure, product description, receiving location, receipt date, the traceability lot code source and the reference document. Those receiving fields are exactly the join keys the case block needs. If the excursion case uses the same lot code and reference document, an auditor or an evaluation harness can walk from the alarm to the receiving record to the disposition without guessing.

Replay breaks in three predictable places: lots split at receipt with no parent and child link, product rework or repacking that creates a new lot without recording the input lot, and dispositions recorded against a pallet or trailer rather than the lot. Fix those links during extraction, or mark the case as not replayable and keep it out of grading sets.

  • Start from the lot code on the disposition and find the matching receiving record.
  • Confirm the receiving date falls after the excursion window and the location matches the trip's destination.
  • Walk back from the receiving record to the shipment and the reefer unit that carried it.
  • Confirm the readings, the limit source and the evaluation cite the same trip.
  • Walk forward from the disposition to the outbound shipment, destruction or later hold record.
  • Record the result as replayable, replayable with a documented split, or not replayable.

Excursion cases to hold back or send to a food safety reviewer

Some cases should never enter a training set, and some can only enter after a food safety reviewer looks at them. Excursions linked to an illness complaint, a recall, a regulatory inspection or litigation stay out. So do cases where the safety evaluation is missing or was made by someone whose role cannot be established.

Cases where the evaluation relied on product testing, an extended shelf-life argument or a specification exception go to a qualified food safety reviewer before annotation, because a labeler cannot judge whether the basis was sound. The annotation records what was decided and on what evidence; it never asserts that a released lot was safe.

Rights need a separate check. The carrier may control the reefer download and telematics account while the shipper controls its loggers and written instructions, so having access to a record does not mean your company may license it. Driver names, phone numbers and GPS traces are personal data and should be removed or generalized. Removing them protects privacy but does not by itself create a right to license the data; the contracts that govern each record decide that.

Licensing excursion histories as decision data

Linked excursion histories are scarce because the blocks live in different companies' systems. A food business, carrier or cold storage operator that can assemble them with the linkage above holds records that show how qualified people actually decide under time pressure, which is hard for AI developers to find anywhere else.

For a food business or carrier that wants to license such histories, SourceX runs the whole path: finding and qualifying the records, reviewing rights, contracting, preparing and delivering the dataset, and paying the company. As a rough guide, buyers look for consistent history of two years or more and tens of thousands of records in one system, although rarer decision records can qualify at smaller volume; each dataset is assessed on its own.

Not one excursion case is handed over until your company has agreed the scope and terms and signed. To learn whether your records could qualify, start with a short assessment of the business; it asks for no reefer downloads or other data.

Frequently asked questions

What is food cold chain excursion data?

It is the set of records created when a refrigerated or frozen food leaves its required temperature range: sensor readings, the limit that applied, alerts, holds, safety evaluations, dispositions and the corrective action. It becomes useful for AI when those records are linked lot by lot so the decision sequence can be followed.

Who decides whether food can be sold after a temperature excursion in transport?

Under 21 CFR 1.908, food affected by a possible material failure of temperature control may not be sold or distributed unless a qualified individual determines the deviation did not make it unsafe. In practice the shipper's procedures and the carrier agreement say who performs that evaluation.

Do cold storage monitoring records have to show every reading?

Not always. FDA's preventive controls rule allows refrigeration monitoring records to be affirmative records showing control or exception records showing a loss of control. Datasets built only from exception records have no normal loads to compare against, which limits what they can teach.

Can GS1 EPCIS carry temperature excursion data?

Yes. EPCIS 2.0 added a sensor element whose reports carry a measurement type such as temperature, values with units, minimum and maximum values, device IDs and an alarm or error exception. Dispositions and persistent dispositions on events can model holds and releases.

Why is an alarm different from an error in excursion data?

An alarm condition says the product may have been exposed to an out-of-range temperature. An error condition says the sensor or device may be faulty. Keeping them separate prevents a model from learning that a broken probe and a warm trailer call for the same response.

Can a carrier license excursion records without the shipper?

It depends on which records the carrier controls and what its contracts say. Reefer downloads and telematics data may sit with the carrier, while shipper loggers and instructions may not, and the governing agreements decide reuse. Counsel should review the specific contracts.

Sources

  • 21 CFR 1.908 bars sale or distribution of food after a possible material failure of temperature control unless a qualified individual determines it was not made unsafe, and sets shipper and carrier temperature duties. Source
  • 21 CFR 117.145 requires written monitoring procedures and records, and allows refrigeration monitoring records to be affirmative or exception records. Source
  • 21 CFR 117.150 lists the elements of corrective action procedures, including evaluating affected food and keeping it out of commerce, and requires corrective actions to be documented. Source
  • 21 CFR 1.1345 lists the receiving records for foods on the Food Traceability List, keyed to the traceability lot code. Source
  • FDA retail guidance states that cut leafy greens must be kept at 41 F (5 C) or less during cold storage and display. Source
  • The EPCIS 2.0 JSON schema defines the sensor report fields (type, value, unit, minimum, maximum, device, business rules, exception) and persistent dispositions that can be set and unset. Source
  • The GS1 sensor alert type vocabulary defines an alarm condition and an error condition. Source
  • The GS1 Core Business Vocabulary disposition list includes in transit, non-sellable other, damaged, destroyed, disposed, recalled and sellable accessible. Source
  • GS1's EPCIS 2.0 conformance test for the sensor element uses a temperature reading of type gs1:Temperature expressed in degrees Celsius. Source

See if your company qualifies

A short company assessment. No data uploads are needed.

See if your company qualifies