The Hidden Contamination Hazards: What the Catalent Warning Letter Reveals About Systemic Aseptic Processing Failures

The November 2025 FDA Warning Letter to Catalent Indiana, LLC reads like an autopsy report—a detailed dissection of how contamination hazards aren’t discovered but rather engineered into aseptic operations through a constellation of decisions that individually appear defensible yet collectively create what I’ve previously termed the “zemblanity field” in pharmaceutical quality. Section 2, addressing failures under 21 CFR 211.113(b), exposes contamination hazards that didn’t emerge from random misfortune but from deliberate choices about decontamination strategies, sampling methodologies, intervention protocols, and investigation rigor.

What makes this warning letter particularly instructive isn’t the presence of contamination events—every aseptic facility battles microbial ingress—but rather the systematic architectural failures that allowed contamination hazards to persist unrecognized, uninvestigated, and unmitigated despite multiple warning signals spanning more than 20 deviations and customer complaints. The FDA’s critique centers on three interconnected contamination hazard categories: VHP decontamination failures involving occluded surfaces, inadequate environmental monitoring methods that substituted convenience for detection capability, and intervention risk assessments that ignored documented contamination routes.

For those of us responsible for contamination control in aseptic manufacturing, this warning letter demands we ask uncomfortable questions: How many of our VHP cycles are validated against surfaces that remain functionally occluded? How often have we chosen contact plates over swabs because they’re faster, not because they’re more effective? When was the last time we terminated a media fill and treated it with the investigative rigor of a batch contamination event?

The Occluded Surface Problem: When Decontamination Becomes Theatre

The FDA’s identification of occluded surfaces as contamination sources during VHP decontamination represents a failure mode I’ve observed with troubling frequency across aseptic facilities. The fundamental physics are unambiguous: vaporized hydrogen peroxide achieves sporicidal efficacy through direct surface contact at validated concentration-time profiles. Any surface the vapor doesn’t contact—or contacts at insufficient concentration—remains a potential contamination reservoir regardless of cycle completion indicators showing “successful” decontamination.

The Catalent situation involved two distinct occluded surface scenarios, each revealing different architectural failures in contamination hazard assessment. First, equipment surfaces occluded during VHP decontamination that subsequently became contamination sources during atypical interventions involving equipment changes. The FDA noted that “the most probable root cause” of an environmental monitoring failure was equipment surfaces occluded during VHP decontamination, with contamination occurring during execution of an atypical intervention involving changes to components integral to stopper seating.

This finding exposes a conceptual error I frequently encounter: treating VHP decontamination as a universal solution that overcomes design deficiencies rather than as a validated process with specific performance boundaries. The Catalent facility’s own risk assessments advised against interventions that could disturb potentially occluded surfaces, yet these interventions continued—creating the precise contamination pathway their risk assessments identified as unacceptable.

The second occluded surface scenario involved wrapped components within the filling line where insufficient VHP exposure allowed potential contamination. The FDA cited “occluded surfaces on wrapped [components] within the [equipment] as the potential cause of contamination”. This represents a validation failure: if wrapping materials prevent adequate VHP penetration, either the wrapping must be eliminated, the decontamination method must change, or these surfaces must be treated through alternative validated processes.

The literature on VHP decontamination is explicit about occluded surface risks. As Sandle notes, surfaces must be “designed and installed so that operations, maintenance, and repairs can be performed outside the cleanroom” and where unavoidable, “all surfaces needing decontaminated” must be explicitly identified. The PIC/S guidance is similarly unambiguous: “Continuously occluded surfaces do not qualify for such trials as they cannot be exposed to the process and should have been eliminated”. Yet facilities continue to validate VHP cycles that demonstrate biological indicator kill on readily accessible flat coupons while ignoring the complex geometries, wrapped items, and recessed surfaces actually present in their filling environments.

What does a robust approach to occluded surface assessment look like? Based on the regulatory expectations and technical literature, facilities should:

Conduct comprehensive occluded surface mapping during design qualification. Every component introduced into VHP-decontaminated spaces must undergo geometric analysis to identify surfaces that may not receive adequate vapor exposure. This includes crevices, threaded connections, wrapped items, hollow spaces, and any surface shadowed by another object. The mapping should document not just that surfaces exist but their accessibility to vapor flow based on the specific VHP distribution characteristics of the equipment.

Validate VHP distribution using chemical and biological indicators placed on identified occluded surfaces. Flat coupon placement on readily accessible horizontal surfaces tells you nothing about vapor penetration into wrapped components or recessed geometries. Biological indicators should be positioned specifically where vapor exposure is questionable—inside wrapped items, within threaded connections, under equipment flanges, in dead-legs of transfer lines. If biological indicators in these locations don’t achieve the validated log reduction, the surfaces are occluded and require design modification or alternative decontamination methods.

Establish clear intervention protocols that distinguish between “sterile-to-sterile” and “potentially contaminated” surface contact. The Catalent finding reveals that atypical interventions involving equipment changes exposed the Grade A environment to surfaces not reliably exposed to VHP. Intervention risk assessments must explicitly categorize whether the intervention involves only VHP-validated surfaces or introduces components from potentially occluded areas. The latter category demands heightened controls: localized Grade A air protection, pre-intervention surface swabbing and disinfection, real-time environmental monitoring during the intervention, and post-intervention investigation if environmental monitoring shows any deviation.

Implement post-decontamination surface monitoring that targets historically occluded locations. If your facility has identified occluded surfaces that cannot be designed out, these become critical sampling locations for post-VHP environmental monitoring. Trending of these specific locations provides early detection of decontamination effectiveness degradation before contamination reaches product-contact surfaces.

The FDA’s remediation demand is appropriately comprehensive: “a review of VHP exposure to decontamination methods as well as permitted interventions, including a retrospective historical review of routine interventions and atypical interventions to determine their risks, a comprehensive identification of locations that are not reliably exposed to VHP decontamination (i.e., occluded surfaces), your plan to reduce occluded surfaces where feasible, review of currently permitted interventions and elimination of high-risk interventions entailing equipment manipulations during production campaigns that expose the ISO 5 environment to surfaces not exposed to a validated decontamination process, and redesign of any intervention that poses an unacceptable contamination risk”.

This remediation framework represents best practice for any aseptic facility using VHP decontamination. The occluded surface problem isn’t limited to Catalent—it’s an industry-wide vulnerability wherever VHP validation focuses on demonstrating sporicidal activity under ideal conditions rather than confirming adequate vapor contact across all surfaces within the validated space.

Contact Plates Versus Swabs: The Detection Capability Trade-Off

The FDA’s critique of Catalent’s environmental monitoring methodology exposes a decision I’ve challenged repeatedly throughout my career: the use of contact plates for sampling irregular, product-contact surfaces in Grade A environments. The technical limitations are well-established, yet contact plates persist because they’re faster and operationally simpler—prioritizing workflow convenience over contamination detection capability.

The specific Catalent deficiency involved sampling filling line components using “contact plate, sampling [surfaces] with one sweeping sampling motion.” The FDA identified two fundamental inadequacies: “With this method, you are unable to attribute contamination events to specific [locations]” and “your firm’s use of contact plates is not as effective as using swab methods”. These limitations aren’t novel discoveries—they’re inherent to contact plate methodology and have been documented in the microbiological literature for decades.

Contact plates—rigid agar surfaces pressed against the area to be sampled—were designed for flat, smooth surfaces where complete agar-to-surface contact can be achieved with uniform pressure. They perform adequately on stainless steel benchtops, isolator walls, and other horizontal surfaces. But filling line components—particularly those identified in the warning letter—present complex geometries: curved surfaces, corners, recesses, and irregular topographies where rigid agar cannot conform to achieve complete surface contact.

The microbial recovery implications are significant. When a contact plate fails to achieve complete surface contact, microorganisms in uncontacted areas remain unsampled. The result is a false-negative environmental monitoring reading that suggests contamination control while actual contamination persists undetected. Worse, the “sweeping sampling motion” described in the warning letter—moving a single contact plate across multiple locations—creates the additional problem the FDA identified: inability to attribute any recovered contamination to a specific surface. Was the contamination on the first component contacted? The third? Somewhere in between? This sampling approach provides data too imprecise for meaningful contamination source investigation.

The alternative—swab sampling—addresses both deficiencies. Swabs conform to irregular surfaces, accessing corners, recesses, and curved topographies that contact plates cannot reach. Swabs can be applied to specific, discrete locations, enabling precise attribution of any contamination recovered to a particular surface. The trade-off is operational: swab sampling requires more time, involves additional manipulative steps within Grade A environments, and demands different operator technique validation.

Yet the Catalent warning letter makes clear that this operational inconvenience doesn’t justify compromised detection capability for critical product-contact surfaces. The FDA’s expectation—acknowledged in Catalent’s response—is swab sampling “to replace use of contact plates to sample irregular surfaces”. This represents a fundamental shift from convenience-optimized to detection-optimized environmental monitoring.

What should a risk-based surface sampling strategy look like? The differentiation should be based on surface geometry and criticality:

Contact plates remain appropriate for flat, smooth, readily accessible surfaces where complete agar contact can be verified and where contamination risk is lower (Grade B floors, isolator walls, equipment external surfaces). The speed and simplicity advantages of contact plates justify their continued use in these applications.

Swab sampling should be mandatory for product-contact surfaces, irregular geometries, recessed areas, and any location where contact plate conformity is questionable. This includes filling needles, stopper bowls, vial transport mechanisms, crimping heads, and the specific equipment components cited in the Catalent letter. The additional time required for swab sampling is trivial compared to the contamination risk from inadequate monitoring.

Surface sampling protocols must specify the exact location sampled, not general equipment categories. Rather than “sample stopper bowl,” protocols should identify “internal rim of stopper bowl,” “external base of stopper bowl,” “stopper agitation mechanism interior surfaces.” This specificity enables contamination source attribution during investigations and ensures sampling actually reaches the highest-risk surfaces.

Swab technique must be validated to ensure consistent recovery from target surfaces. Simply switching from contact plates to swabs doesn’t guarantee improved detection unless swab technique—pressure applied, surface area contacted, swab saturation, transfer to growth media—is standardized and demonstrated to achieve adequate microbial recovery from the specific materials and geometries being sampled.

The EU GMP Annex 1 and FDA guidance documents emphasize detection capability over convenience in environmental monitoring. The expectation isn’t perfect contamination prevention—that’s impossible in aseptic processing—but rather monitoring systems sensitive enough to detect contamination events when they occur, enabling investigation and corrective action before product impact. Contact plates on irregular surfaces fail this standard by design, not because of operator error or inadequate validation but because the fundamental methodology cannot access the surfaces requiring monitoring.

The Intervention Paradox: When Risk Assessments Identify Hazards But Operations Ignore Them

Perhaps the most troubling element of the Catalent contamination hazards section isn’t the presence of occluded surfaces or inadequate sampling methods but rather the intervention management failure that reveals a disconnect between risk assessment and operational decision-making. Catalent’s risk assessments explicitly “advised against interventions that can disturb potentially occluded surfaces,” yet these high-risk interventions continued during production campaigns.

This represents what I’ve termed “investigation theatre” in previous posts—creating the superficial appearance of risk-based decision-making while actual operations proceed according to production convenience rather than contamination risk mitigation. The risk assessment identified the hazard. The environmental monitoring data confirmed the hazard when contamination occurred during the intervention. Yet the intervention continued as an accepted operational practice.

The specific intervention involved equipment changes to components “integral to stopper seating in the [filling line]”. These components operate at the critical interface between the sterile stopper and the vial—precisely the location where any contamination poses direct product impact risk. The intervention occurred during production campaigns rather than between campaigns when comprehensive decontamination and validation could occur. The intervention involved surfaces potentially occluded during VHP decontamination, meaning their microbiological state was unknown when introduced into the Grade A filling environment.

Every element of this scenario screams “unacceptable contamination risk,” yet it persisted as accepted practice until FDA inspection. How does this happen? Based on my experience across multiple aseptic facilities, the failure mode follows a predictable pattern:

Production scheduling drives intervention timing rather than contamination risk assessment. Stopping a campaign for equipment maintenance creates schedule disruption, yield loss, and capacity constraints. The pressure to maintain campaign continuity overwhelms contamination risk considerations that appear theoretical compared to the immediate, quantifiable production impact.

Risk assessments become compliance artifacts disconnected from operational decision-making. The quality unit conducts a risk assessment, documents that certain interventions pose unacceptable contamination risk, and files the assessment. But when production encounters the situation requiring that intervention, the actual decision-making process references production need, equipment availability, and batch schedules—not the risk assessment that identified the intervention as high-risk.

Interventions become “normalized deviance”—accepted operational practices despite documented risks. After performing a high-risk intervention successfully (meaning without detected contamination) multiple times, it transitions from “high-risk intervention requiring exceptional controls” to “routine intervention” in operational thinking. The fact that adequate controls prevented contamination detection gets inverted into evidence that the intervention isn’t actually high-risk.

Environmental monitoring provides false assurance when contamination goes undetected. If a high-risk intervention occurs and subsequent environmental monitoring shows no contamination, operations interprets this as validation that the intervention is acceptable. But as discussed in the contact plate section, inadequate sampling methodology may fail to detect contamination that actually occurred. The absence of detected contamination becomes “proof” that contamination didn’t occur, reinforcing the normalization of high-risk interventions.

The EU GMP Annex 1 requirements for intervention management represent regulatory recognition of these failure modes. Annex 1 Section 8.16 requires “the list of interventions evaluated via risk analysis” and Section 9.36 requires that aseptic process simulations include “interventions and associated risks”. The framework is explicit: identify interventions, assess their contamination risk, validate that operators can perform them aseptically through media fills, and eliminate interventions that cannot be performed without unacceptable contamination risk.

What does robust intervention risk management look like in practice?

Categorize interventions by contamination risk based on specific, documented criteria. The categorization should consider: surfaces contacted (sterile-to-sterile vs. potentially contaminated), duration of exposure, proximity to open product, operator actions required, first air protection feasibility, and frequency. This creates a risk hierarchy that enables differentiated control strategies rather than treating all interventions equivalently.

Establish clear decision authorities for different intervention risk levels. Routine interventions (low contamination risk, validated through media fills, performed regularly) can proceed under operator judgment following standard procedures. High-risk interventions (those involving occluded surfaces, extended exposure, or proximity to open product) should require quality unit pre-approval including documented risk assessment and enhanced controls specification. Interventions identified as posing unacceptable risk should be prohibited until equipment redesign or process modification eliminates the contamination hazard.

Validate intervention execution through media fills that specifically simulate the intervention’s contamination challenges. Generic media fills demonstrating overall aseptic processing capability don’t validate specific high-risk interventions. If your risk assessment identifies a particular intervention as posing contamination risk, your media fill program must include that intervention, performed by the operators who will execute it, under the conditions (campaign timing, equipment state, environmental conditions) where it will actually occur.

Implement intervention-specific environmental monitoring that targets the contamination pathways identified in risk assessments. If the risk assessment identifies that an intervention may expose product to surfaces not reliably decontaminated, environmental monitoring immediately following that intervention should specifically sample those surfaces and adjacent areas. Trending this intervention-specific monitoring data separately from routine environmental monitoring enables detection of intervention-associated contamination patterns.

Conduct post-intervention investigations when environmental monitoring shows any deviation. The Catalent warning letter describes an environmental monitoring failure whose “most probable root cause” was an atypical intervention involving equipment changes. This temporal association between intervention and contamination should trigger automatic investigation even if environmental monitoring results remain within action levels. The investigation should assess whether intervention protocols require modification or whether the intervention should be eliminated.

The FDA’s remediation demand addresses this gap directly: “review of currently permitted interventions and elimination of high-risk interventions entailing equipment manipulations during production campaigns that expose the ISO 5 environment to surfaces not exposed to a validated decontamination process”. This requirement forces facilities to confront the intervention paradox: if your risk assessment identifies an intervention as high-risk, you cannot simultaneously permit it as routine operational practice. Either modify the intervention to reduce risk, validate enhanced controls that mitigate the risk, or eliminate the intervention entirely.

Media Fill Terminations: When Failures Become Invisible

The Catalent warning letter’s discussion of media fill terminations exposes an investigation failure mode that reveals deeper quality system inadequacies. Since November 2023, Catalent terminated more than five media fill batches representing the filling line. Following two terminations for stoppering issues and extrinsic particle contamination, the facility “failed to open a deviation or an investigation at the time of each failure, as required by your SOPs”.

Read that again. Media fills—the fundamental aseptic processing validation tool, the simulation specifically designed to challenge contamination control—were terminated due to failures, and no deviation was opened, no investigation initiated. The failures simply disappeared from the quality system, becoming invisible until FDA inspection revealed their existence.

The rationalization is predictable: “there was no impact to the SISPQ (Safety, Identity, Strength, Purity, Quality) of the terminated media batches or to any customer batches” because “these media fills were re-executed successfully with passing results”. This reasoning exposes a fundamental misunderstanding of media fill purpose that I’ve encountered with troubling frequency across the industry.

A media fill is not a “test” that you pass or fail with product consequences. It is a simulation—a deliberate challenge to your aseptic processing capability using growth medium instead of product specifically to identify contamination risks without product impact. When a media fill is terminated due to a processing failure, that termination is itself the critical finding. The termination reveals that your process is vulnerable to exactly the failure mode that caused termination: stoppering problems that could occur during commercial filling, extrinsic particles that could contaminate product.

The FDA’s response is appropriately uncompromising: “You do not provide the investigations with a root cause that justifies aborting and re-executing the media fills, nor do you provide the corrective actions taken for each terminated media fill to ensure effective CAPAs were promptly initiated”. The regulatory expectation is clear: media fill terminations require investigation identical in rigor to commercial batch failures. Why did the stoppering issue occur? What equipment, material, or operator factors contributed? How do we prevent recurrence? What commercial batches may have experienced similar failures that went undetected?

The re-execution logic is particularly insidious. By immediately re-running the media fill and achieving passing results, Catalent created the appearance of successful validation while ignoring the process vulnerability revealed by the termination. The successful re-execution proved only that under ideal conditions—now with heightened operator awareness following the initial failure—the process could be executed successfully. It provided no assurance that commercial operations, without that heightened awareness and under the same conditions that caused the initial termination, wouldn’t experience identical failures.

What should media fill termination management look like?

Treat every media fill termination as a critical deviation requiring immediate investigation initiation. The investigation should identify the root cause of the termination, assess whether the failure mode could occur during commercial manufacturing, evaluate whether previous commercial batches may have experienced similar failures, and establish corrective actions that prevent recurrence. This investigation must occur before re-execution, not instead of investigation.

Require quality unit approval before media fill re-execution. The approval should be based on documented investigation findings demonstrating that the termination cause is understood, corrective actions are implemented, and re-execution will validate process capability under conditions that include the corrective actions. Re-execution without investigation approval perpetuates the “keep running until we get a pass” mentality that defeats media fill purpose.

Implement media fill termination trending as a critical quality indicator. A facility terminating “more than five media fill batches” in a period should recognize this as a signal of fundamental process capability problems, not as a series of unrelated events requiring re-execution. Trending should identify common factors: specific operators, equipment states, intervention types, campaign timing.

Ensure deviation tracking systems cannot exclude media fill terminations. The Catalent situation arose partly because “you failed to initiate a deviation record to capture the lack of an investigation for each of the terminated media fills, resulting in an undercounting of the deviations”. Quality metrics that exclude media fill terminations from deviation totals create perverse incentives to avoid formal deviation documentation, rendering media fill findings invisible to quality system oversight.

The broader issue extends beyond media fill terminations to how aseptic processing validation integrates with quality systems. Media fills should function as early warning indicators—detecting aseptic processing vulnerabilities before product impact occurs. But this detection value requires that findings from media fills drive investigations, corrective actions, and process improvements with the same rigor as commercial batch deviations. When media fill failures can be erased through re-execution without investigation, the entire validation framework becomes performative rather than protective.

The Stopper Supplier Qualification Failure: Accepting Contamination at the Source

The stopper contamination issues discussed throughout the warning letter—mammalian hair found in or around stopper regions of vials from nearly 20 batches across multiple products—reveal a supplier qualification and incoming inspection failure that compounds the contamination hazards already discussed. The FDA’s critique focuses on Catalent’s “inappropriate reliance on pre-shipment samples (tailgate samples)” and failure to implement “enhanced or comparative sampling of stoppers from your other suppliers”.

The pre-shipment or “tailgate” sample approach represents a fundamental violation of GMP sampling principles. Under this approach, the stopper supplier—not Catalent—collected samples from lots prior to shipment and sent these samples directly to Catalent for quality testing. Catalent then made accept/reject decisions for incoming stopper lots based on testing of supplier-selected samples that never passed through Catalent’s receiving or storage processes.

Why does this matter? Because representative sampling requires that samples be selected from the material population actually received by the facility, stored under facility conditions, and handled through facility processes. Supplier-selected pre-shipment samples bypass every opportunity to detect contamination introduced during shipping, storage transitions, or handling. They enable a supplier to selectively sample from cleaner portions of production lots while shipping potentially contaminated material in the same lot to the customer.

The FDA guidance on this issue is explicit and has been for decades: samples for quality attribute testing “are to be taken at your facility from containers after receipt to ensure they are representative of the components in question”. This isn’t a new expectation emerging from enhanced regulatory scrutiny—it’s a baseline GMP requirement that Catalent systematically violated through reliance on tailgate samples.

But the tailgate sample issue represents only one element of broader supplier qualification failures. The warning letter notes that “while stoppers from [one supplier] were the primary source of extrinsic particles, they were not the only source of foreign matter.” Yet Catalent implemented “limited, enhanced sampling strategy for one of your suppliers” while failing to “increase sampling oversight” for other suppliers. This selective enhancement—focusing remediation only on the most problematic supplier while ignoring systemic contamination risks across the stopper supply base—predictably failed to resolve ongoing contamination issues.

What should stopper supplier qualification and incoming inspection look like for aseptic filling operations?

Eliminate pre-shipment or tailgate sampling entirely. All quality testing must be conducted on samples taken from received lots, stored in facility conditions, and selected using documented random sampling procedures. If suppliers require pre-shipment testing for their internal quality release, that’s their process requirement—it doesn’t substitute for the purchaser’s independent incoming inspection using facility-sampled material.

Implement risk-based incoming inspection that intensifies sampling when contamination history indicates elevated risk. The warning letter notes that Catalent recognized stoppers as “a possible contributing factor for contamination with mammalian hairs” in July 2024 but didn’t implement enhanced sampling until May 2025—a ten-month delay. The inspection enhancement should be automatic and immediate when contamination events implicate incoming materials. The sampling intensity should remain elevated until trending data demonstrates sustained contamination reduction across multiple lots.

Apply visual inspection with reject criteria specific to the defect types that create product contamination risk. Generic visual inspection looking for general “defects” fails to detect the specific contamination types—embedded hair, extrinsic particles, material fragments—that create sterile product risks. Inspection protocols must specify mammalian hair, fiber contamination, and particulate matter as reject criteria with sensitivity adequate to detect single-particle contamination in sampled stoppers.

Require supplier process changes—not just enhanced sampling—when contamination trends indicate process capability problems. The warning letter acknowledges Catalent “worked with your suppliers to reduce the likelihood of mammalian hair contamination events” but notes that despite these efforts, “you continued to receive complaints from customers who observed mammalian hair contamination in drug products they received from you”. Enhanced sampling detects contamination; it doesn’t prevent it. Suppliers demonstrating persistent contamination require process audits, environmental control improvements, and validated contamination reduction demonstrated through process capability studies—not just promises to improve quality.

Implement finished product visual inspection with heightened sensitivity for products using stoppers from suppliers with contamination history. The FDA notes that Catalent indicated “future batches found during visual inspection of finished drug products would undergo a re-inspection followed by tightened acceptable quality limit to ensure defective units would be removed” but didn’t provide the re-inspection procedure. This two-stage inspection approach—initial inspection followed by re-inspection with enhanced criteria for lots from high-risk suppliers—provides additional contamination detection but must be validated to demonstrate adequate defect removal.

The broader lesson extends beyond stoppers to supplier qualification for any component used in sterile manufacturing. Components introduce contamination risks—microbial bioburden, particulate matter, chemical residues—that cannot be fully mitigated through end-product testing. Supplier qualification must function as a contamination prevention tool, ensuring that materials entering aseptic operations meet microbiological and particulate quality standards appropriate for their role in maintaining sterility. Reliance on tailgate samples, delayed sampling enhancement, and acceptance of persistent supplier contamination all represent failures to recognize suppliers as critical contamination control points requiring rigorous qualification and oversight.

The Systemic Pattern: From Contamination Hazards to Quality System Architecture

Stepping back from individual contamination hazards—occluded surfaces, inadequate sampling, high-risk interventions, media fill terminations, supplier qualification failures—a systemic pattern emerges that connects this warning letter to the broader zemblanity framework I’ve explored in previous posts. These aren’t independent, unrelated deficiencies that coincidentally occurred at the same facility. They represent interconnected architectural failures in how the quality system approaches contamination control.

The pattern reveals itself through three consistent characteristics:

Detection systems optimized for convenience rather than capability. Contact plates instead of swabs for irregular surfaces. Pre-shipment samples instead of facility-based incoming inspection. Generic visual inspection instead of defect-specific contamination screening. Each choice prioritizes operational ease and workflow efficiency over contamination detection sensitivity. The result is a quality system that generates reassuring data—passing environmental monitoring, acceptable incoming inspection results, successful visual inspection—while actual contamination persists undetected.

Risk assessments that identify hazards without preventing their occurrence. Catalent’s risk assessments advised against interventions disturbing potentially occluded surfaces, yet these interventions continued. The facility recognized stoppers as contamination sources in July 2024 but delayed enhanced sampling until May 2025. Media fill terminations revealed aseptic processing vulnerabilities but triggered re-execution rather than investigation. Risk identification became separated from risk mitigation—the assessment process functioned as compliance theatre rather than decision-making input.

Investigation systems that erase failures rather than learn from them. Media fill terminations occurred without deviation initiation. Mammalian hair contamination events were investigated individually without recognizing the trend across 20+ deviations. Root cause investigations concluded “no product impact” based on passing sterility tests rather than addressing the contamination source enabling future events. The investigation framework optimized for batch release justification rather than contamination prevention.

These patterns don’t emerge from incompetent quality professionals or inadequate resource allocation. They emerge from quality system design choices that prioritize production efficiency, workflow continuity, and batch release over contamination detection, investigation rigor, and source elimination. The system delivers what it was designed to deliver: maximum throughput with minimum disruption. It fails to deliver what patients require: contamination control capable of detecting and eliminating sterility risks before product impact.

Recommendations: Building Contamination Hazard Detection Into System Architecture

What does effective contamination hazard management look like at the quality system architecture level? Based on the Catalent failures and broader industry patterns, several principles should guide aseptic operations:

Design decontamination validation around worst-case geometries, not ideal conditions. VHP validation using flat coupons on horizontal surfaces tells you nothing about vapor penetration into the complex geometries, wrapped components, and recessed surfaces actually present in your filling line. Biological indicator placement should target occluded surfaces specifically—if you can’t achieve validated kill on these locations, they’re contamination hazards requiring design modification or alternative decontamination methods.

Select environmental monitoring methods based on detection capability for the surfaces and conditions actually requiring monitoring. Contact plates are adequate for flat, smooth surfaces. They’re inadequate for irregular product-contact surfaces, recessed areas, and complex geometries. Swab sampling takes more time but provides contamination detection capability that contact plates cannot match. The operational convenience sacrifice is trivial compared to the contamination risk from monitoring methods incapable of detecting contamination when it occurs.

Establish intervention risk classification with decision authorities proportional to contamination risk. Routine low-risk interventions validated through media fills can proceed under operator judgment. High-risk interventions—those involving occluded surfaces, extended exposure, or proximity to open product—require quality unit pre-approval with documented enhanced controls. Interventions identified as posing unacceptable risk should be prohibited pending equipment redesign.

Treat media fill terminations as critical deviations requiring investigation before re-execution. The termination reveals process vulnerability—the investigation must identify root cause, assess commercial batch risk, and establish corrective actions before validation continues. Re-execution without investigation perpetuates the failures that caused termination.

Implement supplier qualification with facility-based sampling, contamination-specific inspection criteria, and automatic sampling enhancement when contamination trends emerge. Tailgate samples cannot provide representative material assessment. Visual inspection must target the specific contamination types—mammalian hair, particulate matter, material fragments—that create product risks. Enhanced sampling should be automatic and sustained when contamination history indicates elevated risk.

Build investigation systems that learn from contamination events rather than erasing them through re-execution or “no product impact” conclusions. Contamination events represent failures in contamination control regardless of whether subsequent testing shows product remains within specification. The investigation purpose is preventing recurrence, not justifying release.

The FDA’s comprehensive remediation demands represent what quality system architecture should look like: independent assessment of investigation capability, CAPA effectiveness evaluation, contamination hazard risk assessment covering material flows and equipment placement, detailed remediation with specific improvements, and ongoing management oversight throughout the manufacturing lifecycle.

The Contamination Control Strategy as Living System

The Catalent warning letter’s contamination hazards section serves as a case study in how quality systems can simultaneously maintain surface-level compliance while allowing fundamental contamination control failures to persist. The facility conducted VHP decontamination cycles, performed environmental monitoring, executed media fills, and inspected incoming materials—checking every compliance box. Yet contamination hazards proliferated because these activities optimized for operational convenience and batch release justification rather than contamination detection and source elimination.

The EU GMP Annex 1 Contamination Control Strategy requirement represents regulatory recognition that contamination control cannot be achieved through isolated compliance activities. It requires integrated systems where facility design, decontamination processes, environmental monitoring, intervention protocols, material qualification, and investigation practices function cohesively to detect, investigate, and eliminate contamination sources. The Catalent failures reveal what happens when these elements remain disconnected: decontamination cycles that don’t reach occluded surfaces, monitoring that can’t detect contamination on irregular geometries, interventions that proceed despite identified risks, investigations that erase failures through re-execution

For those of us responsible for contamination control in aseptic manufacturing, the question isn’t whether our facilities face similar vulnerabilities—they do. The question is whether our quality systems are architected to detect these vulnerabilities before regulators discover them. Are your VHP validations addressing actual occluded surfaces or ideal flat coupons? Are you using contact plates because they detect contamination effectively or because they’re operationally convenient? Do your intervention protocols prevent the high-risk activities your risk assessments identify? When media fills terminate, do investigations occur before re-execution?

The Catalent warning letter provides a diagnostic framework for assessing contamination hazard management. Use it. Map your own decontamination validation against the occluded surface criteria. Evaluate your environmental monitoring method selection against detection capability requirements. Review intervention protocols for alignment with risk assessments. Examine media fill termination handling for investigation rigor. Assess supplier qualification for facility-based sampling and contamination-specific inspection.

The contamination hazards are already present in your aseptic operations. The question is whether your quality system architecture can detect them.

A 2025 Retrospective for Investigations of a Dog

If the history of pharmaceutical quality management were written as a geological timeline, 2025 would hopefully mark the end of the Holocene of Compliance—a long, stable epoch where “following the procedure” was sufficient to ensure survival—and the beginning of the Anthropocene of Complexity.

For decades, our industry has operated under a tacit social contract. We agreed to pretend that “compliance” was synonymous with “quality.” We agreed to pretend that a validated method would work forever because we proved it worked once in a controlled protocol three years ago. We agreed to pretend that “zero deviations” meant “perfect performance,” rather than “blind surveillance.” We agreed to pretend that if we wrote enough documents, reality would conform to them.

If I had my wish 2025 would be the year that contract finally dissolved.

Throughout the year—across dozens of posts, technical analyses, and industry critiques on this blog—I have tried to dismantle the comfortable illusions of “Compliance Theater” and show how this theater collides violently with the unforgiving reality of complex systems.

The connecting thread running through every one of these developments is the concept I have returned to obsessively this year: Falsifiable Quality.

This Year in Review is not merely a summary of blog posts. It is an attempt to synthesize the fragmented lessons of 2025 into a coherent argument. The argument is this: A quality system that cannot be proven wrong is a quality system that cannot be trusted.

If our systems—our validation protocols, our risk assessments, our environmental monitoring programs—are designed only to confirm what we hope is true (the “Happy Path”), they are not quality systems at all. They are comfort blankets. And 2025 was the year we finally started pulling the blanket off.

The Philosophy of Doubt

(Reflecting on: The Effectiveness Paradox, Sidney Dekker, and Gerd Gigerenzer)

Before we dissect the technical failures of 2025, let me first establish the philosophical framework that defined this year’s analysis.

In August, I published “The Effectiveness Paradox: Why ‘Nothing Bad Happened’ Doesn’t Prove Your Quality System Works.” It became one of the most discussed posts of the year because it attacked the most sacred metric in our industry: the trend line that stays flat.

We are conditioned to view stability as success. If Environmental Monitoring (EM) data shows zero excursions for six months, we throw a pizza party. If a method validation passes all acceptance criteria on the first try, we commend the development team. If a year goes by with no Critical deviations, we pay out bonuses.

But through the lens of Falsifiable Quality—a concept heavily influenced by the philosophy of Karl Popper, the challenging insights of Deming, and the safety science of Sidney Dekker, whom we discussed in November—these “successes” look suspiciously like failures of inquiry.

The Problem with Unfalsifiable Systems

Karl Popper famously argued that a scientific theory is only valid if it makes predictions that can be tested and proven false. “All swans are white” is a scientific statement because finding one black swan falsifies it. “God is love” is not, because no empirical observation can disprove it.

In 2025, I argued that most Pharmaceutical Quality Systems (PQS) are designed to be unfalsifiable.

The Unfalsifiable Alert Limit: We set alert limits based on historical averages + 3 standard deviations. This ensures that we only react to statistical outliers, effectively blinding us to gradual drift or systemic degradation that remains “within the noise.”
The Unfalsifiable Robustness Study: We design validation protocols that test parameters we already know are safe (e.g., pH +/- 0.1), avoiding the “cliff edges” where the method actually fails. We prove the method works where it works, rather than finding where it breaks.
The Unfalsifiable Risk Assessment: We write FMEAs where the conclusion (“The risk is acceptable”) is decided in advance, and the RPN scores are reverse-engineered to justify it.

This is “Safety Theater,” a term Dekker uses to describe the rituals organizations perform to look safe rather than be safe.

Safety-I vs. Safety-II

In November’s post “Sidney Dekker: The Safety Scientist Who Influences How I Think About Quality,“ I explored Dekker’s distinction between Safety-I (minimizing things that go wrong) and Safety-II (understanding how things usually go right).

Traditional Quality Assurance is obsessed with Safety-I. We count deviations. We count OOS results. We count complaints. When those counts are low, we assume the system is healthy.
But as the LeMaitre Vascular warning letter showed us this year (discussed in Part III), a system can have “zero deviations” simply because it has stopped looking for them. LeMaitre had excellent water data—because they were cleaning the valves before they sampled them. They were measuring their ritual, not their water.

Falsifiable Quality is the bridge to Safety-II. It demands that we treat every batch record not as a compliance artifact, but as a hypothesis test.

Hypothesis: “The contamination control strategy is effective.”
Test: Aggressive monitoring in worst-case locations, not just the “representative” center of the room.
Result: If we find nothing, the hypothesis survives another day. If we find something, we have successfully falsified the hypothesis—which is a good thing because it reveals reality.

The shift from “fearing the deviation” to “seeking the falsification” is a cultural pivot point of 2025.

The Epistemological Crisis in the Lab (Method Validation)

(Reflecting on: USP <1225>, Method Qualification vs. Validation, and Lifecycle Management)

Nowhere was the battle for Falsifiable Quality fought more fiercely in 2025 than in the analytical laboratory.

The proposed revision to USP <1225> Validation of Compendial Procedures (published in Pharmacopeial Forum 51(6)) arrived late in the year, but it serves as the perfect capstone to the arguments I’ve been making since January.

For forty years, analytical validation has been the ultimate exercise in “Validation as an Event.” You develop a method. You write a protocol. You execute the protocol over three days with your best analyst and fresh reagents. You print the report. You bind it. You never look at it again.

This model is unfalsifiable. It assumes that because the method worked in the “Work-as-Imagined” conditions of the validation study, it will work in the “Work-as-Done” reality of routine QC for the next decade.

The Reportable Result: Validating Decisions, Not Signals

The revised USP <1225>—aligned with ICH Q14(Analytical Procedure Development) and USP <1220> (The Lifecycle Approach)—destroys this assumption. It introduces concepts that force falsifiability into the lab.

The most critical of these is the Reportable Result.

Historically, we validated “the instrument” or “the measurement.” We proved that the HPLC could inject the same sample ten times with < 1.0% RSD.

But the Reportable Result is the final value used for decision-making—the value that appears on the Certificate of Analysis. It is the product of a complex chain: Sampling -> Transport -> Storage -> Preparation -> Dilution -> Injection -> Integration -> Calculation -> Averaging.

Validating the injection precision (the end of the chain) tells us nothing about the sampling variability (the beginning of the chain).

By shifting focus to the Reportable Result, USP <1225> forces us to ask: “Does this method generate decisions we can trust?”

The Replication Strategy: Validating “Work-as-Done”

The new guidance insists that validation must mimic the replication strategy of routine testing.
If your SOP says “We report the average of 3 independent preparations,” then your validation must evaluate the precision and accuracy of that average, not of the individual preparations.

This seems subtle, but it is revolutionary. It prevents the common trick of “averaging away” variability during validation to pass the criteria, only to face OOS results in routine production because the routine procedure doesn’t use the same averaging scheme.

It forces the validation study to mirror the messy reality of the “Work-as-Done,” making the validation data a falsifiable predictor of routine performance, rather than a theoretical maximum capability.

Method Qualification vs. Validation: The June Distinction

I wrote “Method Qualification and Validation,” clarifying a distinction that often confuses the industry.

Qualification is the “discovery phase” where we explore the method’s limits. It is inherently falsifiable—we want to find where the method breaks.
Validation has traditionally been the “confirmation phase” where we prove it works.

The danger, as I noted in that post, is when we skip the falsifiable Qualification step and go straight to Validation. We write the protocol based on hope, not data.

USP <1225> essentially argues that Validation must retain the falsifiable spirit of Qualification. It is not a coronation; it is a stress test.

The Death of “Method Transfer” as We Know It

In a Falsifiable Quality system, a method is never “done.” The Analytical Target Profile (ATP)—a concept from ICH Q14 that permeates the new thinking—is a standing hypothesis: “This method measures Potency within +/- 2%.”

Every time we run a system suitability check, every time we run a control standard, we are testing that hypothesis.

If the method starts drifting—even if it still passes broad system suitability limits—a falsifiable system flags the drift. An unfalsifiable system waits for the OOS.

The draft revision of USP <1225> is a call to arms. It asks us to stop treating validation as a “ticket to ride”—a one-time toll we pay to enter GMP compliance—and start treating it as a “ticket to doubt.” Validation gives us permission to use the method, but only as long as the data continues to support the hypothesis of fitness.

The Reality Check (The “Unholy Trinity” of Warning Letters)

Philosophy and guidelines are fine, but in 2025, reality kicked in the door. The regulatory year was defined by three critical warning letters—Sanofi, LeMaitre, and Rechon—that collectively dismantled the industry’s illusions of control.

It began, as these things often do, with a ghost from the past.

Sanofi Framingham: The Pendulum Swings Back

**(Reflecting on: Failure to Investigate Critical Deviations and The Sanofi Warning Letter)**

The year opened with a shock. On January 15, 2025, the FDA issued a warning letter to Sanofi’s Framingham facility—the sister site to the legacy Genzyme Allston landing, whose consent decree defined an entire generation of biotech compliance and of my career.

In my January analysis (“Failure to Investigate Critical Deviations: A Cautionary Tale“), I noted that the FDA’s primary citation was a failure to “thoroughly investigate any unexplained discrepancy.”

This is the cardinal sin of Falsifiable Quality.

An “unexplained discrepancy” is a signal from reality. It is the system telling you, “Your hypothesis about this process is wrong.”

The Falsifiable Response: You dive into the discrepancy. You assume your control strategy missed something. You use Causal Reasoning (the topic of my May post) to find the mechanism of failure.
The Sanofi Response: As the warning letter detailed, they frequently attributed failures to “isolated incidents” or superficial causes without genuine evidence.

This is the “Refusal to Falsify.” By failing to investigate thoroughly, the firm protects the comfortable status quo. They choose to believe the “Happy Path” (the process is robust) over the evidence (the discrepancy).

The Pendulum of Compliance

In my companion post (“Sanofi Warning Letter”), I discussed the “pendulum of compliance.” The Framingham site was supposed to be the fortress of quality, built on the lessons of the Genzyme crisis.

The failure at Sanofi wasn’t a lack of SOPs; it was a lack of curiosity.

The investigators likely had checklists, templates, and timelines (Compliance Theater), but they lacked the mandate—or perhaps the Expertise —to actually solve the problem.

This set the thematic stage for the rest of 2025. Sanofi showed us that “closing the deviation” is not the same as fixing the problem. This insight led directly into my August argument in “The Effectiveness Paradox“: You can close 100% of your deviations on time and still have a manufacturing process that is spinning out of control.

If Sanofi was the failure of investigation (looking back), Rechon and LeMaitre were failures of surveillance (looking forward). Together, they form a complete picture of why unfalsifiable systems fail.

Reflecting on: Rechon Life Science and LeMaitre Vascular

Philosophy and guidelines are fine, but in September, reality kicked in the door.

Two warning letters in 2025—Rechon Life Science (September) and LeMaitre Vascular (August)—provided brutal case studies in what happens when “representative sampling” is treated as a buzzword rather than a statistical requirement.

Rechon Life Science: The Map vs. The Territory

The Rechon Life Science warning letter was a significant regulatory signal of 2025 regarding sterile manufacturing. It wasn’t just a list of observations; it was an indictment of unfalsifiable Contamination Control Strategies (CCS).

We spent 2023 and 2024 writing massive CCS documents to satisfy Annex 1. Hundreds of pages detailing airflows, gowning procedures, and material flows. We felt good about them. We felt “compliant.”

Then the FDA walked into Rechon and essentially asked: “If your CCS is so good, why does your smoke study show turbulence over the open vials?”

The warning letter highlighted a disconnect I’ve called “The Map vs. The Territory.”

The Map: The CCS document says the airflow is unidirectional and protects the product.
The Territory: The smoke study video shows air eddying backward from the operator to the sterile core.

In an unfalsifiable system, we ignore the smoke study (or film it from a flattering angle) because it contradicts the CCS. We prioritize the documentation (the claim) over the observation (the evidence).

In a falsifiable system, the smoke study is the test. If the smoke shows turbulence, the CCS is falsified. We don’t defend the CCS; we rewrite it. We redesign the line.

The FDA’s critique of Rechon’s “dynamic airflow visualization” was devastating because it showed that Rechon was using the smoke study as a marketing video, not a diagnostic tool. They filmed “representative” operations that were carefully choreographed to look clean, rather than the messy reality of interventions.

LeMaitre Vascular: The Sin of “Aspirational Data”

If Rechon was about air, LeMaitre Vascular (analyzed in my August post “When Water Systems Fail“) was about water. And it contained an even more egregious sin against falsifiability.

The FDA observed that LeMaitre’s water sampling procedures required cleaning and purging the sample valves before taking the sample.

Let’s pause and consider the epistemology of this.

The Goal: To measure the quality of the water used in manufacturing.
The Reality: Manufacturing operators do not purge and sanitize the valve for 10 minutes before filling the tank. They open the valve and use the water.
The Sample: By sanitizing the valve before sampling, LeMaitre was measuring the quality of the sampling process, not the quality of the water system.

I call this “Aspirational Data.” It is data that reflects the system as we wish it existed, not as it actually exists. It is the ultimate unfalsifiable metric. You can never find biofilm in a valve if you scrub the valve with alcohol before you open it.

The FDA’s warning letter was clear: “Sampling… must include any pathway that the water travels to reach the process.”

LeMaitre also performed an unauthorized “Sterilant Switcheroo,” changing their sanitization agent without change control or biocompatibility assessment. This is the hallmark of an unfalsifiable culture: making changes based on convenience, assuming they are safe, and never designing the study to check if that assumption is wrong.

The “Representative” Trap

Both warning letters pivot on the misuse of the word “representative.”

Firms love to claim their EM sampling locations are “representative.” But representative of what? Usually, they are representative of the average condition of the room—the clean, empty spaces where nothing happens.

But contamination is not an “average” event. It is a specific, localized failure. A falsifiable EM program places probes in the “worst-case” locations—near the door, near the operator’s hands, near the crimping station. It tries to find contamination. It tries to falsify the claim that the zone is sterile, asceptic or bioburden reducing.

When Rechon and LeMaitre failed to justify their sampling locations, they were guilty of designing an unfalsifiable experiment. They placed the “microscope” where they knew they wouldn’t find germs.

2025 taught us that regulators are no longer impressed by the thickness of the CCS binder. They are looking for the logic of control. They are testing your hypothesis. And if you haven’t tested it yourself, you will fail.

The Investigation as Evidence

(Reflecting on: The Golden Start to a Deviation Investigation, Causal Reasoning, Take-the-Best Heuristics, and The Catalent Case)

If Rechon, LeMaitre, and Sanofi teach us anything, it is that the quality system’s ability to discover failure is more important than its ability to prevent failure.

A perfect manufacturing process that no one is looking at is indistinguishable from a collapsing process disguised by poor surveillance. But a mediocre process that is rigorously investigated, understood, and continuously improved is a path toward genuine control.

The investigation itself—how we respond to a deviation, how we reason about causation, how we design corrective actions—is where falsifiable quality either succeeds or fails.

The Golden Day: When Theory Meets Work-as-Done

In April, I published “The Golden Start to a Deviation Investigation,” which made a deceptively simple argument: The first 24 hours after a deviation is discovered are where your quality system either commits to discovering truth or retreats into theater.

This argument sits at the heart of falsifiable quality.

When a deviation occurs, you have a narrow window—what I call the “Golden Day”—where evidence is fresh, memories are intact, and the actual conditions that produced the failure still exist. If you waste this window with vague problem statements and abstract discussions, you permanently lose the ability to test causal hypotheses later.

The post outlined a structured protocol:

First, crystallize the problem. Not “potency was low”—but “Lot X234, potency measured at 87% on January 15th at 14:32, three hours after completion of blending in Vessel C-2.” Precision matters because only specific, bounded statements can be falsified. A vague problem statement can always be “explained away.”

Second, go to the Gemba. This is the antidote to “work-as-imagined” investigation. The SOP says the temperature controller should maintain 37°C +/- 2°C. But the Gemba walk reveals that the probe is positioned six inches from the heating element, the data logger is in a recessed pocket where humidity accumulates, and the operator checks it every four hours despite a requirement to check hourly. These are the facts that predict whether the deviation will recur.

Third, interview with cognitive discipline. Most investigations fail not because investigators lack information, but because they extract information poorly. Cognitive interviewing—developed by the FBI and the National Transportation Safety Board—uses mental reinstatement, multiple perspectives, and sequential reordering to access accurate recall rather than confabulated narrative. The investigator asks the operator to walk through the event in different orders, from different viewpoints, each time triggering different memory pathways. This is not “soft” technique; it is a mechanism for generating falsifiable evidence.

The Golden Day post makes it clear: You do not investigate deviations to document compliance. You investigate deviations to gather evidence about whether your understanding of the process is correct.

Causal Reasoning: Moving Beyond “What Was Missing”

Most investigation tools fail not because they are flawed, but because they are applied with the wrong mindset. In my May post “Causal Reasoning: A Transformative Approach to Root Cause Analysis,” I argued that pharmaceutical investigations are often trapped in “negative reasoning.”

Negative reasoning asks: “What barrier was missing? What should have been done but wasn’t?” This mindset leads to unfalsifiable conclusions like “Procedure not followed” or “Training was inadequate.” These are dead ends because they describe the absence of an ideal, not the presence of a cause.

Causal reasoning flips the script. It asks: “What was present in the system that made the observed outcome inevitable?”

Instead of settling for “human error,” causal reasoning demands we ask: What environmental cues made the action sensible to the operator at that moment? Were the instructions ambiguous? Did competing priorities make compliance impossible? Was the process design fragile?

This shift transforms the investigation from a compliance exercise into a scientific inquiry.

Consider the LeMaitre example:

Negative Reasoning: “Why didn’t they sample the true condition?” Answer: “Because they didn’t follow the intent of the sampling plan.”
Causal Reasoning: “What made the pre-cleaning practice sensible to them?” Answer: “They believed it ensured sample validity by removing valve residue.”

By understanding the why, we identify a knowledge gap that can be tested and corrected, rather than a negligence gap that can only be punished.

In September, “Take-the-Best Heuristic for Causal Investigation” provided a practical framework for this. Instead of listing every conceivable cause—a process that often leads to paralysis—the “Take-the-Best” heuristic directs investigators to focus on the most information-rich discriminators. These are the factors that, if different, would have prevented the deviation. This approach focuses resources where they matter most, turning the investigation into a targeted search for truth.

CAPA: Predictions, Not Promises

The Sanofi warning letter—analyzed in January—showed the destination of unfalsifiable investigation: CAPAs that exist mainly as paperwork.

Sanofi had investigation reports. They had “corrective actions.” But the FDA noted that deviations recurred in similar patterns, suggesting that the investigation had identified symptoms, not mechanisms, and that the “corrective” action had not actually addressed causation.

This is the sin of treating CAPA as a promise rather than a hypothesis.

A falsifiable CAPA is structured as an explicit prediction: “If we implement X change, then Y undesirable outcome will not recur under conditions Z.”

This can be tested. If it fails the test, the CAPA itself becomes evidence—not of failure, but of incomplete causal understanding. Which is valuable.

In the Rechon analysis, this showed up concretely: The FDA’s real criticism was not just that contamination was found; it was that Rechon’s Contamination Control Strategy had no mechanism to falsify itself. If the CCS said “unidirectional airflow protects the product,” and smoke studies showed bidirectional eddies, the CCS had been falsified. But Rechon treated the falsification as an anomaly to be explained away, rather than evidence that the CCS hypothesis was wrong.

A falsifiable organization would say: “Our CCS predicted that Grade A in an isolator with this airflow pattern would remain sterile. The smoke study proves that prediction wrong. Therefore, the CCS is false. We redesign.”

Instead, they filmed from a different angle and said the aerodynamics were “acceptable.”

Knowledge Integration: When Deviations Become the Curriculum

The final piece of falsifiable investigation is what I call “knowledge integration.” A single deviation is a data point. But across the organization, deviations should form a curriculum about how systems actually fail.

Sanofi’s failure was not that they investigated each deviation badly (though they did). It was that they investigated them in isolation. Each deviation closed on its own. Each CAPA addressed its own batch. There was no organizational learning—no mechanism for a pattern of similar deviations to trigger a hypothesis that the control strategy itself was fundamentally flawed.

This is where the Catalent case study, analyzed in September’s “When 483s Reveal Zemblanity,” becomes instructive. Zemblanity is the opposite of serendipity: the seemingly random recurrence of the same failure through different paths. Catalent’s 483 observations were not isolated mistakes; they formed a pattern that revealed a systemic assumption (about equipment capability, about environmental control, about material consistency) that was false across multiple products and locations.

A falsifiable quality system catches zemblanity early by:

Treating each deviation as a test of organizational hypotheses, not as an isolated incident.
Trending deviation patterns to detect when the same causal mechanism is producing failures across different products, equipment, or operators.
Revising control strategies when patterns falsify the original assumptions, rather than tightening parameters at the margins.

The Digital Hallucination (CSA, AI, and the Expertise Crisis)

(Reflecting on: CSA: The Emperor’s New Clothes, Annex 11, and The Expertise Crisis)

While we battled microbes in the cleanroom, a different battle was raging in the server room. 2025 was the year the industry tried to “modernize” validation through Computer Software Assurance (CSA) and AI, and in many ways, it was the year we tried to automate our way out of thinking.

CSA: The Emperor’s New Validation Clothes

In September, I published “Computer System Assurance: The Emperor’s New Validation Clothes,” a critique of the the contortions being made around the FDA’s guidance. The narrative sold by consultants for years was that traditional Computer System Validation (CSV) was “broken”—too much documentation, too much testing—and that CSA was a revolutionary new paradigm of “critical thinking.”

My analysis showed that this narrative is historically illiterate.

The principles of CSA—risk-based testing, leveraging vendor audits, focusing on intended use—are not new. They are the core principles of GAMP5 and have been applied for decades now.

The industry didn’t need a new guidance to tell us to use critical thinking; we had simply chosen not to use the critical thinking tools we already had. We had chosen to apply “one-size-fits-all” templates because they were safe (unfalsifiable).

The CSA guidance is effectively the FDA saying: “Please read the GAMP5 guide you claimed to be following for the last 15 years.”

The danger of the “CSA Revolution” narrative is that it encourages a swing to the opposite extreme: “Unscripted Testing” that becomes “No Testing.”

In a falsifiable system, “unscripted testing” is highly rigorous—it is an expert trying to break the software (“Ad Hoc testing”). But in an unfalsifiable system, “unscripted testing” becomes “I clicked around for 10 minutes and it looked fine.”

The Expertise Crisis: AI and the Death of the Apprentice

This leads directly to the Expertise Crisis. In September, I wrote “The Expertise Crisis: Why AI’s War on Entry-Level Jobs Threatens Quality’s Future.” This was perhaps the most personal topic I covered this year, because it touches on the very survival of our profession.

We are rushing to integrate Artificial Intelligence (AI) into quality systems. We have AI writing deviations, AI drafting SOPs, AI summarizing regulatory changes. The efficiency gains are undeniable. But the cost is hidden, and it is epistemological.

Falsifiability requires expertise.
To falsify a claim—to look at a draft investigation report and say, “No, that conclusion doesn’t follow from the data”—you need deep, intuitive knowledge of the process. You need to know what a “normal” pH curve looks like so you can spot the “abnormal” one that the AI smoothed over.

Where does that intuition come from? It comes from the “grunt work.” It comes from years of reviewing batch records, years of interviewing operators, years of struggling to write a root cause analysis statement.

The Expertise Crisis is this: If we give all the entry-level work to AI, where will the next generation of Quality Leaders come from?

The Junior Associate doesn’t review the raw data; the AI summarizes it.
The Junior Associate doesn’t write the deviation; the AI generates the text.
Therefore, the Junior Associate never builds the mental models necessary to critique the AI.

The Loop of Unfalsifiable Hallucination

We are creating a closed loop of unfalsifiability.

The AI generates a plausible-sounding investigation report.
The human reviewer (who has been “de-skilled” by years of AI reliance) lacks the deep expertise to spot the subtle logical flaw or the missing data point.
The report is approved.
The “hallucination” becomes the official record.

In a falsifiable quality system, the human must remain the adversary of the algorithm. The human’s job is to try to break the AI’s logic, to check the citations, to verify the raw data.
But in 2025, we saw the beginnings of a “Compliance Autopilot”—a desire to let the machine handle the “boring stuff.”

My warning in September remains urgent: Efficiency without expertise is just accelerated incompetence. If we lose the ability to falsify our own tools, we are no longer quality professionals; we are just passengers in a car driven by a statistical model that doesn’t know what “truth” is.

My post “The Missing Middle in GMP Decision Making: How Annex 22 Redefines Human-Machine Collaboration in Pharmaceutical Quality Assurance” goes a lot deeper here.

Annex 11 and Data Governance

In August, I analyzed the draft Annex 11 (Computerised Systems) in the post “Data Governance Systems: A Fundamental Shift.”

The Europeans are ahead of the FDA here. While the FDA talks about “Assurance” (testing less), the EU is talking about “Governance” (controlling more). The new Annex 11 makes it clear: You cannot validate a system if you do not control the data lifecycle. Validation is not a test script; it is a state of control.

This aligns perfectly with USP <1225> and <1220>. Whether it’s a chromatograph or an ERP system, the requirement is the same: Prove that the data is trustworthy, not just that the software is installed.

The Process as a Hypothesis (CPV & Cleaning)

(Reflecting on: Continuous Process Verification and Hypothesis Formation)

The final frontier of validation we explored in 2025 was the manufacturing process itself.

CPV: Continuous Falsification

In March, I published “Continuous Process Verification (CPV) Methodology and Tool Selection.”
CPV is the ultimate expression of Falsifiable Quality in manufacturing.

Traditional Validation (3 Batches): “We made 3 good batches, therefore the process is perfect forever.” (Unfalsifiable extrapolation).
CPV: “We made 3 good batches, so we have a license to manufacture, but we will statistically monitor every subsequent batch to detect drift.” (Continuous hypothesis testing).

The challenge with CPV, as discussed in the post, is that it requires statistical literacy. You cannot implement CPV if your quality unit doesn’t understand the difference between Cpk and Ppk, or between control limits and specification limits.

This circles back to the Expertise Crisis. We are implementing complex statistical tools (CPV software) at the exact moment we are de-skilling the workforce. We risk creating a “CPV Dashboard” that turns red, but no one knows why or what to do about it.

Cleaning Validation: The Science of Residue

In August, I tried to apply falsifiability to one of the most stubborn areas of dogma: Cleaning Validation.

In “Building Decision-Making with Structured Hypothesis Formation,“ I argued that cleaning validation should not be about “proving it’s clean.” It should be about “understanding why it gets dirty.”

Traditional Approach: Swab 10 spots. If they pass, we are good.
Hypothesis Approach: “We hypothesize that the gasket on the bottom valve is the hardest to clean. We predict that if we reduce rinse time by 1 minute, that gasket will fail.”

By testing the boundaries—by trying to make the cleaning fail—we understand the Design Space of the cleaning process.

We discussed the “Visual Inspection” paradox in cleaning: If you can see the residue, it failed. But if you can’t see it, does it pass?

Only if you have scientifically determined the Visible Residue Limit (VRL). Using “visually clean” without a validated VRL is—you guessed it—unfalsifiable.

To: Jeremiah Genest
From: Perplexity Research
Subject: Draft Content – Single-Use Systems & E&L Section

Here is a section on Single-Use Systems (SUS) and Extractables & Leachables (E&L).

I have positioned this piece to bridge the gap between “Part III: The Reality Check” (Contamination/Water) and “Part V: The Process as a Hypothesis” (Cleaning Validation).

The argument here is that by switching from Stainless Steel to Single-Use, we traded a visible risk (cleaning residue) for an invisible one (chemical migration), and that our current approach to E&L is often just “Paper Safety”—relying on vendor data that doesn’t reflect the “Work-as-Done” reality of our specific process conditions.

The Plastic Paradox (Single-Use Systems and the E&L Mirage)

If the Rechon and LeMaitre warning letters were about the failure to control biological contaminants we can find, the industry’s struggle with Single-Use Systems (SUS) in 2025 was about the chemical contaminants we choose not to find.

We have spent the last decade aggressively swapping stainless steel for plastic. The value proposition was irresistible: Eliminate cleaning validation, eliminate cross-contamination, increase flexibility. We traded the “devil we know” (cleaning residue) for the “devil we don’t” (Extractables and Leachables).

But in 2025, with the enforcement reality of USP <665> (Plastic Components and Systems) settling in, we had to confront the uncomfortable truth: Most E&L risk assessments are unfalsifiable.

The Vendor Data Trap

The standard industry approach to E&L is the ultimate form of “Compliance Theater.”

We buy a single-use bag.
We request the vendor’s regulatory support package (the “Map”).
We see that the vendor extracted the film with aggressive solvents (ethanol, hexane) for 7 days.
We conclude: “Our process uses water for 24 hours; therefore, we are safe.”

This logic is epistemologically bankrupt. It assumes that the Vendor’s Model (aggressive solvents/short time) maps perfectly to the User’s Reality (complex buffers/long duration/specific surfactants).

It ignores the fact that plastics are dynamic systems. Polymers age. Gamma irradiation initiates free radical cascades that evolve over months. A bag manufactured in January might have a different leachable profile than a bag manufactured in June, especially if the resin supplier made a “minor” change that didn’t trigger a notification.

By relying solely on the vendor’s static validation package, we are choosing not to falsify our safety hypothesis. We are effectively saying, “If the vendor says it’s clean, we will not look for dirt.”

USP <665>: A Baseline, Not a Ceiling

The full adoption of USP <665> was supposed to bring standardization. And it has—it provides a standard set of extraction conditions. But standards can become ceilings.

In 2025, I observed a troubling trend of “Compliance by Citation.” Firms are citing USP <665> compliance as proof of absence of risk, stopping the inquiry there.

A Falsifiable E&L Strategy goes further. It asks:

“What if the vendor data is irrelevant to my specific surfactant?”
“What if the gamma irradiation dose varied?”
“What if the interaction between the tubing and the connector creates a new species?”

The Invisible Process Aid

We must stop viewing Single-Use Systems as inert piping. They are active process components. They are chemically reactive vessels that participate in our reaction kinetics.

When we treat them as inert, we are engaging in the same “Aspirational Thinking” that LeMaitre used on their water valves. We are modeling the system we want (pure, inert plastic), not the system we have (a complex soup of antioxidants, slip agents, and degradants).

The lesson of 2025 is that Material Qualification cannot be a paper exercise. If you haven’t done targeted simulation studies that mimic your actual “Work-as-Done” conditions, you haven’t validated the system. You’ve just filed the receipt.

The Mandate for 2026

As we look toward 2026, the path is clear. We cannot go back to the comfortable fiction of the pre-2025 era.

The regulatory environment (Annex 1, ICH Q14, USP <1225>, Annex 11) is explicitly demanding evidence of control, not just evidence of compliance. The technological environment (AI) is demanding that we sharpen our human expertise to avoid becoming obsolete. The physical environment (contamination, supply chain complexity) is demanding systems that are robust, not just rigid.

The mandate for the coming year is to build Falsifiable Quality Systems.

What does that look like practically?

In the Lab: Implement USP <1225> logic now. Don’t wait for the official date. Validate your reportable results. Add “challenge tests” to your routine monitoring.
In the Plant: Redesign your Environmental Monitoring to hunt for contamination, not to avoid it. If you have a “perfect” record in a Grade C area, move the plates until you find the dirt.
In the Office: Treat every investigation as a chance to falsify the control strategy. If a deviation occurs that the control strategy said was impossible, update the control strategy.
In the Culture: Reward the messenger. The person who finds the crack in the system is not a troublemaker; they are the most valuable asset you have. They just falsified a false sense of security.
In Design: Embrace the “Elegant Quality System“ (discussed in May). Complexity is the enemy of falsifiability. Complex systems hide failures; simple, elegant systems reveal them.

2025 was the year we stopped pretending. 2026 must be the year we start building. We must build systems that are honest enough to fail, so that we can build processes that are robust enough to endure.

Thank you for reading, challenging, and thinking with me this year. The investigation continues.

Recent Podcast Appearance: Risk Revolution

I’m excited to share that I recently had the opportunity to appear on the Risk Revolution podcast, joining host Valerie Mulholland for what turned out to be a provocative and deeply engaging conversation about the future of pharmaceutical quality management.

The episode, titled “Quality Theatre to Quality Science – Jeremiah Genest’s Playbook,” aired on September 28, 2025, and dives into one of my core arguments: that quality systems should be designed to fail predictably so we can learn purposefully. This isn’t about celebrating failure—it’s about building systems intelligent enough to fail in ways that generate learning rather than hiding in the shadows until catastrophic breakdown occurs.

Why This Conversation Matters

Valerie and I spent over an hour exploring what I call “intelligent failure”—a concept that challenges the feel-good metrics that dominate our industry dashboards. You know the ones I’m talking about: those green lights celebrating zero deviations that make everyone feel accomplished while potentially masking the unknowns lurking beneath the surface. As I argued in the episode, these metrics can hide systemic problems rather than prove actual control.

This discussion connects directly to themes I’ve been developing here on Investigations of a Dog, particularly my thoughts on the effectiveness paradox and the dangerous comfort of “nothing bad happened” thinking. The podcast gave me a chance to explore how zemblanity—the patterned recurrence of unfortunate events that we should have anticipated—manifests in quality systems that prioritize the appearance of control over genuine understanding.

The Perfect Platform for These Ideas

Risk Revolution proved to be the ideal venue for this conversation. Valerie brings over 25 years of hands-on experience across biopharmaceutical, pharmaceutical, medical device, and blood transfusion industries, but what sets her apart is her unique combination of practical expertise and cutting-edge research.

The podcast’s monthly format allows for the kind of deep, nuanced discussions that advance risk management maturity rather than recycling conference presentations. When I wrote about Valerie’s writing on the GI Joe Bias, I noted how her emphasis on systematic interventions rather than individual awareness represents exactly the kind of sophisticated thinking our industry needs. This podcast appearance let us explore these concepts in real-time conversation.

What made the discussion particularly engaging was Valerie’s ability to challenge my thinking while building on it. Her research-backed insights into cognitive bias management created a perfect complement to my practical experience with system failures and investigation patterns. We explored how quality professionals—precisely because of our expertise—become vulnerable to specific blind spots that systematic design can address.

Looking Forward

This Risk Revolution appearance represents more than just a podcast interview—it’s part of a broader conversation about advancing pharmaceutical quality management beyond surface-level compliance toward genuine excellence. The episode includes references to my blog work, the Deming philosophy, and upcoming industry conferences where these ideas will continue to evolve.

If you’re interested in how quality systems can be designed for intelligent learning rather than elegant hiding, this conversation offers both provocative challenges and practical frameworks. Fair warning: you might never look at a green dashboard the same way again.

The episode is available now, and I’d love to hear your thoughts on how we might move from quality theatre toward quality science in your own organization.

Risk Blindness: The Invisible Threat

Risk blindness is an insidious loss of organizational perception—the gradual erosion of a company’s ability to recognize, interpret, and respond to threats that undermine product safety, regulatory compliance, and ultimately, patient trust. It is not merely ignorance or oversight; rather, risk blindness manifests as the cumulative inability to see threats, often resulting from process shortcuts, technology overreliance, and the undervaluing of hands-on learning.

Unlike risk aversion or neglect, which involves conscious choices, risk blindness is an unconscious deficiency. It often stems from structural changes like the automation of foundational jobs, fragmented risk ownership, unchallenged assumptions, and excessive faith in documentation or AI-generated reports. At its core, risk blindness breeds a false sense of security and efficiency while creating unseen vulnerabilities.

Pattern Recognition and Risk Blindness: The Cognitive Foundation of Quality Excellence

The Neural Architecture of Risk Detection

Pattern recognition lies at the heart of effective risk management in quality systems. It represents the sophisticated cognitive process by which experienced professionals unconsciously scan operational environments, data trends, and behavioral cues to detect emerging threats before they manifest as full-scale quality events. This capability distinguishes expert practitioners from novices and forms the foundation of what we might call “risk literacy” within quality organizations.

The development of pattern recognition in pharmaceutical quality follows predictable stages. At the most basic level (Level 1 Situational Awareness), professionals learn to perceive individual elements—deviation rates, environmental monitoring trends, supplier performance metrics. However, true expertise emerges at Level 2 (Comprehension), where practitioners begin to understand the relationships between these elements, and Level 3 (Projection), where they can anticipate future system states based on current patterns.

Research in clinical environments demonstrates that expert pattern recognition relies on matching current situational elements with previously stored patterns and knowledge, creating rapid, often unconscious assessments of risk significance. In pharmaceutical quality, this translates to the seasoned professional who notices that “something feels off” about a batch record, even when all individual data points appear within specification, or the environmental monitoring specialist who recognizes subtle trends that precede contamination events.

The Apprenticeship Dividend: Building Pattern Recognition Through Experience

The development of sophisticated pattern recognition capabilities requires what we’ve previously termed the “apprenticeship dividend”—the cumulative learning that occurs through repeated exposure to routine operations, deviations, and corrective actions. This learning cannot be accelerated through technology or condensed into senior-level training programs; it must be built through sustained practice and mentored reflection.

The Stages of Pattern Recognition Development:

Foundation Stage (Years 1-2): New professionals learn to identify individual risk elements—understanding what constitutes a deviation, recognizing out-of-specification results, and following investigation procedures. Their pattern recognition is limited to explicit, documented criteria.

Integration Stage (Years 3-5): Practitioners begin to see relationships between different quality elements. They notice when environmental monitoring trends correlate with equipment issues, or when supplier performance changes precede raw material problems. This represents the emergence of tacit knowledge—insights that are difficult to articulate but guide decision-making.

Mastery Stage (Years 5+): Expert practitioners develop what researchers call “intuitive expertise”—the ability to rapidly assess complex situations and identify subtle risk patterns that others miss. They can sense when a investigation is heading in the wrong direction, recognize when supplier responses are evasive, or detect process drift before it appears in formal metrics.

Tacit Knowledge: The Uncodifiable Foundation of Risk Assessment

Perhaps the most critical aspect of pattern recognition in pharmaceutical quality is the role of tacit knowledge—the experiential wisdom that cannot be fully documented or transmitted through formal training systems. Tacit knowledge encompasses the subtle cues, contextual understanding, and intuitive insights that experienced professionals develop through years of hands-on practice.

In pharmaceutical quality systems, tacit knowledge manifests in numerous ways:

Knowing which equipment is likely to fail after cleaning cycles, based on subtle operational cues rather than formal maintenance schedules
Recognizing when supplier audit responses are technically correct but practically inadequate
Sensing when investigation teams are reaching premature closure without adequate root cause analysis
Detecting process drift through operator reports and informal observations before it appears in formal monitoring data

This tacit knowledge cannot be captured in standard operating procedures or electronic systems. It exists in the experienced professional’s ability to read “between the lines” of formal data, to notice what’s missing from reports, and to sense when organizational pressures are affecting the quality of risk assessments.

The GI Joe Fallacy: The Dangers of “Knowing is Half the Battle”

A persistent—and dangerous—belief in quality organizations is the idea that simply knowing about risks, standards, or biases will prevent us from falling prey to them. This is known as the GI Joe fallacy—the misguided notion that awareness is sufficient to overcome cognitive biases or drive behavioral change.

What is the GI Joe Fallacy?

Inspired by the classic 1980s G.I. Joe cartoons, which ended each episode with “Now you know. And knowing is half the battle,” the GI Joe fallacy describes the disconnect between knowledge and action. Cognitive science consistently shows that knowing about biases or desired actions does not ensure that individuals or organizations will behave accordingly.

Even the founder of bias research, Daniel Kahneman, has noted that reading about biases doesn’t fundamentally change our tendency to commit them. Organizations often believe that training, SOPs, or system prompts are enough to inoculate staff against error. In reality, knowledge is only a small part of the battle; much larger are the forces of habit, culture, distraction, and deeply rooted heuristics.

GI Joe Fallacy in Quality Risk Management

In pharmaceutical quality risk management, the GI Joe fallacy can have severe consequences. Teams may know the details of risk matrices, deviation procedures, and regulatory requirements, yet repeatedly fail to act with vigilance or critical scrutiny in real situations. Loss aversion, confirmation bias, and overconfidence persist even for those trained in their dangers.

For example, base rate neglect—a bias where salient event data distracts from underlying probabilities—can influence decisions even when staff know better intellectually. This manifests in investigators overreacting to recent dramatic events while ignoring stable process indicators. Knowing about risk frameworks isn’t enough; structures and culture must be designed specifically to challenge these biases in practice, not simply in theory.

Structural Roots of Risk Blindness

The False Economy of Automation and Overconfidence

Risk blindness often arises from a perceived efficiency gained through process automation or the curtailment of on-the-ground learning. When organizations substitute active engagement for passive oversight, staff lose critical exposure to routine deviations and process variables.

Senior staff who only approve system-generated risk assessments lack daily operational familiarity, making them susceptible to unseen vulnerabilities. Real risk assessment requires repeated, active interaction with process data—not just a review of output.

Fragmented Ownership and Deficient Learning Culture

Risk ownership must be robust and proximal. When roles are fragmented—where the “system” manages risk and people become mere approvers—vital warnings can be overlooked. A compliance-oriented learning culture that believes training or SOPs are enough to guard against operational threats falls deeper into the GI Joe fallacy: knowledge is mistaken for vigilance.

Instead, organizations need feedback loops, reflection, and opportunities to surface doubts and uncertainties. Training must be practical and interactive, not limited to information transfer.

Zemblanity: The Shadow of Risk Blindness

Zemblanity is the antithesis of serendipity in the context of pharmaceutical quality—it describes the persistent tendency for organizations to encounter negative, foreseeable outcomes when risk signals are repeatedly ignored, misunderstood, or left unacted upon.

When examining risk blindness, zemblanity stands as the practical outcome: a quality system that, rather than stumbling upon unexpected improvements or positive turns, instead seems trapped in cycles of self-created adversity. Unlike random bad luck, zemblanity results from avoidable and often visible warning signs—deviations that are rationalized, oversight meetings that miss the point, and cognitive biases like the GI Joe fallacy that lull teams into a false sense of mastery

Real-World Manifestations

Case: The Disappearing Deviation

Digital batch records reduced documentation errors and deviation reports, creating an illusion of process control. But when technology transfer led to out-of-spec events, the lack of manually trained eyes meant no one was poised to detect subtle process anomalies. Staff “knew” the process in theory—yet risk blindness set in because the signals were no longer being actively, expertly interpreted. Knowledge alone was not enough.

Case: Supplier Audit Blindness

Virtual audits relying solely on documentation missed chronic training issues that onsite teams would likely have noticed. The belief that checklist knowledge and documentation sufficed prevented the team from recognizing deeper underlying risks. Here, the GI Joe fallacy made the team believe their expertise was shield enough, when in reality, behavioral engagement and observation were necessary.

Counteracting Risk Blindness: Beyond Knowing to Acting

Effective pharmaceutical quality systems must intentionally cultivate and maintain pattern recognition capabilities across their workforce. This requires structured approaches that go beyond traditional training and incorporate the principles of expertise development:

Structured Exposure Programs: New professionals need systematic exposure to diverse risk scenarios—not just successful cases, but also investigations that went wrong, supplier audits that missed problems, and process changes that had unexpected consequences. This exposure must be guided by experienced mentors who can help identify and interpret relevant patterns.

Cross-Functional Pattern Sharing: Different functional areas—manufacturing, quality control, regulatory affairs, supplier management—develop specialized pattern recognition capabilities. Organizations need systematic mechanisms for sharing these patterns across functions, ensuring that insights from one area can inform risk assessment in others.

Cognitive Diversity in Assessment Teams: Research demonstrates that diverse teams are better at pattern recognition than homogeneous groups, as different perspectives help identify patterns that might be missed by individuals with similar backgrounds and experience. Quality organizations should intentionally structure assessment teams to maximize cognitive diversity.

Systematic Challenge Processes: Pattern recognition can become biased or incomplete over time. Organizations need systematic processes for challenging established patterns—regular “red team” exercises, external perspectives, and structured devil’s advocate processes that test whether recognized patterns remain valid.

Reflective Practice Integration: Pattern recognition improves through reflection on both successes and failures. Organizations should create systematic opportunities for professionals to analyze their pattern recognition decisions, understand when their assessments were accurate or inaccurate, and refine their capabilities accordingly.

Using AI as a Learning Accelerator

AI and automation should support, not replace, human risk assessment. Tools can help new professionals identify patterns in data, but must be employed as aids to learning—not as substitutes for judgment or action.

Assess organizational risk literacy not by the presence of knowledge, but by the frequency of active, critical engagement with real risks. Use self-assessment questions such as:

Do deviation investigations include frontline voices, not just system reviewers?
Are new staff exposed to real processes and deviations, not just theoretical scenarios?
Are risk reviews structured to challenge assumptions, not merely confirm them?
Is there evidence that knowledge is regularly translated into action?

Why Preventing Risk Blindness Matters

Regulators evaluate quality maturity not simply by compliance, but by demonstrable capability to anticipate and mitigate risks. AI and digital transformation are intensifying the risk of the GI Joe fallacy by tempting organizations to substitute data and technology for judgment and action.

As experienced professionals retire, the gap between knowing and doing risks widening. Only organizations invested in hands-on learning, mentorship, and behavioral feedback will sustain true resilience.

Choosing Sight

Risk blindness is perpetuated by the dangerous notion that knowing is enough. The GI Joe fallacy teaches that organizational memory, vigilance, and capability require much more than knowledge—they demand deliberate structures, engaged cultures, and repeated practice that link theory to action.

Quality leaders must invest in real development, relentless engagement, and humility about the limits of their own knowledge. Only then will risk blindness be cured, and resilience secured.

When 483s Reveal Zemblanity: The Catalent Investigation – A Case Study in Systemic Quality Failure

The Catalent Indiana 483 form from July 2025 reads like a textbook example of my newest word, zemblanity, in risk management—the patterned, preventable misfortune that accrues not from blind chance, but from human agency and organizational design choices that quietly hardwire failure into our operations.

Twenty hair contamination deviations. Seven months to notify suppliers. Critical equipment failures dismissed as “not impacting SISPQ.” Media fill programs missing the very interventions they should validate. This isn’t random bad luck—it’s a quality system that has systematically normalized exactly the kinds of deviations that create inspection findings.

The Architecture of Inevitable Failure

Reading through the six major observations, three systemic patterns emerge that align perfectly with the hidden architecture of failure I discussed in my recent post on zemblanity.

Pattern 1: Investigation Theatre Over Causal Understanding

Observation 1 reveals what happens when investigations become compliance exercises rather than learning tools. The hair contamination trend—20 deviations spanning multiple product codes—received investigation resources proportional to internal requirement, not actual risk. As I’ve written about causal reasoning versus negative reasoning, these investigations focused on what didn’t happen rather than understanding the causal mechanisms that allowed hair to systematically enter sterile products.

The tribal knowledge around plunger seating issues exemplifies this perfectly. Operators developed informal workarounds because the formal system failed them, yet when this surfaced during an investigation, it wasn’t captured as a separate deviation worthy of systematic analysis. The investigation closed the immediate problem without addressing the systemic failure that created the conditions for operator innovation in the first place.

Pattern 2: Trend Blindness and Pattern Fragmentation

The most striking aspect of this 483 is how pattern recognition failed across multiple observations. Twenty-three work orders on critical air handling systems. Ten work orders on a single critical water system. Recurring membrane failures. Each treated as isolated maintenance issues rather than signals of systematic degradation.

This mirrors what I’ve discussed about normalization of deviance—where repeated occurrences of problems that don’t immediately cause catastrophe gradually shift our risk threshold. The work orders document a clear pattern of equipment degradation, yet each was risk-assessed as “not impacting SISPQ” without apparent consideration of cumulative or interactive effects.

Pattern 3: Control System Fragmentation

Perhaps most revealing is how different control systems operated in silos. Visual inspection systems that couldn’t detect the very defects found during manual inspection. Environmental monitoring that didn’t include the most critical surfaces. Media fills that omitted interventions documented as root causes of previous failures.

This isn’t about individual system inadequacy—it’s about what happens when quality systems evolve as collections of independent controls rather than integrated barriers designed to work together.

Solutions: From Zemblanity to Serendipity

Drawing from the approaches I’ve developed on this blog, here’s how Catalent could transform their quality system from one that breeds inevitable failure to one that creates conditions for quality serendipity:

Implement Causally Reasoned Investigations

The Energy Safety Canada white paper I discussed earlier this year offers a powerful framework for moving beyond counterfactual analysis. Instead of concluding that operators “failed to follow procedure” regarding stopper installation, investigate why the procedure was inadequate for the equipment configuration. Instead of noting that supplier notification was delayed seven months, understand the systemic factors that made immediate notification unlikely.

Practical Implementation:

Retrain investigators in causal reasoning techniques
Require investigation sponsors (area managers) to set clear expectations for causal analysis
Implement structured causal analysis tools like Cause-Consequence Analysis
Focus on what actually happened and why it made sense to people at the time
Implement rubrics to guide consistency

Build Integrated Barrier Systems

The take-the-best heuristic I recently explored offers a powerful lens for barrier analysis. Rather than implementing multiple independent controls, identify the single most causally powerful barrier that would prevent each failure type, then design supporting barriers that enhance rather than compete with the primary control.

For hair contamination specifically:

Implement direct stopper surface monitoring as the primary barrier
Design visual inspection systems specifically to detect proteinaceous particles
Create supplier qualification that includes contamination risk assessment
Establish real-time trend analysis linking supplier lots to contamination events

Establish Dynamic Trend Integration

Traditional trending treats each system in isolation—environmental monitoring trends, deviation trends, CAPA trends, maintenance trends. The Catalent 483 shows what happens when these parallel trend systems fail to converge into integrated risk assessment.

Integrated Trending Framework:

Create cross-functional trend review combining all quality data streams
Implement predictive analytics linking maintenance patterns to quality risks
Establish trigger points where equipment degradation patterns automatically initiate quality investigations
Design Product Quality Reviews that explicitly correlate equipment performance with product quality data

Transform CAPA from Compliance to Learning

The recurring failures documented in this 483—repeated hair findings after CAPA implementation, continued equipment failures after “repair”—reflect what I’ve called the effectiveness paradox. Traditional CAPA focuses on thoroughness over causal accuracy.

CAPA Transformation Strategy:

Implement a proper CAPA hierarchy, prioritizing elimination and replacement over detection and mitigation
Establish effectiveness criteria before implementation, not after
Create learning-oriented CAPA reviews that ask “What did this teach us about our system?”
Link CAPA effectiveness directly to recurrence prevention rather than procedural compliance

Build Anticipatory Quality Architecture

The most sophisticated element would be creating what I call “quality serendipity”—systems that create conditions for positive surprises rather than inevitable failures. This requires moving from reactive compliance to anticipatory risk architecture.

Anticipatory Elements:

Implement supplier performance modeling that predicts contamination risk before it manifests
Create equipment degradation models that trigger quality assessment before failure
Establish operator feedback systems that capture emerging risks in real-time
Design quality reviews that explicitly seek weak signals of system stress

The Cultural Foundation

None of these technical solutions will work without addressing the cultural foundation that allowed this level of systematic failure to persist. The 483’s most telling detail isn’t any single observation—it’s the cumulative picture of an organization where quality indicators were consistently rationalized rather than interrogated.

As I’ve written about quality culture, without psychological safety and learning orientation, people won’t commit to building and supporting robust quality systems. The tribal knowledge around plunger seating, the normalization of recurring equipment failures, the seven-month delay in supplier notification—these suggest a culture where adaptation to system inadequacy became preferable to system improvement.

The path forward requires leadership that creates conditions for quality serendipity: reward pattern recognition over problem solving, celebrate early identification of weak signals, and create systems that make the right choice the easy choice.

Beyond Compliance: Building Anti-Fragile Quality

The Catalent 483 offers more than a cautionary tale—it provides a roadmap for quality transformation. Every observation represents an invitation to build quality systems that become stronger under stress rather than more brittle.

Organizations that master this transformation—moving from zemblanity-generating systems to serendipity-creating ones—will find that quality becomes not just a regulatory requirement but a competitive advantage. They’ll detect risks earlier, respond more effectively, and create the kind of operational resilience that turns disruption into opportunity.

The choice is clear: continue managing quality as a collection of independent compliance activities, or build integrated systems designed to create the conditions for sustained quality success. The Catalent case shows us what happens when we choose poorly. The frameworks exist to choose better.

What patterns of “inevitable failure” do you see in your own quality systems? How might shifting from negative reasoning to causal understanding transform your approach to investigations? Share your thoughts—this conversation about quality transformation is one we need to have across the industry.