When Two Labs Disagree on a Frozen-Food Quality Result: Retest or Resample?

The decision matters most near a contractual limit. A frozen corn Brix result just above a minimum and another just below it can tempt both parties to commission a third test and accept whichever report they prefer. That approach does not locate the source of disagreement. The two results may use different product fractions, units, thawing procedures, methods, calibration conditions or sample locations. The qualified technical teams should define a diagnostic comparison and the applicable decision rule before further material is consumed.

This article concerns physicochemical and other non-pathogen quality measurements such as Brix, acidity, size, moisture or defect percentage. A suspected food-safety positive requires the applicable hold, confirmation and disposition process; a later negative result cannot simply erase it. We can provide GreenLand lot and product documentation, while the buyer, laboratories and responsible QA personnel agree how disputed results will be investigated.

Secure the original evidence before another analysis

Ask both laboratories for the complete result package, not just the reported number. It should identify the lot, submitted pack or sample ID, collection date, receipt condition, test date, method and revision, test portion, preparation, units, calculation, quality controls and the person or system authorizing the report. Preserve raw observations and any retained material under its required conditions. An amended report should retain the original version and explain the correction. Without this evidence, the third test may produce another number but leave the first disagreement unexplained.

The sample chain is part of the measurement. One laboratory may have received individually sealed IQF kernels at stable frozen temperature. The other may have received a thawed composite or a bag opened during an earlier inspection. Those materials do not have the same history. If the product is frozen fruit, losing thaw liquid can alter the portion presented for soluble-solids or acidity measurement. If it is whole corn, selecting different cob positions may change a result. Write these differences down before assuming an instrument is faulty.

GreenLand corn illustrates a frozen-food matrix; no laboratory dispute is represented.

Preservation also protects the commercial decision. When a disputed shipment remains in storage, identify affected lots and cartons, note temperature and location, and avoid mixing them into unrestricted stock before qualified QA gives a disposition. The agreement may allow a hold, additional inspection, conditional use or rejection, but these outcomes depend on the contract, destination requirements and actual evidence. A sample investigation should not silently change the lot available for inspection.

Keep a clear record of who collected each original sample and how it was split. Was one common sample sent to both labs, or did each draw from different cartons? Were the original test portions sealed separately? Is a stable reserve available? The GreenLand quality-inspection sampling plan describes representative carton selection. That selection remains relevant to a lab disagreement because a precise result from a biased location cannot stand for the whole lot.

At receipt, photograph labels and seals before opening disputed material. Note any leakage, excess frost or damage that could affect the attribute under test, then let the laboratory decide whether the sample remains suitable. A photograph does not replace a measurement, but it links the later result to the condition that actually reached the bench. If two labs received different physical states, the investigation can address that difference directly. If the original packs were discarded, a later resample can evaluate the lot but cannot reconstruct what either lab received on the first day.

The first review often costs less and reveals more than a rushed retest. A transposed decimal, wrong dilution factor, mismatched unit or outdated specification revision can create a disagreement without any product difference. The FAO's food-analysis quality-control discussion describes independent calculation review as a useful early check. It does not mean every discrepancy is a clerical error; it means the record should be tested for simple causes before more samples are used.

Locate the disagreement in units, methods and preparation

Put the two reports on a common basis. Confirm whether the measurand is the same: Brix of an expressed juice, soluble solids in a prepared puree, titratable acidity expressed as a named acid, moisture by a particular drying method, or percentage of broken pieces by weight. Similar labels can conceal different measurement definitions. If one result is reported in a different unit or at a different reference temperature, a numerical comparison without conversion or method context can be misleading.

Next compare the product fraction. Did both labs receive the same production lot, cut, grade and packing? Did one test whole kernels while the other included released liquid or broken pieces? Did one composite several packs while the other tested a single retail bag? The result may be accurate for each submitted sample and still not answer a common lot question. An agreement that defines only "test corn Brix" leaves too much of this to improvisation after a dispute.

GreenLand product photo illustrates a possible lot form; it is not laboratory evidence.

Preparation steps can create substantial differences in heterogeneous frozen foods. Homogenization time, particle size, drainage, thawing temperature, time before measurement and the number of increments influence what reaches the instrument. A laboratory may follow a valid internal procedure that differs from the method named in the purchase contract. Ask both laboratories to provide the operative procedure and deviations, then decide whether results can be compared directly. If they cannot, the next experiment should control that specific difference.

For visual defect counts, clarify the denominator and classification rule before comparing percentages. One inspector may count partly broken corn kernels by number, while another reports mass of broken pieces. A fruit pack with a few large damaged berries can look severe by count and less severe by weight, depending on the agreed criterion. The test report needs to state how borderline pieces were classified and whether ice or glaze was removed. These are examples of measurement definition, not claims that any particular laboratory made such an error.

Review method capability and calibration evidence for the relevant range and matrix. A method validated for a clear liquid does not automatically perform identically on pulpy fruit or frozen vegetable slurry. Quality-control failures, out-of-range calibration or a changed reagent lot may require laboratory investigation. These are technical questions for the analysts. The supplier and buyer should avoid calling a lab "wrong" because its number is less favorable; first identify whether it measured the intended material under an accepted method.

Near a limit, the pre-agreed decision rule also matters. ILAC G8 gives guidance on decision rules and conformity statements for accredited laboratory work, but the applicable rule can be specified by regulation, standard or contract. The parties should know whether and how measurement uncertainty enters an acceptance statement. This article does not choose a universal guard band. A third result is not a substitute for agreeing how a result near the limit is interpreted.

What retesting the same homogenate can establish

If a stable, properly retained homogenate exists, equivalent portions drawn from it can help assess analytical repeatability under controlled conditions. The same laboratory might repeat the method, or both labs might receive blinded split portions from one common preparation. If results align on that shared material, some of the original disagreement may have arisen before analysis, such as sample selection or preparation. If results still diverge, the labs should investigate method, calibration, analyst and reporting differences.

The inference is narrower than many commercial summaries suggest. Matching results on one homogenate do not prove that the lot is uniform. The homogenate has already averaged any variation in the source material. Nor does a repeat from the same container erase the first report automatically. It adds evidence about the analytical stage under the new controlled conditions. To compare results fairly, the material must be stable for the target measurement through storage, splitting and transport; otherwise the diagnostic test introduces a new cause of difference.

A retained homogenate addresses measurement repeatability only.

The practical endpoint of this stage is a written question answered or still open. If the two labs converge on common material, mark analytical disagreement as less likely under the controlled comparison, then investigate the original preparation and sampling records. If they diverge, review each lab's controls, calculations and method implementation. Avoid declaring one lab reliable solely because its value agrees with a preferred earlier value. Agreement is evidence only when the material, method and acceptance criteria made the comparison meaningful.

Define the split before opening the reserve. A custodian can identify the source, mix it according to the accepted procedure, prepare equivalent sealed portions and assign blind codes. Document portion mass, container, dispatch conditions and time to analysis. The laboratories should use the same agreed method where the purpose is to compare their execution of that method. If the purpose is to compare two methods, that is a different study and needs a stated interpretation plan. Do not blend these purposes after seeing the numbers.

An illustrative frozen-corn case makes the distinction clear. Suppose two reports disagree on a lot-average Brix value. A common homogenate prepared from the retained composite is divided into two coded aliquots. If both labs report comparable values under the same method, the difference may lie in their original samples or preparation. If the values remain apart, laboratory execution or method details deserve closer review. This example is a proposed diagnostic design, not a GreenLand experiment or proof of a particular error source.

The FAO discussion distinguishes repeat analysis of the same analytical sample from preparing a new analytical sample from the original food material. That distinction is useful because the latter includes another preparation step. When the original homogenate is not stable or unavailable, a same-homogenate retest may not be technically valid. The lab should advise whether it can retain or reprepare material for the intended analyte before anyone describes the test as a replicate.

What independent preparation and fresh lot sampling add

Preparing a new analytical sample from the original sealed pack or laboratory sample tests a broader chain than repeating a portion of an existing homogenate. It includes selection within the pack, thawing, grinding or blending, dilution and the analytical method. If repeated preparation from the same retained pack produces variable results, preparation or within-pack heterogeneity may be contributing. If it produces consistent results that differ between labs, the method or laboratory conditions remain candidates for investigation.

Fresh lot sampling broadens the question again. New cartons may reveal variation across production time, pallet position, pack size or handling route. They can inform the current state of the defined lot and whether the original material was unusual. They do not recreate the exact portion that led to the first report. If the shipment has been handled, repacked or partially used since then, the new sample may also have a different history. The sampling map and condition record must accompany the result.

New primary units test whether the original sample represented the lot.

A useful investigation ladder therefore moves from least disruptive checks to more extensive material collection: confirm identity and calculations; review methods and preparation; examine retained aliquots where technically suitable; reprepare from sealed original material; and finally sample the defined lot under a new plan when needed. This is a diagnostic sequence, not a required number of tests. Stop or change course when a cause is established or when the available material cannot answer the intended question.

There is a statistical consequence to each step. Repeat readings from one preparation chiefly describe within-preparation analytical variation. Independent preparations add preparation variation. New cartons add sampling and lot variation. Eurachem's guide on uncertainty from sampling is a technical reference for separating sampling and physical-preparation contributions. A purchasing team need not calculate every component, but it should avoid treating all three types of replicate as equivalent evidence of lot compliance.

Sampling plans need to account for the product's distribution. Whole frozen fruit, corn kernels and cut vegetables may vary by pack, pallet, production time and maturation. A few convenience cartons cannot estimate that variation reliably. A qualified specialist should set the number and placement of new units from the lot definition, intended inference and decision risk. The GreenLand frozen-corn product information shows that cob, segment and kernel forms are different commercial offers; a new sample of one form cannot validate a disputed result for another.

If the remaining lot is no longer intact, record that limit. An importer may have released some pallets to customers while the dispute was being investigated. New cartons from what remains might overrepresent one production interval or storage position. They may still be worth testing for current stock disposition, but the result should not be presented as a clean rerun of the original lot plan. Involving the warehouse and logistics team early can preserve a usable sampling frame and avoid a technical argument about material that can no longer be located.

How to design a blinded two-laboratory comparison

A controlled comparison is most useful when both labs measure materially equivalent portions and do not know which one is expected to pass. The study plan should identify the measurand, method, matrix, preparation, number of portions, stability condition, blinding, reporting unit and how results will be interpreted. A qualified coordinator should retain the code key and maintain a custody record. If one lab receives a different lot or a warmer shipment, the comparison loses its diagnostic force.

The design can vary with the question. To compare analytical execution, use one stable homogenate split into equivalent portions. To compare preparation, send equivalent original material with a common preparation instruction and then review how each lab performed it. To evaluate lot variation, draw independently identified cartons under a defined sampling map and avoid hiding those identities inside one untraceable mixture. More than one stage may be needed, but each stage should have a reason.

Blind comparison can isolate procedural differences when the protocol is agreed.

The NIST overview of interlaboratory comparisons describes study designs used to examine random and systematic differences among laboratories and emphasizes stable comparable material. A two-laboratory commercial investigation is smaller and cannot claim the precision of a broad proficiency study. The relevant principle remains: shared material and documented conditions make a difference interpretable. If the split is unstable, no amount of analytical sophistication repairs that initial weakness.

Agree on the decision path before results arrive. Will the labs compare raw values, uncertainty statements, method control data or a contractual conformity outcome? Who reviews a discrepancy? When would a third independent laboratory be involved, and what material would it receive? A pre-agreed process reduces the temptation to select the report that favors one party. If the original dispute concerns a customer complaint, keep the customer-facing disposition with qualified QA rather than letting a convenient laboratory number drive it alone.

The comparison record should be readable by someone who did not attend the calls: sample IDs, source lot, pack and portion history, test requests, actual methods, reports, deviations, calculations, interpretation and conclusion. Retain reports that do not support the final decision as well as those that do. Removing an inconvenient value from the file makes future audits and repeat-order decisions less trustworthy.

The laboratories should also distinguish an ordinary method-control problem from a genuine difference in food material. If a control failed, the lab may have an internal rule for invalidating a run, and that rule should be documented. If controls passed but independently prepared portions differ, the variation may be real at the sampling scale. A third laboratory cannot resolve that by measuring an unrelated carton. The next step must be designed around the uncertainty that remains, with enough retained or newly sampled material to support it.

How to apply the agreed decision rule and close the record

The closure should identify which part of the chain was resolved and which remains uncertain. A calculation correction may fully explain a difference. A shared homogenate may show that the laboratories agree analytically but leave lot sampling unresolved. New carton results may reveal product heterogeneity without proving exactly why the original reports differed. State the inference supported by the evidence, rather than giving every investigation a simple winner and loser.

Apply the specification revision and decision rule effective for the order. A limit for a lot average should be compared with a result generated under the agreed lot-level plan. A limit for individual packs should not be settled by averaging results from different units unless the contract allows that interpretation. If the result lies close to a limit, have the qualified parties apply the agreed treatment of uncertainty. Do not invent a rule from ILAC guidance or borrow one from an unrelated destination market after seeing the data.

GreenLand product photo illustrates physical product identity, not a retest result.

GreenLand product photo illustrates a different presentation; it is not a test result.

The final file can name any corrective action: a clarified test request, revised sampling map, common preparation method, laboratory method review, additional retain procedure or an amendment to the purchase specification. These changes protect future orders only when they are communicated to the teams collecting and testing samples. If the present lot remains on hold, record who has authority to release, rework, redirect or reject it and the evidence on which that decision rests.

For GreenLand-food supply discussions, we can confirm the product form, size or cut, grade, packaging, quantity, application, destination and requested documents before an order is placed. Those inputs help the buyer and laboratory define the correct measurement and sample unit. When two labs disagree, the fastest useful action is often to identify what they compared. Retest a shared material when the analytical stage is the question; reprepare when preparation is the question; resample when the defined lot is the question. Keep the original evidence and a pre-agreed decision rule through every step.