Recall Forensics

Recall Forensics: The Lewis Sausage Warning Letter — the validation failure behind it

Team ValidLeaf · October 2026 · 6 min read

Recall Forensics is our series tracing real product recalls and enforcement actions to the specific validation gap behind them. Every fact comes from public regulator records or reporting of those records — no speculation, no invented causes. Last week we traced the warning letter sent to the brand in this supply chain. This is the other half: the letter sent to the plant that made the food. Its problem was not a missing kill step. Its problem was validation studies that could not prove the control worked — because they never measured what the control had to overcome.

What happened

On October 13, 2026, the U.S. Food and Drug Administration posted a warning letter to Lewis Sausage Company Inc. of Burgaw, North Carolina. The letter is dated August 5, 2026, is addressed to the company's general manager and owner, and carries CMS number 717432. FDA's Center for Veterinary Medicine issued it after a joint inspection with the North Carolina Department of Agriculture and Consumer Services (NCDACS) (FDA warning letter).

That inspection ran from August 25 to 29, 2025, as a follow-up to an Untitled Letter FDA sent the facility on March 14, 2025. Investigators found significant violations of the Current Good Manufacturing Practice, Hazard Analysis, and Risk-Based Preventive Controls for Food for Animals requirements in 21 CFR Part 507, and concluded the products were adulterated — prepared, packed, or held under conditions where they may have been contaminated (FDA).

The trigger was sampling done before the inspection began. On August 11, 2025, NCDACS, working with FDA, collected unopened samples of pet food Lewis contract manufactures, taken from a third-party wholesale distributor. Analysis found Salmonella Enteritidis in a chicken pet food mixer and a chicken dog food, and Listeria monocytogenes in a turkey and beef puppy food — samples FDA states were collected after the plant's preventive control was in place (FDA). Independent reporting by eFoodAlert identifies Lewis as the contract manufacturer for Viva Raw LLC, whose products were recalled on August 25, 2025, the day the follow-up inspection opened. FDA redacted the client's identity from both letters as confidential commercial information, so that link rests on eFoodAlert's reporting, not the letter itself (eFoodAlert).

The plant's paperwork named the right hazards. Its hazard analysis, dated July 28, 2025, identified Salmonella, L. monocytogenes, and Escherichia coli O157:H7 as hazards requiring a preventive control. FDA's first citation, under 21 CFR 507.34(a)(1), is that the control chosen was not adequate — the finished-product positives are its evidence. A smaller detail tells the same story: the written procedure did not establish the mixing time the company described verbally, and on August 27, 2025 investigators watched mixing times of 153, 142, and 75 seconds during production of a turkey food for cats (FDA).

Lewis responded twice, on September 22, 2025 and March 19, 2026. The second response replaced the original in-process control with a treatment applied to finished products in their final packaging, carried out by a contracted third-party facility. FDA answered that it could not fully evaluate that change, because the company provided no documentation showing the new control had actually been implemented for the finished products it manufactures (FDA).

The second citation, under 21 CFR 507.47(a), is the centre of this case: validation itself. The original control rested on studies of a probiotic ingredient from a third-party consulting laboratory. FDA reviewed them and concluded they do not support the claim that the probiotic is an effective preventive control — and they never addressed E. coli O157:H7, though the hazard analysis had named it. The replacement control's studies tested inactivation of Salmonella, L. monocytogenes, and Shiga toxin-producing E. coli, but did not determine the pathogenic load in the product before treatment. Without the typical starting load, FDA said, it cannot judge whether the reduction achieved is enough to significantly minimize the pathogens. The studies did report general microbial counts — which, FDA noted, describe the overall microbial population, not the pathogens of concern (FDA).

An honesty note, because we hold ourselves to it: FDA redacted the brand and product names, lot numbers, and the names of both treatment steps from this letter, so this article does not name those steps beyond what the letter discloses. The letter states no illness count, and we infer none. What is on the record is enough: a hazard analysis that named three pathogens, a control that finished-product testing contradicted, and two rounds of validation studies that could not carry the claim built on them.

The validation gap: a reduction claim with no starting point

A log reduction is a subtraction. It only means something if you know the number you started from. That arithmetic is where this food safety plan failed, and it fails in ways any pet food manufacturer — or any maker of a no-kill-step product — can repeat:

• Validate against a measured baseline, not an assumed one. A study that reports a reduction without first determining the load the product actually carries into treatment cannot show the treatment is strong enough for real production. Worst-case challenge levels must come from the product, not from levels the study happens to pass at.
• Cover every hazard your own analysis named. The hazard analysis listed three organisms; the validation addressed two. A hazard-coverage check — each hazard requiring a control, mapped to the study, parameter, and acceptance criterion that controls it — would have exposed the missing third organism on paper, before an inspector did.
• Indicators are not pathogens. General microbial counts are useful trend tools, but they measure the overall population, not the specific organisms the control exists to defeat. An acceptance criterion written around an indicator can pass every lot while the hazard it stands in for goes unmeasured.
• A parameter that matters must be written down. Mixing times on the floor ranged from 75 to 153 seconds for a step whose written procedure set no time at all. If a parameter affects whether a control works, it belongs in the procedure with limits, monitoring, and a record — a verbal number cannot be trained, audited, or defended.
• Outsourcing a step does not outsource the evidence. When Lewis replaced its control with a treatment performed by a third-party facility on finished product, FDA still asked Lewis for documentation that the treatment had been implemented. Contracting the work changes who runs the step; it does not change who must hold the validation file and the lot-by-lot proof.

None of this is exotic. It is the ordinary structure of validation: define the hazard, measure the starting condition, set the criterion in advance, measure the hazard itself, and keep records a stranger can audit. This plant had studies — plural, from a consulting laboratory, submitted across two responses. What it lacked was a study designed to answer the question FDA actually asked.

How a GMP Readiness Assessment catches it

A GMP Readiness Assessment is built to find this gap while it is still a paper problem. Over four weeks, it reviews:

• Validation study design — does each study determine the baseline load, challenge the control at worst case, and measure the named hazard itself rather than an indicator?
• Hazard-to-control coverage — is every hazard in the analysis mapped to a validated control with preset acceptance criteria, with no organism left named but unaddressed?
• Procedure-versus-practice alignment — do the critical parameters in your procedures match what operators actually run on the floor, within written limits?
• Contracted-step evidence — for any control performed by a third party, can you produce the validation file and lot-level implementation records on request?
• Corrective-action quality — when a control is replaced, is the replacement validated and documented before it becomes the plan's answer?

It closes with a gap assessment, a remediation roadmap, and starter IQ/OQ/PQ templates — so the next study is designed to survive the question, not just to exist.