Safety Gaps That Keep Shipping

The modern drone industry was born in garages and hobby shops. In the early 2000s, remote-controlled (RC) aircraft were the province of engineers, model-plane enthusiasts, and a few university labs experimenting with GPS and autopilot systems. The idea of an “unmanned aerial vehicle” (UAV) sounded like the province of the military, far from everyday life. That all began to change as the smartphone supply chain matured. Accelerometers, gyroscopes, and cameras—components that were once expensive and rare—became cheap and accessible to anyone. Chris Anderson, then editor of Wired and later founder of 3D Robotics, called this “the peace dividends of the smartphone war.” The miniaturized components first developed for iPhones and Androids laid the groundwork for drones. What began as a hobby was now poised to become a product category. By 2010, the pieces had come together: open-source flight controllers like ArduPilot, lightweight GoPro cameras, and compact brushless motors. American startups such as 3D Robotics, Airware, and GoPro imagined a future of consumer drones that would change photography, surveying, and eventually transportation. It looked, for a moment, like Silicon Valley might own the skies. Parrot from France introduced the Parrot AR.Drone flying hardware piloted over Wi-Fi with a smartphone in January 2010 at CES Las Vegas. The rise of DJI and the first American retreat While U.S. startups were designing prototypes, DJI was learning to manufacture. Founded in 2006 in Shenzhen, DJI built on China’s unique advantage: a complete hardware ecosystem within a single city. Parts could move from concept to production in days. By 2015, the company’s Phantom and Mavic models had become the face of consumer drones. American companies could not match DJI’s speed or cost. GoPro’s Karma drone failed after recalls. Airware and 3D Robotics ran out of capital. The American drone boom of the early 2010s ended almost as quickly as it began. But that collapse prompted a shift in thinking. Competing with DJI on price was off the table. Competing on intelligence might not be. DJI’s Phantom drone has long been a favorite of photographers for aerial imaging. Building autonomy as a differentiator Skydio emerged in 2014 with a different philosophy: drones could (and should) fly themselves. The company’s founders, including CEO Adam Bry, came from MIT’s robotics program, where they had studied computer vision and control systems. Instead of depending on a pilot’s skill, their drones would understand the environment and make decisions in real time. This early focus on autonomy gave Skydio a natural edge in the new era of AI. “We make flying robots that give people superpowers,” Bry said in an interview with Jon Bruner on the Go/No-Go podcast. He charts the evolution of the industry as moving from toys to tools to infrastructure: drones that were once just for fun now perform critical work in law enforcement, energy, and defense. Skydio CEO Adam Bry outlined the progression of drones from toys to tools to infrastructure in his 2025 Ascend keynote speech. That evolution also had implications for the business model. Skydio’s products are designed, built, and tested in the United States, which is almost unheard-of in modern electronics. This vertical integration can lengthen the product development timeline, but the strategic importance of a secure supply chain outweighs its drawbacks. As AI becomes central to navigation and safety, having full control over hardware and software reduces risk and builds trust with institutional customers. Manufacturing trust By the late 2010s, DJI’s dominance had drawn political scrutiny. U.S. agencies grew uneasy about using Chinese-made equipment for sensitive data collection. The Department of Defense responded with the Blue sUAS program to certify domestically produced drones. The conversation shifted from market competition to national security. However great the risks, reshoring an entire supply chain cannot happen overnight, and it might not even be possible within our lifetimes. Most drone components, including batteries, motors, image sensors, and plastics, still originate in China. Batteries are the hardest link to replace. China controls roughly three-quarters of global cell production and nearly all of the raw-material refining. Korea, Japan, and Taiwan offer partial alternatives, but the industrial gap is wide. U.S. manufacturers are now rebuilding local ecosystems, combining domestic assembly with allied sourcing. New facilities in California, Texas, and Mexico focus on short-run, high-reliability production rather than mass volume. This follows the model of American dominance in aerospace manufacturing, doubling down on precision rather than competing on price. Drones as critical infrastructure When drones move into public safety, defense, and energy, they become enmeshed in national infrastructure. Bry often draws this parallel directly: drones are no longer gadgets or tools but “critical physical infrastructure” in their own right. They monitor grids, assess storm damage, and assist in policing and firefighting. As drones evolve from cameras to critical infrastructure, reliability becomes non-negotiable. A failure in flight can interrupt power delivery, compromise inspection data, or delay an emergency response. For autonomous systems, the standard is even higher: the AI cannot guess, hallucinate, or forget. Every component must perform predictably despite heat, dust, and fatigue. Ensuring that consistency now depends on advanced inspection methods, including industrial CT, which allows engineers to see inside complex assemblies, verify structural integrity, and detect hidden defects before they become field failures. A Skydio drone inspects electrical infrastructure, using onboard AI to navigate complex environments and capture detailed visual data safely and autonomously. This is where American firms have an advantage, especially in the Bay Area where everyone from hard tech startups to power players like OpenAI and Anthropic are building the future of AI. Building autonomy into hardware is less about cheap components and more about tightly integrating the full stack of training, testing, and iteration between software and production. It requires the kind of engineering culture that thrives close to the factory floor. Policy and the next industrial race The federal government has begun to treat drone manufacturing as strategically important. The FCC now has authority to restrict or retroactively ban transmitters from companies tied to national-security risks. Reports of disguised DJI products, sold under new brand names to evade import rules, have reinforced the urgency of domestic production. Replacing China’s dominance will take years. It means building not just factories but the networks of suppliers, machinists, and engineers that make quick iteration possible. The lessons from electric vehicles and semiconductors are clear: reshoring works only when supported by steady procurement, tax incentives, and long-term demand. The Federal Communications Commission announces new rules expanding its authority to restrict technology from foreign manufacturers, including potential limits on DJI drones. The emerging model is regional, not isolationist. Allied sourcing across Japan, Korea, and Taiwan can provide resilience without abandoning cost efficiency. A new phase of flight The first drone era was about invention, but the next is about comprehensive industrialization. Drones’ transition from toys to tools to infrastructure captures more than a shift in technology. It marks the moment when drones stop being products and start being part of the built environment, flying machines that belong to the same world as power lines and cell towers. And these hard-won lessons learned by the drone industry offer a blueprint for the rest of American manufacturing. For U.S. companies like Skydio, that evolution may prove decisive. Autonomy is where America is already leading, not by racing to the bottom on price, but through the smarter integration of AI into real, physical systems. The products that will define the next decade will be those whose proven reliability at every level can’t help but earn our trust.

Scan of the Month

Apple AirPods Pro (2nd Generation) CT Teardown

The second-generation AirPods Pro arrived in September 2022 with Apple's H2 chip at its center. The H2 brought meaningfully better active noise cancellation, Adaptive Transparency mode, improved battery life, and the processing headroom for personalized spatial audio. The high-level layout of the earbuds, though, looked familiar. We scanned them to find out how much had actually changed. The charging case The case follows the same basic architecture as the first generation: two batteries flanking a central logic board, magnets to hold the lid closed, and a magnet in each charging well to pull the earbuds into position. The new case adds four magnets around the inductive charging coil, which align it with a MagSafe or Apple Watch charger. It also gains a U1 chip for precise Find My tracking and an outward-facing speaker for audio cues and location pings. A metal eyelet on the side of the case is nominally a lanyard attachment point. Our scan reveals that it is 4 mm deep and 18 mm long, with a tail that connects to a metal pad around the Lightning port. The geometry suggests it may also function as an antenna supporting the improved Find My capability, though Apple has not confirmed that function. The hinge on the back of the case appears to be metal, matching the finish of the eyelet. The scan shows otherwise. The visible hinge components are electroplated plastic with a metal pin inside. Apple matched the electroplated finish to the aluminum eyelet closely enough that the difference is invisible without a scan. The result is a premium appearance at lower cost and weight than solid metal would require. One detail in the lid mechanism is worth noting. The first-generation case used a spring-loaded pin assembly to control the lid. The second generation replaces it with a thin bistable metal strap that snaps between two stable positions. Fewer parts, same function, lower assembly complexity. The earbuds The earbud layout is essentially unchanged from the first generation. A battery sits in the head of the earbud with the logic board positioned behind it. The antenna and charging contacts run through the stem. A large magnet below the battery seats the earbud in the charging case. Three microphones are arranged as before: one at the bottom of the stem for voice pickup during calls, one facing outward at the top of the stem for ambient sound collection used in noise cancellation, and one pointed into the ear canal to tune audio response. The second-generation model adds a sensor along the flat surface of the stem to support the new swipe gesture for volume control. The performance improvements Apple advertises, better noise cancellation, longer battery life, cleaner audio, come primarily from the H2 chip rather than from any fundamental change in how the earbuds are built. The physical architecture is an iteration, not a redesign. What the scan shows The second-generation AirPods Pro are a tightly packaged piece of consumer electronics with meaningful detail work in the case and measured refinement in the earbuds. What the scan makes visible is the continuity with the first generation: Apple changed the chip, updated the case, and left the physical earbud architecture largely intact. The performance gains are real. So is the conservatism of the underlying design.

Materials World

Apple Rethinks Paper Packaging

Apple’s product experience has always started long before a device is powered on. The box matters, so much so that we probably owe the entire “unboxing” phenomenon to Apple. For decades, the company has treated packaging as an extension of the product itself, engineered to guide the user’s hands, stage the reveal, and convey precision through sound and resistance. Today, that same design discipline has been redirected toward sustainability. The changes here are almost imperceptible, a far cry from the effect of compostable straws replacing plastic ones. Apple’s latest packaging, seen through the eyes of industrial CT, shows how far the company has gone to replace plastic with paper, down to the smallest fold and insert. What the scans show Industrial CT imaging exposes what we normally can’t see: the layered structure that gives this iPhone 17 Pro box its rigidity. Its walls are built from stacked sheets of laminated paperboard, compressed and bonded under heat and pressure. The result is a square, seamless enclosure, its visible surfaces wrapped in a continuous paper layer that conceals every joint. CT cross-sections reveal the layered construction of the iPhone 17 Pro box, made from laminated and stacked paperboard with scored folding corners and a molded fiber tray. Inside sits a molded fiber pulp tray, formed by vacuum suction in a heated aluminum tool. The fibers, drawn from a mix of recycled and virgin sources, are pressed to a density that rivals injection-molded plastic. This tray holds the phone in place and absorbs shock during shipping, performing the same structural role as plastic but with a lower carbon footprint. Industrial CT top-down view showing molded fiber tray and stray paper fiber, evidence of the natural variation in fiber-based materials. Engineered paper, not cardboard What strikes us as simple is the result of several refined paper-engineering techniques. The outer shell’s score lines are cut before lamination, allowing it to fold sharply without cracking. Corners are glued and stacked for thickness, then wrapped by a laminated outer layer that gives the box its continuous surface. In effect, Apple’s packaging combines two crafts: bookbinding and molded pulp forming. The outer box is built like a miniature hardcover book, while the inner tray is the product of advanced fiber-molding technology, similar in principle to an egg carton but executed with the signature finesse of a company with a $4 trillion market cap. A hardcover book, a molded pulp egg carton, and a generic smartphone box show the lineage of Apple’s current fiber-based design. From excess to refinement A decade ago, opening an iPhone box meant peeling away multiple layers of plastic and lifting molded trays holding chargers, cables, and earbuds. The new iPhone 17 Pro box is smaller by nearly half, contains no charger, and is made entirely from fiber-based materials. The difference is more than aesthetic. Apple has replaced every plastic component with laminated paperboard and molded fiber pulp while preserving the magic of unboxing. [Embed video here: Original iPod Classic packaging comparison — showing the size, materials, and inserts of Apple’s early product boxes.] Before Apple’s boxes became fiber sculptures, they were cavernous theaters of plastic and foam. This CT scan of the original iPod box shows how much empty space, weight, and material Apple has since engineered away. The fact that the original iPod packaging is now a collectible shows just how far we’ve come. Marketplaces like StockX use Lumafield industrial CT to verify the authenticity of sealed vintage Apple products, a testament to how packaging has become part of the brand’s cultural and material legacy. The sustainability shift According to Apple’s 2025 Environmental Progress Report, the company is on track to eliminate all plastic from its packaging by the end of 2025. In 2024, products including the iPhone 16, Apple Watch, and MacBook shipped in 100 percent fiber-based boxes. Packaging now consists of roughly 60% recycled fiber, 39% responsibly sourced virgin fiber, and less than 1% plastic. All fiber is either recycled or sourced from forests managed under FSC, PEFC, or Bonsucro standards. Apple’s goal is for every box to be recyclable through mixed-paper streams, the same process (without disassembly or sorting) used for cereal boxes. Apple’s 2025 Environmental Progress Report shows packaging plastic reduced by 20 percentage points since 2015, replaced by recycled and responsibly sourced fiber, part of its goal to eliminate plastic packaging entirely by the end of 2025. The company’s report notes: “Along our journey, we’ve addressed many packaging components that typically rely on plastic, including large product trays, screen films, wraps, and foam cushioning. We’ve replaced each with fiber-based alternatives and implemented innovative alternatives to the small uses of plastics across our packaging — like labels and lamination. At the same time, we’re taking steps to confirm that our packaging is recyclable and that the fiber we source comes from recycled sources or responsibly managed forests.” (Apple Environmental Progress Report 2025, p. 21) Engineering at scale Transitioning to fiber packaging is far more complex than replacing one material with another. Paper behaves differently than plastic; it absorbs moisture, expands unevenly, and varies in thickness. To preserve the precise fit and slow-release “air cushion” feel of an iPhone box, Apple reengineered its production tools and quality-control systems to manage those variations. That transformation extends across its global supply chain: more than 70 packaging vendors have joined Apple’s redesign program, with 90 percent expected to complete the fiber transition by mid-2025. Apple’s work also reflects a larger movement reshaping electronics packaging. Google committed to plastic-free designs for Pixel, Nest, and Fitbit by 2023, Samsung pledged a full switch to recycled paper packaging for the Galaxy S24, and Xiaomi (the so-called “Apple of China”) began cutting plastics from its boxes in 2020. Yet Apple’s influence is unmatched. When it standardizes a solution, the industry tends to follow. This path builds on groundwork laid by Dell more than a decade ago, when they began experimenting with bamboo, mushroom-based foams, and molded pulp. “Almost every person on this planet of eight billion people touches packaging multiple times a day,” said Dell’s Director of Procurement & Packaging Engineering Oliver Campbell in a 2023 interview with Forbes. “If you’re anything like me, you wake up, stagger to the shower, and it’s packaging—everything.” That omnipresence is what makes even small changes in the CPG space have massive consequences. The material frontier Material science is continually expanding what fiber packaging can do. By blending virgin wood fibers with bamboo, agricultural waste, or recycled pulp, engineers can fine-tune stiffness, texture, and strength. Today, some of these composites can match the polish of plastic while remaining fully recyclable. For Apple, that progress is both technical and symbolic. The box that once defined the company’s design precision now speaks to its environmental commitments too, proof that sustainability has become part of the product itself.

Design to Reality

Apple vs. Meta: Same Problem, Different Answers

Every VR headset is trying to solve the same impossible brief: put a powerful computer on someone's face, keep it cool, keep it light, and make it comfortable enough to wear for hours. The hardware decisions that follow from that brief reveal more about a company's design philosophy than almost any other product they make. We scanned the Apple Vision Pro, the Meta Quest Pro, and the Meta Quest 3 using an industrial CT scanner to see how each team answered that brief. The scans don't settle any arguments about which headset is better. What they show is how differently Apple and Meta think about the problem. The layout problem In our CT scans of the Apple Vision Pro, we find a product where the external form dictated the internal arrangement. Everything inside is canted to fill the available space within the curved enclosure without detracting from the aluminum frame and laminated glass front. The main circuit board is built around a flexible PCB ribbon, and components are packed at different angles throughout. Steve Jobs described a principle that is visible in the scan: at Apple, designers define the shape they want, and engineers make the electronics fit inside it. The Quest Pro and Quest 3 both use a traditional flat main board with components stacked on a single plane. This is a well-established approach that works reliably with off-the-shelf components and conventional manufacturing processes. It is also the approach you take when the internal layout drives the external form rather than the other way around. The Meta headsets suggest not so much that the emphasis is reversed, but rather that the products have different priorities. Apple, at least for now, is going all in for early adopters willing to spend $3,500 on a new kind of computer. Meta is pursuing a design philosophy aimed at making VR accessible: at $500, the Quest 3 is one-seventh the price of the Vision Pro. Displays Even though all three headsets feature video passthrough, none of them are glasses. They are screens strapped to a face. The Vision Pro uses two Sony micro-OLED displays at 3,680 x 3,140 pixels per eye, delivering more than 23 million pixels combined. The Quest Pro offers dual LCD panels at 1,800 x 1,920 pixels per eye; the Quest 3 runs at 2,064 x 2,208. Sensors The Meta headsets pair inside-out cameras for spatial tracking with handheld controllers that add haptic feedback and pressure sensing. The Quest 3 includes a Time of Flight sensor for depth sensing. The Vision Pro relies entirely on eye tracking and hand tracking for input. There are no physical controllers. That requires a substantial sensor array: four eye-tracking IR cameras, a LiDAR scanner, the TrueDepth camera system from Face ID, downward-facing cameras for hand tracking, and a six-microphone array with directional beamforming. The R1 chip processes all of it with 12-millisecond latency, fast enough that virtual content feels attached to the physical world rather than lagging behind it. The Quest Pro had originally planned to include a Time of Flight sensor but omitted it in the final product, a detail that reflects how development timelines shape shipping hardware in ways that market positioning doesn't always explain. Processors The Vision Pro runs on Apple's M2 and R1 chips: the M2 handles computing, and the R1 handles real-time sensor processing at 256 GB/s memory bandwidth. The Meta Quest headsets use the Qualcomm Snapdragon XR2 platform. Batteries The Vision Pro offloads its battery to an external pack connected by cable, a decision that keeps weight off the face at the expense of a tethered power source. The Quest Pro integrates a curved battery that follows the headset's rear contour. The Quest 3 places its battery at the front of the headset, the simplest and most direct approach of the three. Thermal management If you pick up a Vision Pro and look at it, you won't see a heat sink. There is no obvious thermal solution on the outside. That restraint is intentional, and it makes the engineering challenge considerably harder. Inside, two micro-blowers handle cooling for the M2 and R1 chips along with their associated display boards. Each fan is doing double duty: managing both a processor and a display in a device where those components sit directly adjacent. The airflow path follows a nautilus-shaped channel that draws air from the bottom of the headset and vents it out the top. The metal chassis functions as a passive thermal element, and thermal compound bonds the fans to the chip packages. When you wear the Vision Pro, you are largely unaware that any of this is happening. That is the engineering achievement. The Quest Pro uses a copper heat pipe and two conventional cooling fans, a configuration recognizable to anyone who has opened a laptop. The Quest 3 simplifies further to a single fan with no heat pipe. Both are effective at their respective thermal loads. Neither is concealed. Audio systems The Vision Pro powers audio through two AudioPods with a dual-driver setup, capable of delivering a surround sound experience calibrated to the user's head and ear geometry. The Quest headsets use a more minimal arrangement, with drivers positioned to channel sound directly toward the ear. What the scans show The Vision Pro and the Quest headsets cost $3,500 and $500 respectively, and the scans show where that difference goes. The Vision Pro's internal layout, thermal architecture, display technology, and sensor array are all purpose-built for this device. The Quest headsets are built from more conventional components in a more conventional layout, optimized for a price point that makes VR accessible rather than aspirational. Neither approach is wrong. The Meta headsets deliver genuine capability at a fraction of the cost, and their economy of means is its own engineering accomplishment. The Vision Pro is a statement about what spatial computing could become, built without the cost constraints that govern most consumer electronics. The scans make both facts visible simultaneously. A note on currency: the original Vision Pro used the M2 chip. Apple updated the headset to M5 in October 2025. The physical design, thermal architecture, and sensor array are unchanged; everything described here applies to both versions.

The Quality Gap

Behind the Battery Report

Behind the Battery Report Today, we released Lumafield’s Battery Quality Report, an investigation into the hazards that hide inside the lithium-ion battery supply chain. For this study, I CT-scanned 1,054 cylindrical 18650 cells from ten different brands, spanning the gamut from reputable, well-known battery manufacturers to ultra-discounted cells from Temu and several sources in-between. As battery users, we all intuitively expect alternative-brand cells to be lower quality, but it’s easy to gloss over how much worse they may really be, especially when a cheaper look-alike promises the same or better specs, and the name brands feel overpriced. However, our CT scans revealed much bigger quality gaps than I could’ve imagined, and once we factored in how dangerous a bad battery can be, the story turned from curious to genuinely concerning. Diving into battery quality questions As a Technical Product Marketing Manager at Lumafield, I design experiments to test and show how engineers can leverage CT to solve real problems. I’m constantly looking for areas to double-click on, and as a gadget-lover, batteries are all-too-often on my mind. A few months ago, I wrote a blog post about the Anker A1263 power bank recall, and it achieved a surprising level of virality given the relatively unglamorous product it covered. But perhaps the practical pervasiveness of batteries in all of our daily lives is exactly what drove so much interest in the piece. In that article, I scanned five power banks: enough to make a few concrete observations, but a tiny sample against the realities of production scale. I wanted the insights that could come with a larger dataset, and 18650 cells seemed like the obvious choice. 18650s are standardized battery cells, their name reflecting that they measure 18 mm in diameter and 65 mm long. These cells are absolutely everywhere. Over five billion are produced each year, and they can be found in everything from electric toothbrushes and cordless drills to even some electric vehicles. For example, each of the Anker power banks I scanned contained 3 18650s, and my colleague’s e-bike battery contained a whopping 39. Though the majority of 18650s are assembled into packs of multiple cells and deeply integrated into devices, there are a few applications where the cells are user-replaceable, such as vapes or flashlights, making them relatively easy to procure. Study methodology I ordered at least 100 cells from each of 10 brands, ending up with 1,054 cells total, given purchase increment minimums and free extras received. Every cell was scanned. Three sets came from the well-known OEMs—Murata, Samsung, and Panasonic—sourced from reputable online stores specializing in 18650 batteries. Three sets came from the rewrap brands Efest, Vapcell, and Trustfire, with Efest and Vapcell coming from reputable 18650 online storefronts, and the Trustfire cells coming from their brand website. The final four were more suspicious. I chose three of the brands for being lower cost: Treasurecase and Maxiaeon came from Temu, Benkia from Amazon. The SOOCOOL cells were not cheap, but were advertised on Amazon as “authentic” Samsung 30Q cells, their wraps closely but not perfectly mimicking the aesthetic of the genuine Samsung cells. A counterfeit SOOCOOL “Authentic 30QP” next to an OEM Samsung 30Q cell. Table of cells scanned for the experiment. Every battery was labeled on arrival for traceability, then scanned as received. We scanned on a Lumafield industrial CT system with a 130 kV microfocus source. Using Ultra-Fast CT, we scanned each cell in under a minute, then processed the data with our Battery Analysis Module, which automatically finds electrode edges and extracts study metrics in bulk. Lumafield has two industrial CT product lines: Neptune, a compact and easy-to-install solution ideal for R&D offices and labs, and Triton, optimized for high-volume environments. We built the workflow to reflect production realities, because annual battery output is enormous and speed determines practical value. In production, Triton paired with Ultra-Fast CT can scan cylindrical cells in under five seconds, enabling more than 720 cells per hour and making battery inspection feasible at scale. To support what we saw in the CT scans, I added a few supplementary tests. I measured capacity on one sample from each set. Marketing claims as high as 9,900 mAh appeared on some of the cell listings, a number that far exceeds what this format can deliver, and the measurements confirmed the mismatch. We also ran longer scans on a handful of cells to resolve more fine details and validate the automated measurements. Defining characteristics of battery quality The analysis focused on what governed safety. I prioritized three parameters for anode overhang and two for alignment. Median anode overhang per cell gave us a baseline for each brand. I used median rather than mean because a few extreme outliers would have distorted the averages. I also compared the maximum and minimum anode overhang values for each cell. Insufficient overhang increases the likelihood of lithium plating during charge, which accelerates aging and raises the probability of internal shorts. Alternatively, excessive anode length can reach the can and create a different short path. For alignment, we calculated the delta between the highest and lowest positions of both the cathode and the anode. Cylindrical cells are wound, and telescoping introduced during winding can become a seed for shorts and other defects. Large deltas point to weak process control and higher risk elsewhere in the line. Comparison of good and bad anode overhang. Side-by-side of ideal and suspicious edge alignment. Findings: Far more variance than expected The results aligned generally with our expectations, but the degree of difference was surprising. OEM cells showed tight control, with anode overhang centered near the 0.50 mm industry average and modest telescoping. The rewraps generally matched the OEM medians quite closely, though their spreads were wider, and the minimum overhang tails in particular stretched lower. Trustfire stood apart from Efest and Vapcell with a significantly worse distribution, a reminder that rewraps are blind boxes that carry real risk. Unless you scan and measure what you buy, you cannot be sure what exactly is hiding under the sleeve. The low-cost and counterfeit group was in a league of its own. Distributions were broader across every metric, and each brand in this tier produced at least one unit with negative anode overhang. The worst results came from the Temu brands. We found fourteen Treasurecase and fifteen Maxiaeon cells with cathode overhang. Across the entire low-cost group, thirty-three of 424 cells presented cathode overhang, which suggests that as many as one in thirteen such cells may carry geometry that accelerates aging and drives a higher chance of internal shorting. The standard deviations ran roughly seven times larger than the OEMs, which signals much weaker process control and, by extension, a higher likelihood of other hidden defects. Implications and recommendations These findings have implications for every stakeholder that interacts with batteries, from battery manufacturers and device integrators to end users. Original battery manufacturers prioritize quality, and the study results speak to their tight specifications and strong execution. The risk for these companies sits downstream, with scrap potentially making its way into the rewrap market, or OEM cells being “recycled” and resold when end-of-life devices are dismantled. Scrutiny of scrap procedures and channel partners, as well as improved traceability features, can protect these brands from being tarnished by the actions of less scrupulous players. Device integrators can reduce their exposure to dangerous batteries by validating their supply with incoming inspection, especially when buying through distributors or less established sources. Battery incidents can seriously mar the reputations of device integrators, making verification critical. Pack design is also essential to ensure battery safety, as building in sufficient spacing and protections can keep potential failures limited to a single cell. Consumers must also actively manage their individual battery risk exposure, given the gaps in supply chain and regulatory oversight. It’s important that users don’t mix brands, capacities, or ages in a shared device, that they protect cells from physical abuse and extreme temperatures, and that end-of-life batteries are recycled properly. Despite the popularity of “dupes,” when it comes to batteries, the savings aren’t worth the risk. Consumers should also familiarize themselves with the warning signs of battery incidents. Heat during or after charging, swelling or a soft feel, hissing, sharp chemical odors, discoloration, liquid residue, sudden drops in capacity, repeated shutdowns, or a charger that refuses to start all suggest internal damage. If you see any of these indicators, stop using the battery immediately and dispose of it at a proper facility. Parting thoughts Today’s volatile trade and tariff environment is complicating cell provenance more than ever, and it’s becoming easier and more attractive to slip gray-market substitutes into supply chains. In such a tumultuous ecosystem, CT becomes a practical tool for certainty. Our 754 non-OEM cells were mysteries until we scanned them, and while some matched OEM behavior, others fell far outside any reasonable specification. Most failures will never cause fires, but even a common reliability failure can disrupt the functionality and experience of a product. The rare battery incidents that do ignite are catastrophic, and this is why battery quality is so fundamentally important. Negative anode overhang and large irregularities in edge alignment do not guarantee failure, but they put a heavy hand on the scale in the wrong direction. Everyone who interacts with batteries, from cell manufacturers and device integrators to the people who use their products, should be intentional with managing these inherently dangerous devices. The Lumafield Battery Quality Report reveals the magnitude of risk in an uncontrolled battery supply chain. Industrial CT inspection allowed us to unearth the risky realities of the 18650 cells we ordered, by turning some of those key quality indicators into measurable geometry. Ultimately, CT is a powerful tool that can make the invisible visible by bringing much-needed clarity to the battery supply chain’s unknowns. SEO Meta Description: A closer look at the Battery Quality Report: methodology, surprising results, and what CT scanning uncovered inside low-cost lithium-ion cells. CT scans of 1,054 18650 batteries revealed major quality differences between OEM, rewrap, and low-cost cells, with defects concentrated in the cheapest brands. Negative anode overhang and poor alignment were key risks, raising the chance of internal shorts, accelerated aging, and thermal events. The study highlights supply chain vulnerabilities and shows how CT can give manufacturers, integrators, and consumers clearer visibility into battery safety.

The Quality Gap

Blind Spots in Electronics Quality

Anyone who has debugged a circuit board knows the feeling: one cold joint or hidden bridge can take down an entire system. In mass production, that same problem scales into yield loss, warranty claims, and painful late-night root-cause hunts. What makes it worse is that many of the defects causing intermittent or field failures aren’t visible to the naked eye or even to the inspection systems most factories depend on.This is the reality of modern electronics manufacturing. Solder defects are no longer surface problems. They’re buried under packages, inside layers, and beneath metal shields. And while inspection tools have improved, most of the ones in daily use were designed for a world of through-hole and early SMD components, not for today’s dense, fine-pitch assemblies.Evolution of a hidden problemThrough-hole soldering and early surface-mount components gave engineers plenty of visual feedback. You could see whether a joint was smooth, shiny, and well-formed, or cold, cracked, or bridged. A digital microscope and a steady hand were often enough to assess quality.That changed in the 1990s when Motorola introduced the Ball Grid Array (BGA). Instead of leads around the package edge, BGAs used a grid of solder balls underneath. It was a breakthrough: thousands of electrical connections in a tiny footprint. But it also created a new kind of invisibility. The solder joints that mattered most were now completely hidden.Optical inspection couldn’t reach them. Electrical tests could confirm continuity, but not explain why a joint might fail under heat or stress. The industry turned to X-ray imaging for answers.Hidden solder joints beneath a BGA packageWhat X-rays can and can’t seeTraditional 2D X-ray and automated optical inspection (AOI) remain essential tools on any electronics line. AOI quickly flags missing components, bridges, and surface anomalies. X-ray systems can reveal hidden solder joints and flag major defects like voids or bridging under BGAs. But these techniques share the same limitation: they collapse three-dimensional data into a single projection.That projection hides as much as it shows. Two defects may overlap in the same line of sight, making them look like one. Tiny voids, cracks, or head-in-pillow separations can sit just a few microns apart, enough to pass a 2D X-ray check while still creating a ticking time bomb for field reliability.For example, voids inside solder joints reduce both electrical conductivity and thermal transfer. A 2D image can suggest their presence, but not quantify their true size or location. A void near the center of a solder ball may be harmless, while one near the interface can cause catastrophic detachment after a few hundred thermal cycles. The difference is invisible without depth.CT cross-section showing voids within a BGA solder ball array (Void fraction = 0.3%).The problem of scale and densityAs packaging density increases, so does risk. A modern smartphone board can have more than a thousand solder joints per square inch, many under stacked memory, RF modules, or metal cans that never come off again after final assembly.Quad Flat No-Lead (QFN) devices compound the issue. Their large exposed center pads act as both thermal and ground planes. Poor solder coverage here increases resistance, leading to localized heating and sometimes runaway failures. Industry guidelines recommend keeping voiding below 50 percent of the pad area, but without three-dimensional data that is only a guess.Volumetric rendering of a QFN package analyzed using Lumafield VoyagerThe Head-in-Pillow (HiP) defect is even more deceptive. It occurs when a solder ball touches the pad without fully bonding. The joint can pass electrical testing at room temperature, only to fail later when the product heats up or bends. To an AOI or 2D X-ray system, HiP looks like a perfect joint. To the customer, it looks like an RMA.Inspection blind spotsThe inspection toolbox most factories rely on (AOI, in-circuit test, and 2D X-ray) was never designed to see these issues in context. Each method provides one piece of the puzzle:AOI: Fast, inexpensive, and effective for surface-level defects, but blind to hidden joints.2D X-ray: Reveals internal geometry, but lacks depth information and can’t distinguish stacked layers or subtle separations.Electrical test: Confirms whether a circuit is open or shorted, but says nothing about why or where mechanical failure will begin.Cross-sectioning: Offers a definitive view, but it’s destructive, slow, and limited to one small sample.Each tool has its place, but none can fully capture how a real solder joint behaves under heat, stress, and time.Seeing in three dimensionsComputed tomography, or CT scanning, fills that gap not because it replaces the tools above, but because it provides the depth they lack. CT builds a true 3D reconstruction of a part by rotating it under an X-ray beam and combining hundreds or thousands of projection images. The result is a volumetric model that can be sliced, measured, and analyzed from any angle.Comparison between a 2D X-ray and a 3D CT reconstruction of a Raspberry Pi Zero.For engineers dealing with solder defects, that means being able to see exactly what’s happening inside the joint. Is a void near the interface or the center? Is a bridge a solid connection or just surface wetting? Has the pad fully bonded, or is it a head-in-pillow separation waiting to fail?CT data turns those questions into measurements. You can calculate void fraction, measure ball alignment, and trace microcracks that would otherwise go unseen. Most importantly, you can do it on a fully assembled board without cutting it apart.From failure analysis to process controlUntil recently, CT was confined to the lab: expensive, slow, and limited to small samples. But faster reconstruction algorithms and more compact systems have changed that. Today, CT is used for first article inspection and even production sampling to verify solder quality across builds.2D inspection remains essential, but modern electronics demand escalation from surface checks to full-volume analysis to explain why a board passes a test one day and fails in the field the next. That is the quality gap that CT closes.CT has also become a powerful process-improvement tool. Engineers can overlay CT data across multiple builds to pinpoint where voiding increases, where temperature profiles drift, or how pad plating or paste variation influence joint integrity.In that sense, three-dimensional inspection is less about defect hunting and more about process learning. Every scan feeds back into how reflow profiles are tuned, how stencils are cut, and how reliability standards are enforced.What’s next for quality managementThe electronics industry has spent decades chasing better throughput and automation. Inspection kept pace with that evolution until components went fully hidden. Now, quality management faces a different challenge: not moving faster, but going deeper. As assemblies become smaller and denser, failure visibility becomes the defining metric. Optical systems see the surface, two-dimensional X-ray sees the outline, but only three-dimensional scanning sees the whole truth.Every tool has its place in the line, and understanding their limits is as important as knowing how to use them. The future of solder quality is not about adding more automation. A defect you can’t see is one you can’t prevent. Choosing the right level of visibility for the risk you are managing is how you prevent the next hidden flaw from escaping your line and reaching your customer.

Design to Reality

Building for the Brain: Pioneering a Long-Term Neural Implant

When you’re building an implanted medical device for the brain, you can’t cut corners. Engineering for performance is a given when you’re designing for a living system. To ensure that there are no adverse effects, an active implantable device cannot dissipate excessive heat, or there could be a risk to the surrounding tissue. Every decision, from materials, to power draw, and even packaging, is shaped by the need to do no harm. Designing a medical device is a totally different world from consumer electronics.Stakes of the SystemAt Paradromics, we’re building a brain-computer interface (BCI) to restore speech for people with severe communication disorders: conditions like ALS or spinal cord injury that disrupts the connection between intention and expression. Our first product, Connexus® BCI, records directly from the brain’s speech centers and synthesizes intended speech in real time.My focus as an electrical engineer is on the Cortical Module itself. That’s the component that sits on the surface of the brain, and it’s also the point where the technical demands are highest. The Cortical Module must be reliable, power-efficient, and manufacturable at scale. And it has to last a long time; because we are not interested in devices that require users to have brain surgery to replace their brain implant every two years as with some technologies being explored by others. Our goal is a device that never has to be replaced. That means every material choice and assembly step matters.People often compare implants like ours to wearable electronics, such as a smartwatch or continuous glucose monitor, as they share some similar goals: small size, clean data, minimal intrusion. But once you’re under the skin, everything changes. Now you’re dealing with strict FDA regulations, biocompatibility constraints, and failure modes with the highest possible stakes.Incompressible TimelinesSome timelines are just incompressible, and you have to come to terms with that. You can move fast in design, but once the project moves into manufacturing and validation, your pace becomes capped to that of regulation and biology. I wish someone had told me going into this field, that even with unlimited resources and engineering expertise, there are parts of the process you simply can’t speed up.Regulatory requirements shape design from day one. Connexus BCI is classified as a Class III medical device, the same category as a pacemaker. Implanted medical devices are typically limited to known biocompatible materials, and to support a minimum of a 10 year implant life that means materials like titanium, platinum, and gold. You can try to get new materials approved, but it’s a much longer road. And when you’re already working on a novel device, it’s wiser to avoid adding extra unknowns.Building the Plane While Flying ItThere’s no off-the-shelf playbook for what we’re doing. Our system isn’t life-supporting like a pacemaker, but it is life-enhancing. We work closely with the FDA to define safety and benefit to the patient for a device that doesn’t quite fit their existing categories. We’re being evaluated, while also helping define the rules.The metaphor we use internally is “building the plane while flying it.” It may sound chaotic, but it’s just the reality of working in a space this new. You have to keep moving forward, even while you’re clearing the path beneath your feet.Testing for the FutureThat pressure to get it right shows up most in testing. Our devices go through a long chain of inspection, step by step. Some processes require equipment so specialized there’s no commercial equivalent, meaning we have to do it ourselves. That’s why we use Lumafield's industrial CT scanner to evaluate process outcomes and observe internal structures that can’t be seen any other way. It’s one of the few tools that can keep up with the level of precision we need.In addition to testing in a large animal model, we also run accelerated aging tests, environmental stress simulations, and strenuous mechanical validation to make sure our devices can survive in the body for years. This includes checking for things like thermal performance, electrode durability, and even the consistency of raw materials from one vendor to another.Milestones That MatterThis work isn’t theoretical. Earlier this year, we completed our first human implantation in a patient undergoing unrelated brain surgery. The successful recording and data collection demonstration confirmed that our device performs well in the real world. We saw the kind of neural activity we expected and it validated years of animal studies and lab work. That milestone puts us on the path toward an Early Feasibility Study (EFS), a longer-term clinical investigation, where the device stays in place and is used regularly by a small number of participants. It’s the next big step toward FDA approval.Designed to StayI joined Paradromics right out of college in 2022, drawn to the cutting-edge intersection of engineering and neuroscience. It’s still a young field, and there’s so much left to figure out. But if we’re going to make a real difference for patients, we have to meet the body where it is. That means durability, reliability, and empathy for the people at the center of it all.This Cortical Module should feel less like a medical device and more like a part of someone’s life. If we can get there, we’ll have done our job.

Design to Reality

CT Teardown: AirPods Pro (3rd Generation)

Apple’s new AirPods Pro (3rd Generation) continues the company’s pattern of incremental but carefully executed engineering updates. With changes such as longer earbud battery life, improved noise cancellation and new health-monitoring features, Apple markets these as their most advanced earbuds so far. This teardown builds on our earlier analysis of the AirPods Pro 2.Using our Neptune industrial CT scanner, we conducted a nondestructive teardown of both the earbuds and the charging case to examine how Apple has updated the design internally. You can explore the full-resolution scans yourself inside our Voyager software by creating a free account.EarbudsInternally, the AirPods Pro 3 earbuds retain the proven architecture from the previous generation while integrating new silicon and sensors to support additional features. CT imaging quickly reveals Ball-Grid-Arrays (BGAs) integrated circuit packages distributed along the stem. This is a type of solder joint commonly used in microcontrollers to increase pin count and functionality.These integrated circuits originate from the H2 System-in-Package used in the AirPods Pro 2 but have been broken out and redistributed the integrated circuits along the stalk. Their new positions are identifiable in the CT images by their distinctive BGA solder patterns.CT imaging reveals how Apple redistributed the H2 chip functions from a compact System-in-Package in the AirPods Pro 2 to multiple BGA modules along the Pro 3’s stem, optimizing internal geometry and signal routing.BatteriesWe had initially thought that this change was made to free additional space for a larger battery as Apple had reported an increase in runtime from 6 hours (Pro 2) to 8 hours (Pro 3). However, Voyager’s Dimensioning tools show that the battery size of the Pro 3 is marginally larger at: Ø10.8 × Ø2.3 × 4.8 mm. This suggests that these improvements in runtime stem from changes in the battery’s active materials, or system-level power optimizations, rather than a dramatic increase in battery volume. The redistributed circuit board is therefore more likely driven by the new external geometry of the earbud.Cross-sections show the Pro 3 battery’s tightly wound internal structure. Despite only a small increase in size, efficiency gains and system-level optimization deliver longer playback time.New sensorsA notable addition in the AirPods Pro 3 is the photoplethysmography (PPG) sensor on the outer surface of the earbud, which detects changes in blood flow to estimate heart rate. The system uses an infrared LED to illuminate blood vessels and a photodetector to measure the reflected light, allowing it to infer blood-flow patterns. In the CT scan, the LED can be identified by the wirebond connected to its semiconductor die, with the photodetector positioned below it, together forming a reflective optical sensing path.A new blood-flow sensor integrates an infrared LED and photodetector on the outer surface of the earbud, enabling reflective optical measurements for future health-tracking features.Aside from these changes, much of the internal architecture of the earbuds such as the printed antenna, touch sensor, and the location of the three MEMS microphones remains familiar, consistent with Apple’s strategy of incremental optimization. Charging caseMoving on to the charging case, we encounter a counterintuitive change: total battery life drops from 30 hours (AirPods Pro 2) to 24 hours (AirPods Pro 3). CT imaging reveals why. In the AirPods Pro 2, Apple used two batteries inside the case; in the AirPods Pro 3, this has been reduced to a single cell.Our scans suggest that this is a packaging-driven tradeoff as the new earbud geometry appears to force more of the case electronics onto a centralized main board, with the USB-C connector now soldered on top of the PCB instead of being recessed “in-line” through a cutout as in the Pro 2 case. This taller, more centralized board must now accommodate the Ultra-Wideband (UWB) chip, the wireless-charging power circuitry, and the USB-C power-delivery hardware. In other words, Apple seems to have prioritized packaging the new earbud form-factor at the expense of some charging-case capacity.The Pro 3 charging case consolidates to a single battery cell and moves the USB-C connector to the top of the PCB, trading some capacity for a slimmer, more integrated design.Ear tipsThe ear tip packaging is entirely paper-based, using laminated stacks of paperboard to build up the height of the posts that hold the tips. This replaces the vacuum-formed plastic used in earlier generations and reflects Apple’s broader shift toward sustainable, fiber-based packaging. The laminated construction is clearly visible in CT, where multiple layers of paperboard form the structural core that holds the ear-tip.The ear tips themselves introduce a new silicone-foam hybrid construction. CT imaging suggests that the “foam” Apple describes is not a separate material but rather a micro-cellular version of the same silicone, filled with tiny gas bubbles. One possible way of producing this could be through a two-shot injection molding process: the mold is first partially filled with standard silicone, then rotated for a second injection containing silicone mixed with a blowing agent. During curing, the blowing agent releases gas, forming a soft, foamed interior while the outer silicone skin remains dense and smooth where it contacts the mold walls.Inside each tip, a rigid plastic armature provides the mechanical snap-fit interface that attaches the tip securely to the earbud and supports a mesh that prevents debris from entering the sound channelCT scans of the redesigned ear tips reveal laminated paperboard packaging and hybrid silicone-foam construction, balancing comfort, acoustic performance, and sustainability.You can explore the full scans in Voyager, our web-based industrial CT viewer. If you’re evaluating CT for your own workflows, we’re happy to help.

Recall Radar

Defective On Arrival

March was heavy with consequential recalls. Medical devices in particular stand out, with high-stakes recalls across surgical tools, and impacted products in drug delivery and wound care. But the rest of the list isn't light either. Other recalls this month encompass structural failures in life-safety equipment, fire hazards in children's products, and a gas connector defect affecting nearly 200,000 homes. Across categories, the common thread is a product that reached a consumer or a patient before the problem that defined it was found. Faulty feedback Surgical recalls are always serious and can completely destroy a medical device manufacturer’s brand. Erbe USA's flexible cryoprobe recall traces to insufficient adhesive applied during assembly, which allows pressure to build until the probe ruptures during activation. Ruptures have caused burns and hearing damage to patients and nearby staff, with five serious injuries reported as of February 24. Medline is expanding an earlier recall of reprocessed electrophysiology and ultrasound catheters where certain lots may carry residual particulates on patient-contacting surfaces. The reprocessing step meant to make these devices safe for reuse introduced contamination, and once those particulates reach circulating blood they can trigger thrombus formation or granulomatous reactions. Integra LifeSciences' Codman Microsensor and Cerelink ICP Sensor kits were shipped with potential corrosion on the included Tuohy needle. The needle is used to access intracranial space for pressure monitoring, and the defect raises the risk of inflammation and infection at a site where either is serious. One injury had been reported. Leaking lines Everyday medical devices have also seen a recall this month. Certain Insulet Omnipod 5 Pod lots contain a small tear in the internal insulin delivery tubing. This defect is small enough that nothing about the pod's external appearance or initial behavior signals a problem. However, as a result insulin leaks inside the device rather than reaching the patient, and the user may not know anything has gone wrong until glucose levels have already climbed to dangerous levels or a hazard alarm finally triggers. In the worst cases, the outcome is diabetic ketoacidosis, a condition that can escalate quickly if insulin delivery has silently failed for hours. Eighteen serious injuries had been reported as of March 12. Broken barriers Integra LifeSciences appears twice more this month, with separate recalls covering MediHoney Wound and Burn products and certain CVS Wound Gel lots. 14 serious injuries have been reported combined, both stemming from compromised sterile barriers. The sterile barrier is the critical mechanism by which the product is safe to apply to broken skin. When it fails, the product actively introduces infection risk to a wound that already has one. The MediHoney recall covers all lots across multiple product lines, not isolated batches. A scope that wide doesn't point to a bad day on the line, but rather to a process that was consistently producing products that shouldn't have shipped. Avoidable accidents The automotive recalls this month are less severe than the medical ones, but they cover a wide range of failure types. Mercedes-Benz is recalling certain 2026 GLA 250 vehicles where a Bowden cable was incorrectly installed, preventing the door from being opened from inside. Kia is recalling certain 2027 Telluride Hybrids where second-row power seats can fail to detect an occupant and continue moving after contact, with children specifically flagged at higher risk. Vision Wheel's recall spans five aftermarket aluminum wheel part numbers where material inconsistencies can cause the wheel to crack on road impact, an unacceptable structural failure in a component that exists entirely to absorb that kind of load. Pebble Mobility is recalling certain 2026 Flow trailers where the adhesive securing roof-mounted solar panels may fail, posing a hazard to other drivers. None of these require unusual circumstances to trigger–a door opened from inside, a child in a back seat, a wheel hitting a pothole are all normal conditions these products were sold into. Escaping energy Two household product recalls this month involve out-of-control heat. The Wagner 900 Series Power Steamer recall covers roughly 700,000 units across three models, where the hose can overheat and the nozzle can expel water after the trigger is released. More than 50 burn injuries have been reported, including first- and second-degree burns to hands, arms, feet, and face. The Navajo Manufacturing heating pad recall is a simpler device with the same basic problem: when the pad is folded during use, heat concentrates in the fold to hazardous levels, and the device has no mechanism to detect this nor an automatic shutoff. Four serious injuries had been reported. The steam cleaner was producing steam and the heating pad was producing heat, but the failure in both cases was that neither could keep it where it belonged. Two more recalls this month carry fire hazards that were present from the moment the products shipped. The smaller of the two is the FUNTOK 24V 2-Seater Ride-On Truck, where the circuit board can overheat and ignite across roughly 1,980 units. The larger is the DuraTrac stainless steel gas connector recall at 196,800 units, where a manufacturing defect can lead to a gas leak. These are two very different products in different contexts, but both put the people nearest to them at risk without any warning that something is wrong. Fatigue failures Specialized is recalling approximately 5,720 Turbo Como SL electric bicycle forks sold between 2021 and early 2026, where the steerer tube is at risk of fatigue cracking. These aren't new products sitting in a warehouse. Some have years of accumulated riding loads, and fatigue cracking is precisely the failure mode that accumulates quietly over time before becoming visible all at once. So far, one failure has been reported. The Tainoki office chair recall covers roughly 2,200 chairs where the five-star base can bend under ordinary sitting loads, which is a shorter timeline to failure but the same basic problem. A structural component wasn't validated against what the product actually experiences in use. Petzl's recall of approximately 4,200 Nomic and Ergonomic ice climbing axes, with another 1,160 units sold in Canada, involves the shaft breaking at the handle during use. The serial number on each axe identifies whether it's affected. The stakes are different from the bicycle fork or the office chair. In ice climbing the axe is a primary anchor, and a shaft failure mid-climb is a fall. Petzl is offering a repair remedy. Like the bicycle fork and the office chair, this is a structural component that failed under the load it was built to carry. Problematic protection ProRider Economy and BMX helmets and Aisstxoer adult bike helmets account for a combined 9,746 units that fail mandatory standards for impact attenuation, positional stability, and certification. Neither recall had associated injuries at the time of filing, which is the best-case version of this kind of failure, caught before someone found out the hard way. A helmet that doesn't meet these standards doesn't offer reduced protection. It offers the appearance of protection, which may be worse than nothing if it changes how a rider behaves. Takeaways What connects March's recalls across categories isn't the product type or the industry. It's that in nearly every case, the problem was already present when the product left the factory. For the medical device recalls, that meant defects forming during manufacturing that inspection processes weren't designed to find, with harm that only became visible at the point of use. For the automotive and consumer product recalls, it meant verification processes that didn't cover the conditions that actually mattered. Some of those problems were found quickly. Others accumulated for years before a failure made them visible. A few only became known because a patient was harmed. The list looks varied on the surface, but the origin story is usually the same.

The Quality Gap

Do Water Filters Actually Work?

Most people replace their water filters according to the manufacturer's schedule, or when they remember to, or when the indicator light turns red. Very few replace them because they have seen what happens inside a filter over time. That gap between recommended practice and informed decision is exactly what these scans address. We CT scanned four types of water filters, a Brita pitcher filter, a refrigerator filter, a reverse osmosis membrane, and a LifeStraw, before and after extended use. The before-and-after comparison shows something that no filter packaging communicates: what a filter actually looks like when it is working, and what it looks like when it is done. What CT shows that a filter test can't Filter certifications from NSF International are the standard most consumers encounter on packaging. NSF/ANSI 53 covers health-related contaminants like lead, cysts, and volatile organic compounds. NSF/ANSI 58 covers reverse osmosis systems. NSF/ANSI 42 covers aesthetic improvements like chlorine taste and odor. These certifications test performance at the beginning of a filter's rated life under controlled conditions. What they don't show is what happens inside a filter as it accumulates contaminants over weeks and months of real-world use. CT scanning fills that gap. By rotating the X-ray source around the object and reconstructing a three-dimensional density map, it produces a model you can slice in any plane and analyze for both geometry and material density. A denser area means more mass in that location, and in a water filter, more mass means more accumulated contaminants. Pitcher filters: the flow channel problem A new Brita filter in CT shows a porous, uniform core of activated carbon and ion-exchange resin. The activated carbon traps chlorine, sediment, and some heavy metals through adsorption: contaminants adhere to the enormous surface area of the carbon particles as water passes through. The ion-exchange resin works differently, attracting specific ions and substituting less harmful ones in their place. The used filter tells a different story. Flow channels have formed through the media, defined paths where water has repeatedly followed the same route through the cartridge. That preferential flow matters: water moving through established channels bypasses the full filter media, reducing contact time with the carbon and limiting how thoroughly contaminants get captured. Denser particles have also migrated toward the outer edges of the core, visible evidence of accumulated sediment strained from the water supply over time. Lumafield's inclusion analysis tool isolates the densest material in the scan and highlights it as distinct specks against the surrounding media. The filter's accumulated load is visible. So is the fact that its remaining capacity is uneven across the cartridge. Refrigerator filters: the density reversal Refrigerator filters use a dense activated carbon core enclosed in a plastic housing, connected directly to a water line. They filter continuously rather than in batches, and most manufacturers recommend replacement every six months. The new filter scan shows the plastic casing as the densest element, with the carbon core registering as less dense and more porous. In the used filter, left in service for more than two years, that relationship has reversed. The carbon core has become markedly denser than the outer shell as it absorbed contaminants day after day. The filter media has changed its material properties in a way that is entirely invisible from the outside. The water's flow pattern is also visible in the used filter: water enters along the outer edges of the cartridge, passes through the carbon medium toward a central channel, and exits toward the ice maker. That flow geometry is embedded in the structure of the used media in a way that is absent in the new one. Reverse osmosis: scaling under pressure Reverse osmosis filters operate on a different principle than activated carbon. A semi-permeable membrane, wound in tight spiral layers, forces water molecules through microscopic pores while blocking dissolved contaminants including salts, minerals, and many chemicals. The new scan looks almost like a lithium-ion battery cross-section: clean, regular, tightly wound layers surrounding a central core, with clear pathways for water to move through the membrane structure. We ran this filter for two months on a hard water supply high in mineral content, alternating 12 hours on and 12 hours off, then scanned it again. The used membrane shows visible scaling deposits throughout the wound layers. Minerals that precipitated out of solution as water passed through the membrane have accumulated in the gaps between layers, partially blocking the pores and restricting flow. The geometry that was open and regular in the new scan is interrupted and dense in the used one. This is what reduced filtration efficiency looks like from the inside. Reverse osmosis is also the most effective filter type for PFAS, the so-called forever chemicals that the EPA established its first national drinking water standard for in April 2024, removing 95 to 99 percent of PFAS across chain lengths where activated carbon performance is more variable. LifeStraw: clogging at the inlet The LifeStraw uses hollow fiber membranes rather than activated carbon or a semi-permeable membrane. Microscopic pores in the fiber walls physically block bacteria, parasites, and microplastics while allowing water molecules through. It requires no electricity and no installation, which makes it the filter of choice for field use, emergency preparedness, and travel in areas where water quality is unreliable. The new LifeStraw scan shows open, uniform hollow fibers throughout the membrane. The used scan shows the degradation pattern specific to this filter type: clogging begins at the inlet end, where water and its suspended contaminants first contact the membrane, and progresses inward. The fibers near the inlet are visibly more restricted than those further along the flow path. That progressive clogging explains why backflushing, blowing air through the filter from the outlet end, is the recommended maintenance step for restoring flow rate. It physically dislodges the accumulated debris from the inlet fibers. Recommendations exist for a reason NSF certifications test performance at the start of a filter's rated life under laboratory conditions. What the scans show is that real-world performance depends on what is actually in your water, how much water has run through the filter, and whether the specific degradation mode of that filter type has begun to compromise effectiveness. The refrigerator filter left in service for two years had clearly exceeded its useful life in ways that were invisible from the outside. The reverse osmosis membrane showed mineral scaling that restricts flow before the membrane's rated life was reached. The Brita filter showed flow channeling that limits contact with the full filter media. None of these conditions are detectable by looking at the filter, smelling the water, or waiting for an indicator light. The standard recommendation, replace filters on schedule, is the right advice for exactly this reason. The scans show what is at stake when you don't.

Design to Reality

Eight Years to Redesign a Ketchup Cap

The Heinz ketchup cap has worked the same way for decades. A silicone valve inside a polypropylene shell allows ketchup to flow at a predictable rate when you’re squeezing the bottle and stop cleanly when you’re not. The silicone is the reason that mechanism works. It’s also why the cap can’t be recycled. Silicone and polypropylene have to be separated before either can be processed, and separating them from a small, complex cap at scale is not economically viable for recycling facilities. In the United States, there is effectively one silicone recycling plant. That leaves these caps in the landfill. Heinz decided to change that. What followed was eight years of development, 45 prototypes, more than 185,000 hours of engineering work, and $1.2 million in investment, in partnership with Berry Global (now Amcor). The goal was a cap made entirely from one material that dispensed ketchup the same way the old one did. The constraint turned out to be harder than it looked. Why silicone is hard to replace The original cap's silicone valve does two things at once: it flexes to allow flow under squeeze pressure, and it closes passively when pressure is released, preventing drips. The geometry is simple, with the material doing most of the work. Polypropylene is stiffer, less elastic, and does not self-seal the same way. You can’t substitute one for the other and keep the same cap geometry. You have to rethink the cap entirely. The scans make the difference visible. In our density mapping, the original cap appears in two distinct colors: the polypropylene shell in one range, the silicone valve in another, denser range. The new cap is uniform throughout. Same material, top to bottom. From valve to chambers The new design replaces passive silicone closure with geometry. The cap uses a series of internal channels and an antechamber that route ketchup on an indirect path to the nozzle. Squeezing the bottle forces ketchup through the channels, where it builds pressure against the antechamber wall before passing through the nozzle. When squeezing stops, the higher viscosity of ketchup at rest, combined with the indirect path, prevents the kind of drip the silicone valve used to stop mechanically. Ketchup's shear-thinning properties matter here: it becomes less viscous under applied force and more viscous at rest. The new cap is designed to lean into that behavior rather than override it with a mechanical seal. The channel geometry took 45 iterations to get right because changing any one dimension affects all the others. Tighten the channel, and the cap resists flow. Widen it, and you lose controlled dosing. The antechamber volume determines how much pressure builds before dispense. Every variable is coupled, and there’s no way around that fact. Inside the new cap CT shows us the internal structure of both caps without disassembly. The old cap's valve sits at the center of the nozzle, clearly distinct in density from the surrounding plastic. The new cap shows no such discontinuity: a single-material part with complex internal geometry that cannot be seen from the outside. The channel walls are thin, uniform, and precisely positioned relative to the nozzle exit. The kind of dimensional variability that would cause inconsistent dosing or leaks is exactly what CT is suited to detect during development. The scope of the problem The new cap launched in the UK and Europe in late 2023, where it replaced approximately 300 million non-recyclable caps per year on 400ml and larger bottles. A global rollout had not been announced at the time of publication. The cap is made entirely of polypropylene, which is accepted by any facility that processes plastic number 5 and is compatible with standard curbside collection in markets where that infrastructure exists. The development cost and timeline reflect something true about sustainable packaging that is easy to understate: recyclable materials do not perform identically to the materials they replace, and the gap has to be closed in both the design and the manufacturing. A silicone valve and a polypropylene channel array offer starkly different engineering solutions. Heinz’s new approach goes back to first principles, and builds a novel solution from there. Getting to the finish line required the kind of iterative physical testing that is expensive in time and material, and where CT can compress cycles by showing what is happening inside a prototype without destroying it.

Recall Radar

Fall on Fire

October runs hot. Pocketable gadgets smolder, wall-powered gear misbehaves where doors, ducts, and consoles meet heat, and vehicles remind us how small omissions ripple into busy spaces. A few welds and joints call it quits under real loads, while kid gear and clinical tools expose the quiet failures of access control and visibility. None of this is exotic. It is energy looking for an exit and interfaces aging faster than their spec sheets. Read the month as a tour of thin margins, and you will see why the headlines keep catching fire. Lithium-Ion batteries linger on the list Each month, portable lithium-ion batteries are inevitably recalled. ESR’s HaloLock wireless power banks, Zyntony’s Kogalla power banks, Anker power banks, Living Glow portable waist fans, IcyBreeze Buddy portable misting fans, and Arizer Solo II portable vaporizers were all recalled for fire or burn hazards tied to lithium-ion batteries. EcoFlow’s Delta Max 2000 power stations were recalled for overheating and ignition risk. These recalls are a testament to the sheer volume of lithium-ion battery powered devices that are sold on a daily basis. The Kogalla power banks represent the low end, with 2,400 devices impacted. In the middle, EcoFlow’s Delta Max 2000 lands near 25,030 and Living Glow’s portable waist fans sit around 48,000. Anker’s power bank recall, which encompasses five models, spans a whopping 481,000 units. x The concentration of portable products with overheating language signals a category risk where compact packaging, varied charge profiles, and real-world usage stack the deck against thermal margins. These parts might pass inspection initially, but for devices that live in pockets, bags, and hot rooms, reliability is key. One more entry belongs here even though it plugs in. iFIT’s NordicTrack rowing machines report screen console electronics that can overheat and ignite, about 44,800 units in the United States and about 700 in Canada. It is a reminder that small, insulated electronics mounted on larger mains-powered products can face the same overheating risks as portable devices. Seams under stress Where heat and load meet, weak links show themselves. Sunbeam’s Oster French Door countertop ovens carry a burn hazard when doors can close unexpectedly, with about 1,290,000 units in the United States and about 104,195 in Canada affected. VESTA.DS VST tankless water heaters can crack at the exhaust duct and release gases, including carbon monoxide, indoors, about 36,700 units in the United States, plus 3,500 Canadian sales. RH’s Byron tables and desks can collapse when a gap opens at the leg to top interface, about 750 units, and Super Wheels’ Vaast A/1 bicycle frames can fracture near a weld, about 1,860 units. Ordinary parts can carry extraordinary consequences. Doors that drift, ducts that fatigue, and welds that see side loads become the story once heat and cycling pile up. Whether the tally is a million units or a few hundred, the same pattern appears: failures concentrate at thermal interfaces and load-bearing joints. This is familiar physics at work, with thin margins revealed by use, time, and temperature. Hazards at any speed Here the risks meet bystanders. Textron E-Z-GO personal transportation vehicles can leak gasoline at a quick-connect near the injector, about 90,800 units in the United States plus Canadian sales. Yamaha fields two golf car campaigns, one for missing stop lights in about 19,300 vehicles from model years 2021 to 2025, and one for potential brake failure in about 4,300 vehicles from model years 2017 to 2024. Wilteexs bioethanol fuel bottles lack flame-mitigation devices and carry “Non Toxic” claims, about 1,100 units, raising flash-fire risk. Different mechanics, same exposure: small misses in connectors and signals become public hazards in parking lots, paths, and garages. These are ordinary environments, which makes them unforgiving. Parking lots, paths, and garages leave little separation between a small defect and a person. People read cues by habit; they assume brake lights exist and fuel stays contained. When those assumptions fail, risk extends beyond the user to everyone nearby. Small configuration splits across trims, suppliers, or model years can escape notice until they show up in the field. Fuel containers sit under the same quiet expectation, and without flame mitigation a routine pour can turn into a flash event. Small batteries, big risks Two children’s products turn on a single access point. About 740 LED tutu skirts sold by Bmrwtg allow easy access to coin cells and lack the warnings required by Reese’s Law. About 3,000 Youbeien crib mobiles have remotes whose battery doors can be opened without a common household tool. The scale is modest compared with appliance recalls, but the hazard is concentrated and severe. Small parts compliance is binary, and real-world handling includes repeated openings, flex, and pry attempts by curious kids. Battery enclosures and labeling must be treated as safety-critical, with lot-to-lot checks and post-use fit tests. Patient safety at stake Two medical device recalls sit squarely on reliability at the point of care. On September 22, 2025, Olympus flagged ViziShot 2 FLEX 19G EBUS-TBNA needles for components that may detach during procedures. On October 7, 2025, Trividia Health recalled TRUE METRIX blood glucose meters for LCD defects that can affect performance. Unit counts are not stated here, but both products inform clinical decisions in real time, which raises the consequence of even limited populations. In practical terms, devices must keep their components secure and present clear, trustworthy readings. When either falters, clinicians lose signal or confidence at the moment they need it most. Patterns worth testing Look across October and the same physics keep showing up. Thermal headroom and honest derating decide whether packs and console boards stay uneventful. Containment and vent paths determine if fuel and exhaust stay where they belong. Doors, latches, ducts, and weld toes carry more safety load than their bill of materials suggests. On low-speed vehicles, stop lights and braking are the guardrails for shared space. In children’s products, safety turns on controlled access to coin cells and correct warnings. In clinical tools, it depends on parts that stay attached and displays that stay readable. The misses rarely sit in the marquee component. They live in a hinge that lost hold-open force, a quick-connect that seeped under vibration, an exhaust duct that thinned with heat, a display interconnect that drifted, a battery door that loosened after use, a weld toe that carried side load. Treat this month as a map of where engineering time tends to matter most: thermal interfaces, joints under side load, basic vehicle signaling, and battery access. Use it to prioritize test time and supplier controls in the next build cycle. Curious how I keep track of this constant barrage of recalls? Check out our new Recall Roundup tool to access a comprehensive database of product recalls, updated daily from CPSC and NHTSA records. We also discuss recalls in every episode of our podcast, Go/No-Go.

From The Floor

Finding Lead in Stanley's Quencher

Social media influencers discovered lead in Stanley's Quencher tumbler in early 2024, setting off a consumer panic that reached far beyond the usual product controversy. Home lead testing kits went viral. Stanley confirmed its cups contained lead. But some critical questions were going unasked. Where exactly is it? How did it get there? Does it actually reach the person drinking from the cup? We put a Quencher in our Neptune CT scanner to find out. [embed: full scan] How it’s made The Quencher's insulation works the way all vacuum-insulated drinkware works: two layers of stainless steel, inner and outer, separated by a vacuum that slows heat transfer. The manufacturing process starts by forming those two layers separately, then welding them together at the rim to create a sealed unit. At that point, the cup goes into a vacuum chamber, and air is pulled out through a small hole left in the bottom of the outer layer. Once the vacuum is established, that hole has to be sealed before the chamber opens. The method Stanley and most of the industry uses is a small lead pellet placed above the hole during assembly. The vacuum chamber heats until the lead melts, flows into the hole, and seals it. When the chamber opens, the vacuum between the walls is locked in. [embed: bottom detail] Why lead Lead is an ideal process material for this application. Its melting point is low and predictable, its flow characteristics under heat are consistent, and the seal it forms is reliable. The same properties that made it a long-running standard in electronics soldering make it useful here. Unleaded alternatives exist, including noncrystalline silica beads and proprietary sealants used by brands like Owala, Hydro Flask, and Klean Kanteen, but switching requires process development and carries higher risk of defects during the transition. It is a genuine manufacturing tradeoff: a well-understood material with known risks, versus a less-characterized one with unknown process behavior at scale. What we found The CT scan makes the geometry clear. In Lumafield's density mapping, denser materials absorb more X-rays and appear in the orange-to-red range. The lead solder appears as a solid red mass at the base of the cup, completely enclosed between the outer stainless steel layer and a steel disc that caps it from below. Above it, the inner wall. Below it, the cap. The lead has no path to the cup's interior under normal conditions. [image: CT cross-section, base of cup] Stanley's statement is consistent with what the scan shows: "Once sealed, this area is covered with a durable stainless steel layer, making it inaccessible to consumers. Rest assured that no lead is present on the surface of any Stanley product that comes into contact with the consumer nor the contents of the product." The company also noted that if the base cap comes off through ordinary use and exposes the seal, the product is covered under its lifetime warranty. Experts who spoke to NBC's Today program put the risk plainly: the odds of meaningful lead exposure from an intact cup are negligible. A class action lawsuit filed in early 2024 alleged that Stanley failed to disclose the lead's presence; a federal judge dismissed the consolidated complaint in January 2025, finding that plaintiffs had not shown sufficient harm from the specific amount of lead in the cups. An amended complaint has since been filed. An inescapable tradeoff What the Stanley situation actually illustrates is how common this kind of engineering compromise is and how rarely it gets explained to consumers. Lead solder in a sealed, inaccessible location in a consumer product is a different risk calculus than lead paint on a toy or lead in drinking water pipes. The scan shows the geometry. The geometry shows the containment. Whether that containment is sufficient is a question the courts are still working through, but the manufacturing logic behind the decision is not mysterious. Most products that live in kitchens and garages contain materials that would be alarming in other contexts. CT imaging is one of the few ways to show exactly where those materials are and what separates them from the person using the product. This study was covered by Fast Company, Business Insider, and The Verge.

Design to Reality

Furbo and KONG: Two Ways to Give a Dog a Treat

There are two ways to give a dog a treat. You can be there to hand it over, or you can engineer something that does it for you. The Furbo and the KONG represent opposite ends of that spectrum, one a connected device with motors and a camera, the other a piece of molded rubber that has barely changed since 1976. CT scanning both of them reveals how much engineering goes into making something look effortless. The Furbo: three motors and a pinwheel [Furbo embed] The Furbo's job is to find your dog, aim at it, and launch a treat across a room. The CT scan shows how it does all three. The base contains a motor dedicated to rotation, which gives the camera a 360-degree field of view. That motor doesn't connect to the PCB directly above it; that board handles the USB-C power connection. The speaker is here too, facing downward behind a plastic mesh on the Furbo's lower edge, positioned to project sound toward floor level where the dog actually is. [Image: Furbo internal layout] The main PCB sits higher in the device, connecting to the camera and microphone assembly via offset boards angled to match the placement of the camera lens in the Furbo's hourglass housing. Two more DC motors handle treat delivery: one turns a pinwheel that loads treats from the hopper into the launcher one at a time, and a second drives a paddle that fires them out of the launch chute. This is a meaningful design change from an earlier spring-loaded mechanism that was prone to jamming. The two-motor system separates the loading and firing functions, which is cleaner mechanically and more reliable in practice. The KONG: from VW bus to dog toy [KONG empty embed] [KONG with treats embed] Joe Markham was an auto shop owner in Denver in the 1970s. His dog Fritz became obsessed with a rubber bump stop from a 1960 Volkswagen Type 2 Bus, the kind of component that cushions the suspension against extreme compression. Fritz chewed it for hours without damaging his teeth or the rubber. Markham spent six years refining the material, shape, and internal cavity, and by 1976 had the KONG. We scanned both the original VW part and the KONG to see what Markham changed. The comparison is direct. The bump stop's internal cavity is narrow and tapered, not centered with the outer rubber: wall thickness varies by nearly 3 mm at the same point in the circumference. The rubber itself shows visible density variation, with pockets of inconsistency throughout. For a suspension component absorbing road impacts, this is probably fine. For a dog toy that needs to dispense treats and survive aggressive chewing across millions of units, it is not. [Image: KONG vs. bump stop comparison] The KONG's internal chamber is wider, more cylindrical, and open on both ends for easier treat loading and cleaning. Wall thickness varies by only about 0.3 mm, one-tenth the variation of the VW part. The rubber density is consistent throughout. Whether this reflects sixty years of manufacturing advancement or simply that Markham was making a better product to tighter tolerances, the scans make the difference visible. Two devices, one problem Neither product looks like much from the outside. The Furbo uses three motors, two PCBs, a camera, a microphone, and a speaker to deliver a treat on command from anywhere with a cell signal. The KONG uses a hollow piece of rubber to turn a treat into an engineering challenge the dog has to solve. The CT scans don't reveal a winner. They reveal how much deliberation goes into both, and how differently two products can find a way to reward man’s best friend.

Recall Radar

Hidden Failures of Everyday Interfaces

The new year has kicked off with a wide swathe of recalls. Compact electronics carried more energy than their housings could tolerate. Ceiling hardware relied on connections that shifted dangerously under real installs. Helmets and pedals failed at the exact points meant to carry impact and torque. In hospitals, small discontinuities in breathing circuits and needles became patient events. Across categories, these recalls remind us that quality must run through the whole product development and ramp cycle, from first drawings to production proof. Compact, closed, and cooking Hazardous batteries have shipped to consumers yet again. Energizer’s rechargeable lanterns can overheat inside a sealed handheld body, a small item with a field count of about 4,100. TJX’s Isla Rae magnetic wireless chargers raise the same flag at a larger scale, with about 13,200 sold in the U.S. and 7,000 in Canada. On the wiring side, Curtis International’s expanded recall of Frigidaire-brand minifridges due to an internal short circuit that can ignite the plastic housing has been expanded to include an additional, 330,000 units, on top of the 634,000 previously recalled in 2024. Two power strip lines failed at fundamentals. HEZI’s metal enclosure lacks a ground path across roughly 1,320 strips. ANNQUAN power strips omit supplementary overcurrent protection at a scale of about 11,200. The common cause across these electronics recalls is unmanaged fault energy in tight spaces, where design must decide where heat and current go during real failures. Childproofing challenges Three beard serums from RootStim (~16,900), Ruahouine (~25,000), and Feel The Beard (~840) contain minoxidil yet shipped without child-resistant packaging, and Plantimex’s Mamisan lidocaine jars show the same gap across roughly 50,330 containers. Under the Poison Prevention Packaging Act, packaging is defined by the active ingredient, which means closures must resist repeated child-opening attempts and continue to perform after real handling and wear. In these cases, the packaging did not meet that requirement, turning a labeling problem into an access problem that puts children at risk. Backyard breakage Three very different products share the same setting and the same test of endurance. DR Power’s leaf vacuums continue to surface material ejection during use, with pieces breaking loose inside the impeller path or debris piercing the chute. The campaign now spans about 60,250 units in the U.S. plus 1,023 in Canada, an expansion from a larger family first flagged in 2024. The campaign now spans about 60,250 units in the U.S. and 1,023 in Canada, an expansion of the 2024 recall. The failure concentrates at the chute, where repeated hits in one spot thin the wall, start cracks, and open a path for fragments. When that happens, chips leave the machine at speed toward the operator. It is a durability problem at a single hot spot. Across the yard, patio seating failed under routine loads. Adams Manufacturing’s Adirondack chairs can crack and collapse across roughly 6,100 units, and Academy Sports + Outdoors’ Magellan Odyssey Rocker chairs have legs that can break on about 35,300 units. The stress lives at leg and seat interfaces that see point loads, twisting, and outdoor exposure. UV and cold change how plastic and fasteners behave, and rocker motions drive cyclic bending at the same locations. If prototypes and production parts are not tested in that combined environment, the rib that looks stout on paper becomes the crack starter in the field. Faulty fixtures overhead Two chandelier families failed at the ceiling interface. RH’s Natural Antler units can detach when installed on vaulted or sloped ceilings, a rare mode but meaningfully dangerous at ~320 units. Currey & Company’s Electra chandeliers use a connection component with improper threads across about 260 fixtures, which can let the assembly drop. Real ceilings are rarely level, which shifts weight across the hardware, and thread form decides how much metal is truly engaged. In both cases, a small part at the canopy sets the outcome. Questionable cycling PNW Components’ Loam Pedal Gen 2 uses an axle that can crack and release the pedal from the crank, a focused field count near 1,200 that still matters because failure happens at speed. Pedego’s Fat Tire Trikes can develop a hairline fracture near a weld that progresses to tube separation across ~400 units. R.X.Y bicycle helmets fail the mandatory standard on impact attenuation, positional stability, and required labeling and certification for about 170 helmets. These are classic weak points. Axles concentrate stress at thread roots, weld toes gather cyclic strain, and helmets that do not hold position or manage impact energy leave the head underprotected. Fatigue testing, notch control, and retention checks are needed to prevent ordinary rides from turning into failure demonstrations. Risks on the road Several RV campaigns point to simple electrical controls that did not hold up. Winnebago and Newmar found water tank heating pads that can develop high resistance and fail. Forest River’s Freelander routed a solar panel lead to a breaker without proper over-current protection. Gulf Stream Coach reversed wires in a taillight harness, so a turn signal can point the wrong way. These are the kinds of issues that routine polarity checks, protection-device validation, and functional lamp tests should catch before a vehicle ships. Ford reported engine block heaters that can crack and leak coolant, then short when the heater is plugged in. The behavior appears during use, which puts the focus on thermal cycling and material choice in the heater assembly. Wanli Tire Corporation recalled Aptany Eco Sendero M/T2 tires in size LT265/75R16, a fitment common to Toyota Tacoma, for sidewalls that can separate and fail to meet FMVSS 139, with the risk of rapid air loss and poor handling. In each case the basics decide outcomes, from correct wiring and working fault cutoffs to sidewalls that stay bonded under load. Medical material mishaps Olympus broadened its action on ViziShot 2 FLEX 19G EBUS-TBNA needles after reports that parts of the needle assembly can eject or detach during use, including one death. The weak point sits at the sleeve–hypotube interface, which can break down and lead to fluid leakage, poor sample movement, or a loose fragment in the airway. Olympus traced contributors to heat-shrink degradation and use factors, which makes this a design-and-materials issue across all lots rather than a limited batch problem. Draeger found supplier residues inside certain Vapor 2000 and Vapor 3000 anesthetic vaporizers. Metal left after soldering can react with volatile fluorinated anesthetics and enter the patient circuit, with risks that include airway irritation, lung injury, and pulmonary edema. Fortunately, as of early December, no serious injuries or deaths had been reported. Meanwhile, Medline identified cracks and leaks in 120-inch expandable anesthesia circuit tubing used in some kits. A leak in that circuit can drop ventilation and dilute anesthetic delivery at the patient, and it can also vent anesthetic vapor into the room. The FDA has seven reports tied to oxygen desaturation from this issue. Across these cases, safety turns on small interfaces and material compatibility under temperature, motion, and anesthetic gases. When those details slip, patient care can change in seconds. Takeaways January ties one idea through very different products. Reliability is earned where everyday forces meet small parts. In compact electronics that means heat and fault current have a defined way to escape. Move to the ceiling and the same rule applies, with brackets and threads that keep their grip when the mount sits on a slope instead of a test stand. On bikes and trikes it shows up at axle roots and weld toes that take thousands of small hits instead of one big pull, and on the road it depends on wiring that matches the drawing every time with joints tightened to values you can prove. In clinical tools it becomes materials and seals that stay stable with the gases, temperatures, and motions seen in practice. Design for the conditions products live in, validate under those conditions during ramp, and keep watching as volume grows so routine use stays routine.

Design to Reality

How Does a Car Cigarette Lighter Work?

If you designed this product today, you would probably reach for a microcontroller. A temperature sensor, a timer circuit, a small actuator. The engineering would be straightforward, the components cheap, and the result predictable. But the car cigarette lighter was designed in 1956, when microcontrollers didn't exist, when the device needed to be manufacturable by the million at minimal cost, and when every function had to be handled by geometry and material properties alone. The result is one of the most elegant examples of analog engineering in everyday use. We put one in a CT scanner to show you how it works. What it has to do The requirements are simple to state and harder to solve than they appear. Push the lighter in, hold it against a spring, complete a 12V DC circuit, heat a coil to ignition temperature, then release the lighter automatically when it's ready without any timer, thermostat, or electronic control of any kind. Do all of this with stamped metal parts, a few springs, and a ceramic insulator. The solution Casco arrived at for their 1956 patent solved every one of those requirements without a single active component. Circuit path When the lighter is pressed into its socket, it snaps into a spring clip and closes a circuit. Current flows in through a bolt at the base of the assembly, which is isolated from the surrounding housing by a ceramic insulator. From there it travels through the arms of the spring clip, into the outer rim of the heating element, through the nickel-chromium coil itself, into a bolt at the coil's center, and back out through the housing to the negative terminal. The CT scan makes this path visible. Stripping away the plastic and isolating the metal components by density, you can trace the circuit through the assembly: the ceramic insulator at the base appears as a distinctly darker region against the surrounding steel, the coil sits at the center of the removable lighter, and the bolt that completes the circuit runs through it. What forces the current through the high-resistance coil rather than finding a shorter path? A thin strip of paper insulates the heating element shield from the rest of the assembly, leaving the center bolt as the only available path. The coil itself is oxidized, insulating each loop from its adjacent loops, which forces the current to travel the full length of the resistive filament rather than jumping across it. That resistance is how the coil generates heat. Pop-out mechanism The spring clip that holds the lighter in its depressed position does two things: it conducts current into the heating element, and it releases the lighter when the coil is hot. Its arms are bimetallic, with steel on the outer surface and copper on the inner surface. As the coil heats up, it also heats the arms of the clip. Copper expands faster than steel under heat, so the inner surface of each arm grows faster than the outer surface, pushing the arms apart. When they open far enough, they release the lighter, and the coil spring in the housing pushes the handle out. No sensor detected the temperature. No timer measured elapsed heating time. The bimetallic strip is calibrated such that the differential expansion of its two metals opens the clip at approximately the right moment, determined entirely by the geometry of the strip and the thermal properties of the materials. The engineering judgment that went into specifying those dimensions is embedded in the part itself. Engineering vestiges You can understand the cigarette lighter conceptually without a CT scan. The mechanism has been described in patents and engineering references for decades. What the scan adds is the spatial reality of the solution: how compact the assembly is, how the ceramic insulator is positioned relative to the current path, how the coil is bonded to its shield at one end and the center bolt at the other, and how little material is used to accomplish the full function. The coil spring, the bimetallic clip, the ceramic insulator, the oxidized nickel-chromium filament—none of these components do one thing. Each is doing two or three things simultaneously, because that is what constraint-driven design produces. When you cannot add a control system, the physics of your materials have to do the work instead. The socket outlasted the lighter by thirty years. It became the de facto standard for 12V DC power in vehicles, a connector standard that still exists in every car made today. The lighter itself is gone from most dashboards, but the engineering logic that produced it is not.

Scan of the Month

How Four Pens Solve the Same Problem

Writing is a fluid dynamics problem. Ink has to travel from a reservoir through a delivery mechanism and onto paper at a controlled rate, all without flooding, clogging, or running dry. Every pen design is an engineering answer to that problem, and the answers vary considerably. CT imaging makes it possible to examine those answers at the scale where the real decisions are made. We scanned four pens: a Majohn A1 fountain pen, a BIC ballpoint, a Pilot rollerball, and a Paper Mate Flair felt tip. What the scans reveal is not just how each pen is built, but why each design makes the tradeoffs it does. Majohn fountain pen: capillary action at work The fountain pen is the oldest design in this comparison, and in many ways the most mechanically elegant. Its core principle, capillary action, requires no moving parts and no active pressure. Surface tension and adhesion draw ink through a narrow channel from the reservoir to the tip without any external force. The scan shows the full assembly of the Majohn A1: the retractable body, the spring-loaded cover that protects the nib when closed, and the nib itself. In the grayscale detail view, we can measure the nib slit directly. At the base, it is 0.128 mm wide. At the tip, it narrows to 0.016 mm, roughly one-eighth of its width at the base. That taper is what regulates ink flow. A wider slit would flood the paper; a narrower one would choke it. The Majohn achieves that geometry at a fraction of the price of the Pilot Vanishing Point it resembles. BIC ballpoint: precision manufacturing at commodity scale The ballpoint pen solved a problem fountain pens couldn't: reliable writing on rough surfaces, in varying orientations, with ink that dries fast enough not to smear. The mechanism is a tungsten carbide ball seated in a brass socket, rotating freely as the pen moves across paper. The ball transfers oil-based ink from the reservoir to the writing surface with each rotation. The scan shows the ball clearly: denser than the surrounding brass housing and therefore appearing dark red here in the CT image. Also visible are the five ink channels that route ink toward the ball from the reservoir above. The assembly sequence is precise: the ball goes in through the tip opening, and the tip is then crimped inward to trap it, holding it in place while still allowing free rotation. The tolerances on that final crimp are extremely tight, on the order of microns, because the gap between ball and socket controls ink flow. Too tight and the pen skips; too loose and it makes a mess. Pilot rollerball: water-based ink, different constraints The rollerball uses water-based ink, which is thinner than ballpoint ink and flows more freely. That thinner viscosity produces a smoother writing experience with less pressure, but it also means the ink delivery system has to be more carefully controlled to prevent flooding. The scan shows the ball dropped into a stainless steel housing tube and crimped from three directions, a three-point crimp that holds the ball in place while leaving enough clearance for ink to reach it consistently. Like the fountain pen, the rollerball relies primarily on capillary action to deliver ink to the tip, with the low viscosity of the water-based ink making the flow more immediate and sensitive to orientation than a ballpoint. The trade-off is ink consumption: rollerballs run out faster, and water-based inks are more prone to bleeding through paper. Paper Mate Flair felt tip: the simplest mechanism, a complex material story The felt tip replaces machined metal components with a porous fibrous tip that both carries and regulates ink. There is no ball, no socket, no slit. Ink wicks through the fiber structure to the writing surface by capillary action, and the density and orientation of the fibers determine how much ink flows and how the line looks. The scan reveals something the surface can't: how the tip structure changes across decades of the same pen. We scanned Paper Mate Flair pens from 1966 through 2023, work we originally did for The New York Times Wirecutter after readers noticed the pen seemed to be running out of ink faster than before. The CT images show why. The 1966 Flair had a large, fully exposed felt tip with considerable fiber volume. By 1983, the tip had been reduced and a separate plastic point guard added. The 1989 version shows further changes to the collar construction. The 2023 version has a visibly smaller exposed tip surface than the original, with a tighter collar. No single redesign explains the change. The cumulative effect across fifty-seven years does. The fiber orientation is also visible in the scans, and with it the ink distribution through the tip material. Even in a design this simple, the fluid dynamics are real and the manufacturing decisions matter. What the comparison shows Each of these pens solves the ink delivery problem with a different mechanism: capillary action through a precision-tapered slit, a rotating ball in a crimped metal socket, a ball in a three-point crimp with low-viscosity ink, or a fibrous wick. The engineering ranges from sub-micron tolerances in a commodity product to the organic irregularity of a fiber matrix. What CT adds to the comparison is the ability to see those decisions directly, to measure the nib slit, observe the crimp geometry, and track changes in tip construction across decades without cutting anything open.

From The Floor

How Ground Truth Data Builds Trust Between OEMs and Suppliers

For all the advances in automation, AI, and robotics, manufacturing still runs on relationships. Behind every finished product is a network of suppliers, subcontractors, and integrators who depend on trust: trust that the part delivered matches the design, that materials meet specification, and that failures become opportunities for improvement rather than casting blame.But trust alone can’t keep pace with the scale and complexity of global supply chains. What manufacturers need now is a shared foundation of truth. They need data that both OEMs and suppliers can use to see exactly what was built and delivered.That foundation is visibility. By visibility data, we mean a shareable, high-resolution 3D inspection record: typically an industrial CT volume plus analysis outputs (CAD comparison, defect segmentation, dimensional results) packaged with metadata so both parties can review the same evidence.A common view changes everythingFor most of the history of manufacturing, suppliers and OEMs relied on separate inspection workflows and quality assumptions. When something went wrong, determining who was responsible took weeks of back-and-forth analysis, finger-pointing, and blame. Every step introduced more friction and eroded trust.Shared visibility of an as-built inspection record eliminates that uncertainty. Both sides can look at the same dataset and see how a manufactured part compares to its digital design. The information is objective, measurable, and mutually verifiable.This alignment builds speed and confidence. OEMs can qualify suppliers more quickly and diagnose problems earlier. Suppliers can prove quality, show compliance, and protect themselves when issues arise. The result is a faster, more transparent relationship built on facts.Shared visibility connects OEMs and suppliers through a single source of truth that strengthens trust, proves quality, and accelerates qualification.For OEMs: faster decisions and faster recoveryFor OEMs, every delay carries cost. Supplier qualification and root cause analysis both take time, and time lost to investigation often also translates to production downtime. Better data changes that equation. Qualification and RCA move from ‘send samples + wait’ to ‘review the same dataset within hours.’OEMs can:Verify vendors and build trust with measurable evidenceIdentify root causes quickly when issues appearAccelerate qualification of new suppliers to respond to market shiftsMake sourcing decisions based on consistent dataIn a global environment shaped by tariffs, pandemics, and shifting trade policies, the ability to pivot to a new supplier without compromising quality has become a strategic advantage. Shared visibility data is what makes that possible.For suppliers: proof, protection, and differentiationSuppliers face their own set of pressures. They’re expected to deliver faster, reduce costs, and meet increasingly detailed documentation standards. Visibility data helps them meet these demands while strengthening their reputation. Suppliers can:Prove and document quality to a higher standardDifferentiate through transparency and measurable precisionDemonstrate compliance and traceability for auditsProtect against blame when a problem originates elsewhereIn practice, this becomes a kind of insurance policy. Delivering visibility data with a shipment shows the customer exactly what was built and why it meets specification. It provides confidence beyond “take my word for it.”How alignment solves real problemsThe clearest examples of this value appear when something fails. We see this again and again with the quality programs we support.A consumer packaged goods manufacturer used visibility data to track down the source of a leaky bottle. The problem turned out not to be with the bottle closure vendor but with a misalignment on the OEM’s own line.An automotive team used industrial CT scans to compare a latch that failed in the field to an unused part from the same lot. The data showed that the defect originated upstream and gave both the OEM and vendor a path to prevent it in the future.In medical devices, a supplier missed a small radius on a metal component, which led to field failures. A single CAD Comparison during first article inspection would have caught the issue before it was too late.In each case, visibility allowed both OEM and supplier to act on the same evidence and reach the same conclusion. The outcome was collaboration instead of conflict.Comparison of an automotive connector’s CAD model and CT scan reveals and quantifies subtle deviations between design intent and manufactured reality.The foundation of modern manufacturing relationshipsAs products become more complex and supply networks more distributed, the relationships that support them must evolve. Shared visibility gives OEMs and suppliers a reliable point of reference based on the physical reality of what was actually produced.That clarity builds accountability, confidence, and speed. It helps suppliers strengthen their credibility while allowing OEMs to qualify partners more efficiently. Most importantly, it reduces uncertainty at a time when agility matters most.Visibility is becoming the common language of modern manufacturing. It allows both sides to move faster, with greater accuracy, and with a shared understanding of quality that scales across every link in the supply chain.

Design to Reality

How I Think About R&D (and Turning Ideas Into Products)

R&D in a startup is not a separate function. It’s an integral part of the product development process, especially in the early stages when requirements are uncertain, the solution is unclear, and the risk is high. At Lumafield, I’ve worked on everything from early prototypes to complete product features like offset scanning and calibration systems. Over time, I’ve learned that effective R&D is not about having a perfect process. It is about thinking clearly, experimenting deliberately, and staying grounded in first principles. Here’s how I approach the work. 1. Start with first principles When tackling a new technical problem, I begin by reducing it to the fundamentals: physics, geometry, mechanical constraints. Early on, you often cannot rely on precedent. You need to understand what is objectively true and work forward from there. This mindset helped us build scanning capabilities when we were still new to tomography. We didn’t import solutions from other companies. Instead, we reasoned through the challenges ourselves. This took more effort up front but allowed us to invent solutions that fit our specific needs and constraints. 2. Prototype to learn In the early stages of product development, the purpose of a prototype is not to impress. It is to test a hypothesis. A prototype should reduce uncertainty. Can we hit the resolution target? Will the calibration approach scale? Can we build this cost-effectively? Sometimes the result is positive. Sometimes not. Either way, it moves the work forward. The key is to be clear about what you are trying to prove and avoid refining the details too early. 3. Use literature to build smarter and faster We often begin with a literature review to understand the problem space. Research papers can clarify what has already been done, what techniques are available, and where current limitations exist. This helps frame our approach. There is no need to reinvent the wheel. When researchers have already solved part of a problem or developed a method that fits into our pipeline, we use it. Applying proven components can accelerate development and avoid unnecessary work. However, most published research is not optimized for product integration. Methods may assume ideal conditions, ignore cost constraints, or require infrastructure that is not practical in our environment. That is where adaptation comes in. We borrow selectively, simplify when possible, and focus on what works reliably within our system. Understanding the principles behind a solution is more valuable than reproducing every detail. Our goal is not to replicate research but to apply it thoughtfully to build products that perform in the real world. 4. Align R&D with product outcomes Our work is driven by product needs. Sometimes a request comes from the product team or a customer. Other times, we explore an internal idea that we believe could unlock value. In both cases, we scope experiments tightly. We aim to answer key questions within a few days or weeks. This helps us avoid chasing technically interesting ideas that do not translate into real product improvements. If a solution is imperfect but solves the problem in a useful way, that is often enough. 5. Focus on what customers actually need In many cases, product requirements aren’t fixed. Customers may request specific performance targets, but in practice their needs are more flexible. We’ve tested configurations that technically fall short of a stated requirement. Yet the results were good enough that the customer was satisfied, and even impressed. Part of product development is figuring out where precision matters and where it doesn’t. That doesn’t mean lowering the bar. It means delivering meaningful outcomes based on real constraints. At Lumafield, we’re making advanced analysis software so accessible that you can even access our 3D CT scans on your phone. 6. Collaborate across functions R&D is rarely isolated. We work closely with teams across the company—engineering, product, customer success—to connect our experiments to real problems. Sometimes we initiate new product ideas. Other times, we help refine or extend features that already exist. In both cases, collaboration is critical. It ensures we’re not solving problems in a vacuum and helps us integrate what we build into the broader system. At a startup, where responsibilities overlap, this cross-functional alignment becomes even more important. 7. Tie research to real impact Good R&D demonstrates what’s possible, but great R&D leads to measurable improvements. For example, I’ve spent the past two years developing a new calibration method. The goal was not theoretical accuracy. It was to improve every scanner in the field with a software-only update. This opened new use cases, improved customer outcomes, and increased the value of our existing products. It’s easy to get lost in technical complexity. Staying focused on real product outcomes keeps the work grounded. 8. Tools help, but mindset matters more We use a range of tools. Cursor has helped me write software more efficiently. Mechanical prototyping tools let us test physical ideas quickly. These tools matter. But they’re not the deciding factor. What really drives R&D is the right mindset. Are you curious? Are you willing to test assumptions? Do you stay focused on solving the right problem? A thoughtful, iterative mindset will take you further than any single piece of software or hardware. Final thought R&D in a startup requires balancing speed, rigor, and practicality. You’re working with uncertainty. You’re building things that may not work. But you’re also learning quickly and making decisions that shape the final product. I have found that the most effective approach is to start from first principles, define clear questions, and work in tight feedback loops. You don’t need to be certain about everything. You just need to make consistent progress toward something real. That mindset (not just technical ability) is what moves a product from concept to reality.

The Quality Gap

How People Drive Quality

When people think about quality in manufacturing, they often think of documentation like certifications, processes, and forms. That stuff definitely matters, but it’s not the core of it. What really drives quality is people: how they work, how they think, and whether the system around them helps them succeed or sets them up to fail. That’s the lens I brought to Lumafield after working in medical devices, where the bar for quality is high, and for good reason. What Medical Devices Taught Me About Quality Before Lumafield, I worked in a highly regulated environment under ISO 13485, which is the quality standard for medical device manufacturers. In that world, trusting that someone made a good product isn’t enough; you have to prove it, every step of the way. What’s interesting about medical device manufacturing is that it doesn’t control the certification of the individuals designing a product, like the certified PE (Professional Engineers) in civil engineering. Technically, anyone can sketch out a medical device. What the certification does control is how that product gets built and tested. That means heavy documentation: from ideation and risk analysis, to design verification and validation, to production release and traceability. Every requirement needs to be documented, and every product needs to be tested against those requirements, with a full paper trail to back it up. I spent a lot of time in that system, and while some parts felt rigid, it also taught me how robust a good quality system can be. Every step exists to build confidence, minimize risk, and create repeatability. That experience gave me a foundation for how I think about quality now, even in a startup where we don’t have formal certifications (yet). What Quality Looks Like at a Startup At Lumafield, we’ve had to figure out what quality means without that external structure. We go beyond the safety standards our product is certified for, despite it not being required. Because even without a certifying body breathing down our neck, we care about building reliable products, and we know that just “building it right the first time” doesn’t scale. So instead of a formal QMS, we rely on a culture where every individual with hands on our hardware cares about product quality, and we enable those individuals with practical implementations of the bread-and-butter tools of manufacturing: detailed work instructions, in-process checks, and a responsive NCR process. Our focus is always on improving quality where the action is happening, not somewhere buried deep in a stack of SOPs. We want our operators to have the tools and information they need to reliably build quality products safely. That sounds simple, but we’ve come a long way since we started. When I joined, there was no structure yet—no standard for how instructions were written, no consistent way to track build issues, no process for root cause analysis. We’ve built that all from the ground up. What We Catch, and When Ideally, we’d catch every problem before it gets to the floor. That’s the goal with things like first article inspections and incoming quality checks. But at this stage, with how lean our manufacturing team is, we’ve had to make strategic tradeoffs. We can’t inspect every part, so we need to take calculated risks on where it’s more efficient to catch issues later on the production line and where we need to spend the resources to catch issues upon receipt. It’s a burn rate we accept. We’ll pay a little more for extra parts or expedited shipping if it means we can stay flexible and move quickly to meet our customer’s needs. Eventually, that’ll change. As we scale, it’ll make more sense to increase our investment in prevention. But right now, we pick our battles. Some parts are high-risk or high-cost, and we’ll spend the time checking those. Others, we let go and monitor downstream. That balance will keep shifting as we grow, but our culture where everyone, every step of the way, is invested in making quality products will remain the bedrock of how we build our hardware. Helping Build Techs Check Their Work One thing we’re pushing on with Triton is giving our build techs the ability to check their own work earlier. Right now, they’re really good at understanding how parts go together, because good mechanical design communicates that on its own. But when we get into more complex subsystems, like motion platforms wrapped in shielding and packed into tight spaces, things get murky. If it doesn’t work at the end, there are too many places to look. Instead of having our technicians blindly checking off items on a list, we’re building in subsystem testing to enable them to “check their own work” and learn from their mistakes. If we can test a motion platform before it’s wrapped up in the system, we isolate any issues and fix them while things are still accessible. It’s a huge time saver, and it makes debugging much more predictable. Designing Quality In The biggest factor that enables us to take this practical approach to quality is the thought that goes into the design of our hardware products. As Murphy’s Law states “if something can go wrong, it will”. In hardware that means if a part can be installed incorrectly, it will be installed incorrectly. It’s just a matter of time. That means the right move is to design the part so that incorrect isn’t possible. That may sound like just common sense, but it’s actually a core principle in any good quality system. At Lumafield, our hardware engineers live this mantra by heavily prioritizing getting hands-on with what they design. They’re on the floor turning wrenches, seeing what’s intuitive and what isn’t, and quickly making improvements. That feedback loop is gold. It mitigates all but the most manageable risks so that our manufacturing team is set up for success. It allows us to continue manufacturing consistent products as the people and the production environment change. Systems, Not Just Screws At Lumafield, we’re building scanners. But more importantly, we’re building the system that builds the scanners, and it’s a system that will have to scale, adapt, and improve over time. That’s why we’re hiring more manufacturing engineers now. Not to turn screws, but to support the whole process: the practical tools and the problem-solving systems that make it possible to build the same thing well, over and over. That’s what quality really is. Compliance and control are just the beginning. What matters most is taking pride in the little things that bring joy to your customers. It’s about making right-sized solutions so those on the front lines can do their jobs well and get better over time. All too often, manufacturing teams lose sight of this; getting the paperwork/documentation right starts to overshadow getting the product right. Only through thoughtful systems, good design, and people who care can you consistently deliver quality to your customers.

From The Floor

How Saucony Uses CT to Build Better Running Shoes

Luca Ciccone is the Director of Product Engineering at Saucony, which means he spends a lot of time thinking about things runners never see. Heel counters. Foam bonding. The precise position of a carbon plate between two layers of midsole. These are the details that determine whether a shoe delivers on its promises, and until recently the only way to inspect them was to cut the shoe apart. Saucony started working with CT scanning as part of their development process, and it changed how Ciccone's team approaches consistency. We sat down with him and scanned three shoes from Saucony's current lineup: the Omni 9, a stability trainer for everyday wear; the Ride 18, a cushioned shoe built for long miles; and the Endorphin Elite 2, their elite marathon racing shoe. The Omni 9: starting with the heel counter The Omni 9 is designed for stability, which means the heel counter, a rigid internal insert that locks the foot in place, is doing significant work. It also sits inside the upper where no one can see it. "The first thing that I look at when I look at a CT scan of this particular shoe is where the heel counter is being placed on the inside of the upper," Ciccone said. "You cannot see it from the outside, and the only way that you're able to look at it is by cutting it in half." [Omni 9 embed] A heel counter that sits even slightly off-center changes how the shoe fits and how it supports the foot. The variation doesn't have to be dramatic to matter. Runners notice when something is subtly wrong before they can articulate why, and by then the damage to the product's reputation is done. CT scanning lets Saucony verify counter placement across production without destroying shoes to do it. The Ride 18: watching for breakdown before it happens The Ride 18 is a durability story. A runner might put three hundred miles on a pair before they notice that a section of cushioning feels different than it did on day one. By that point, the design decision that caused the breakdown is already in production. "If you're able to look at all of your products throughout testing, you can simply see where those breakdowns occur, and you're able to fix the design down the road," Ciccone said. [Ride 18 embed] The scans show glue application, foam density, and structural bonding in full cross-section. Inconsistencies that would only reveal themselves after miles of wear are visible in the scan from the first prototype. That moves the feedback loop from field returns to the development bench, which is where Ciccone's team can actually do something about it. The Endorphin Elite 2: the carbon plate has to be exactly right The Endorphin Elite 2 is built around a carbon fiber plate sandwiched between layers of PWRRUN PB foam. The plate is what gives the shoe its propulsive return, but only if it is positioned correctly and bonded cleanly to the foam on both sides. An air pocket, a shift in placement, a gap in adhesion: any of these changes how the plate loads and returns energy, which changes how the shoe performs at race pace. [Endorphin Elite 2 embed] The scan shows the full bond between the foam layers and the plate. There is no ambiguity about whether the components are aligned or whether voids are present. Ciccone's team can see exactly what they built and compare it against what they intended to build, shoe by shoe, before a single pair reaches a runner. Consistency at scale "Being able to see the CT scan and digitally cut the shoe in half to look at each one of the components and how everything came together, we can truly understand the consistency that we're getting in our R&D phases, all the way up into production," Ciccone said. That's the practical argument for CT in footwear development. Running shoes are complex assemblies, and complexity means more opportunities for variation. The components that matter most for fit, support, and durability are the ones that are hidden inside the upper or buried in the midsole stack. Making them visible without destroying the shoe changes what engineers can know about their own products, and when they can know it.

Design to Reality

How SawStop Stops a Saw Blade in 5 Milliseconds

Table saws send more than 30,000 people to the hospital in the United States every year. More than 10 of those visits involve amputations, every single day. The SawStop safety brake was invented in 1999 to address exactly that problem. More than two decades later, it remains an optional feature rather than a standard one, and the engineering that makes it work is still worth understanding. We scanned a new SawStop brake cartridge and a deployed one side by side to see what the mechanism looks like before and after it fires. [YouTube video embed] How it works The SawStop system runs a low-voltage electrical signal through the saw blade continuously while the saw is running. Wood is a poor conductor; skin is not. When a finger contacts the blade, the change in the electrical signal is detected by the brake cartridge's digital signal processor in less than a millisecond. The DSP triggers a discharge from four capacitors on the circuit board, sending current through a thin fuse wire. The wire melts, releasing a lever pin that has been holding a powerful spring under compression. The spring drives an aluminum pawl upward into the spinning blade. The blade's teeth bite into the pawl, the blade loses momentum, and a second mechanism retracts the blade below the table surface. The entire sequence takes less than 5 milliseconds, faster than a car airbag deploys. [Embed: new brake cartridge] What the scan reveals The CT scan of the new cartridge shows the pawl sitting above the circuit board, with the compressed spring assembly beneath it. Filtering out the lower-density plastic housing in Voyager isolates the metal components and makes the fuse wire visible — a thread so fine it barely registers until you know to look for it. We measured the wire's diameter in the scan at less than 0.4 mm. That dimension is not arbitrary. The spring it is holding back exerts roughly 150 pounds of force. The fuse wire manages that load through a 3:1 mechanical advantage based on its distance from the fulcrum point of the lever pin, which means the wire only needs to resist about 50 pounds of tensile force. The lever pin has ridges along its surface to keep the wire from slipping under vibration, and is sized precisely small enough to fall clear of the pawl's path when released. The current travels between two electrodes on the PCB across a gap of less than 1.5 mm of wire. The pawl itself is aluminum, chosen because it needs to be soft enough for the blade's teeth to penetrate it and hard enough to arrest the blade. It includes two designed collapse zones: a series of small holes along its edge that compress when struck by the blade's teeth, and larger openings that absorb the blade's kinetic energy. Weight-reducing holes also make the pawl lighter, which means the spring deploys it faster. The rotation point of the pawl is designed with intentional clearance in the plastic housing, allowing the plastic ring to expand or contract with temperature changes. Sawdust accumulation or thermal expansion that binds the pawl would be a failure mode; the design anticipates it. At the base of the cartridge, a locking pin with cam action secures the brake to the saw's trunnion, ensuring consistent contact through years of vibration. A bent tab on the pin maintains that contact without transmitting force directly to the PCB. New vs. deployed [Embed: deployed brake cartridge] The scan of the deployed cartridge shows what two teeth of a spinning blade do to the primary collapse zone in less than 5 milliseconds. The zone has compressed as designed. The secondary collapse zone has also deformed. The fuse wire, taut in the new cartridge, is broken. The lever pin is gone, dropped into the cavity reserved for it in the housing below. [Image: comparison of fuse wire before and after] [Image: pawl comparison before and after] The geometry of the blade's path through the pawl is visible in cross-section: a precise record of where the teeth engaged and how the material responded. The cartridge is ruined and requires replacement, along with the blade. That is the intended outcome. The engineering tradeoff The SawStop brake cartridge is a single-use safety device that costs between $95 and $130 to replace, along with a new blade at $60 to $80. The system can also fire on false triggers. Wet pressure-treated wood, static discharge, or certain voltage spikes can disrupt the electrical signal enough to activate the brake. Those false activations are a real cost for production woodworkers. The opposing argument, that the technology could prevent tens of thousands of injuries annually and that the cost per cartridge is substantially lower than the cost of a hospital visit, has been made repeatedly in congressional testimony and before the CPSC for more than twenty years. The Biden administration advanced a proposed rule that would have mandated blade-contact safety technology on all new table saws sold in the United States. In August 2025, the Trump administration withdrew it. The 840 patent, the key remaining SawStop patent covering the Active Injury Mitigation system, doesn't expire until 2033. SawStop had offered to dedicate it to the public if the federal rule passed. It didn't. Thirty thousand injuries a year continue to happen on a schedule that hasn't changed since voluntary blade guard standards were adopted in 2010. The engineering problem was solved in 1999. The cartridge in the scan works exactly as designed. Everything since has been a different kind of problem.

Materials World

How the Wine Industry Engineered Around Cork's One Flaw

Cork has sealed wine bottles for roughly four centuries, and for most of that time the material was taken for granted. It came from the bark of a tree, it worked, and that was enough. Then winemakers started noticing that some bottles tasted like wet cardboard, and the industry began paying very close attention to what cork actually is and how it actually works. The problem had a name: TCA, or trichloroanisole, a compound that forms when certain mold species interact with chlorine compounds in the cork during production. A contaminated cork doesn't ruin a wine dramatically. It muffles it, replacing fruit and complexity with a damp, musty flatness. Estimates of how often this happens range from one to three percent of bottles sealed with natural cork, which at global wine production volumes represents an enormous amount of spoiled wine. The industry response was two decades of materials engineering, producing alternatives that now compete directly with the original. CT imaging lets us compare all three. Natural cork Natural cork is harvested from the bark of the cork oak every nine to twelve years without harming the tree, then cut into bottle stoppers against the grain. That orientation is deliberate. Cork's cellular structure includes channels called lenticels, porous structures that facilitate gas exchange in the living tree. If a stopper were cut with the grain, those channels would run straight through the cork from bottle to atmosphere. Cut perpendicular to the grain, the lenticular channels are interrupted by denser cell wall material, and the cork provides a seal. The CT scan shows that structure directly. The faint diagonal lines running across the cork are growth rings, and we can count nine of them, representing nine years of growth. The speckled pattern throughout the body of the cork is the lenticel network: lower-density voids surrounded by higher-density cell walls. The material is not uniform, and it was never meant to be. Natural variability is inherent to the biology. That variability is also the source of the TCA problem and of inconsistent oxygen ingress, the rate at which small amounts of air permeate the cork and interact with the wine during aging. Some oxygen is beneficial and necessary for wine development. Too much oxidizes the wine prematurely; too little and certain wines never open up. With natural cork, winemakers are making a bet on a biological material that varies from tree to tree and harvest to harvest. DIAM technical cork DIAM was developed in the early 2000s to solve both problems. The process starts with natural cork, ground into granules, then blasted with supercritical CO2 to extract TCA and other volatile compounds. The cork's suberin, the elastic biopolymer that gives cork its compressibility, is separated from lignin, which is less elastic and discarded. The suberin granules are then combined with microscopic polymer microspheres that fill the voids between particles, reducing porosity in a controlled way, and the mixture is formed into uniform stoppers. The CT scan of a DIAM 5 cork shows the result: a strikingly even distribution of material throughout the stopper, with none of the irregular void structure of the natural cork. The density is consistent from one end to the other. The corkscrew's path through the material is visible and unambiguous, where in the natural cork the self-healing cellular structure tends to close around the screw after extraction. The DIAM process appears to reduce that self-healing behavior, which is visible in the scan as a cleaner, more defined channel. The tradeoff is that DIAM is still accumulating long-term aging data. Natural cork has centuries of evidence behind it. DIAM has decades. For wines intended to age for thirty or forty years, that gap matters to some winemakers. For everything under twenty years, the case for DIAM's consistency is strong. NOMACORC synthetic cork The NOMACORC is manufactured through a co-extrusion process that produces a dense outer skin over a porous foam core. The skin is printed with a pattern that mimics the growth lines and lenticel texture of natural cork, which is visible on the surface but has no functional significance. The CT scan shows the internal structure clearly: the highly porous core is immediately distinguishable from the denser outer layer, and the corkscrew cuts a straight, clean path through both. The NOMACORC Select Green 500 that we scanned is rated for 3.1 mg of oxygen ingress after twelve months, then 1.7 mg annually thereafter. That precision is the product's central claim: where natural cork and even DIAM offer a range of oxygen transmission rates, the synthetic cork delivers a specified value that a winemaker can design around. There is no TCA risk, no biological variability, no vintage-to-vintage inconsistency in the stopper itself. The limitation is that higher baseline oxygen transmission rate. Synthetic corks allow more oxygen into the bottle than natural or DIAM, which accelerates aging. For wines meant to be consumed within a few years of release, that acceleration is irrelevant or even beneficial. For wines built to develop over decades in the bottle, it is a meaningful constraint. What the scans show The three corks represent three different answers to the same question: how do you seal a wine bottle reliably? Natural cork answers with biology, accepting variability as the cost of a material that has proven itself across four hundred years of use. DIAM answers with engineering, taking natural cork's functional properties and removing the failure modes. NOMACORC answers with manufacturing precision, trading biological origin for complete control over oxygen transmission. The CT scans make all three answers legible in a way that external inspection cannot. The lenticel network of the natural cork, the even granular structure of the DIAM, and the foam core of the synthetic are all directly visible, and the differences between them explain exactly why the wine industry spent two decades engineering alternatives to something that worked well enough for centuries but not well enough for the standards modern winemakers hold.

From The Floor

How to Read a Plastic Bottle

If you’ve ever picked up a water bottle and noticed the little numbers, symbols, or codes stamped into the plastic, you’ve already seen the secret language of packaging. Most people don’t realize it, but every mark on that bottle tells a story about where and how it was made. I spent years as an R&D and design engineer at Niagara Bottling, one of the largest vertically integrated beverage manufacturers in North America. We made everything in-house, “from pellets to pallets,” as we used to say. Once you’ve worked inside that process, you start seeing the manufactured world differently. You learn to read the bottles (and even some other products) around you. Plastic recycling symbols explained Let’s start with the biggest misconception: that little triangle made of arrows with a number inside it. Most people think it means the package is recyclable. It doesn’t. That symbol is called a Resin Identification Code, and all it tells you is what type of plastic the product is made from. A “1” means PET (polyethylene terephthalate), which is what nearly every water and soda bottle is made of. “2” means HDPE, often used for caps, milk jugs and detergent bottles. “5” is polypropylene, the material behind many yogurt cups and cream cheese containers. None of these numbers guarantee recyclability. Whether a bottle is actually recycled depends far more on local infrastructure than on the material itself. The chasing arrows were originally meant for manufacturers, not consumers. Somewhere along the way, they were mistaken for a promise instead of an ingredient label. Deciphering manufacturing codes If you flip a bottle over, you’ll usually find other small numbers and marks near the base. Each one corresponds to a step in the manufacturing process. At Niagara, we started with resin pellets, injection-molded them into preforms, blow-molded those into bottles, filled and capped them, shrink-wrapped the cases, and palletized them for shipping—all in one continuous process. The mold number identifies which cavity in a multi-cavity mold produced that particular bottle, for example cavity number 44 out of 96. If a defect ever shows up, the production team can trace it back to the exact mold that caused it. Every plastic bottle includes information about its manufacturer, material, and exact location of its making. You might also see a small logo or symbol from the equipment manufacturer or a co-packer who blew the bottle before it was filled. In vertically integrated companies like Niagara, that all happens under one roof, but many beverage brands outsource to partners who make bottles in one place and fill them somewhere else. The printed date and lot codes near the neck or label are part of the same traceability system. They let a company track every unit back to a specific line, shift, and time. If there’s ever a contamination issue or recall, those codes are how you identify which batches to pull from shelves. Inside PET bottle design A bottle’s design is its own kind of language. The features you can see, and the ones you can’t, tell you what that package was designed to withstand. Those faint seam lines running up the sides show where the two halves of the blow mold met. The tiny rough circle in the center of the base is called the gate mark, where molten PET entered the preform during injection molding. The neck ring, the rigid ring just below the cap threads, is what allows the preform to be handled through the production line before being blown into shape. If you look closely, you can sometimes see knife marks or scuffs from the gripping tooling that carried it. PET plastic bottles bear the traces of being injection-blow-molded with thinner walls indicating seam lines on the mold. Once you learn to recognize these, you can tell not just how a bottle was made, but how many times it has been handled, where the most stress lives, and which design choices were made for automation versus performance. Why bottles look the way they do Nothing in a modern bottle is purely aesthetic. Every ridge, curve, and indentation has a purpose. The label panel, for instance, often sits between raised “bumpers.” Those protect the label from friction as bottles move across filling lines at thousands of units per minute. The five-petal base (called a petaloid) distributes pressure evenly for carbonated beverages. Earlier water bottles sometimes reused the same carbonated mold bases, which is why the original Aquafina bottles just looked like Pepsi bottles. Each bottle’s shape reflects its function: ribs for strength, panels for efficiency, curves for grip, and round bases for pressure resistance. In hot-fill products like juice or sports drinks like Gatorade, you’ll see rectangular side panels that can flex inward as the bottle cools, preventing it from collapsing unevenly. Engineers call these intentional deformation zones that are built to move in predictable ways. Caps and closures As someone who has worked on closures, I can tell you the cap might be the most over-engineered part of the whole bottle. It has to create a perfect seal, apply evenly during high-speed capping, stay child-safe and tamper-evident, and be removable by anyone from a kid to a senior. That’s why you’ll notice larger caps on products like juice or apple sauce packets because they’re easier to grip and twist, even if they use more plastic. Look closely and you might see letters like “CSI” or a crown icon on the top, in addition to a mold number, just like the preforms and bottles. That is the cap manufacturer, not the brand. Companies such as CSI or Crown Holdings specialize in closures, and their marks show which supplier’s tooling was used. The future of packaging After decades of lightweighting and process optimization, we’re approaching the physical limits of how thin and efficient a PET bottle can be. The next big steps probably will not come from redesigning bottles but from improving recycling infrastructure and supply chain circularity. New materials such as bioplastics sound promising, but they introduce trade-offs like limited shelf life, degradation during storage, or contamination in existing recycling streams. Until those challenges are solved, the smartest improvements may come from better systems rather than different chemistries. People think packaging is simple just because it’s disposable. But the opposite is true. A plastic bottle represents one of the most refined pieces of everyday engineering. Every ridge, mark, and number testifies to a problem solved. You just have to know where to look.

Materials World

How We Learned to Hold a Bit Still

Every storage format is a physical answer to the same question: how do you keep a pattern intact long enough to read it back? The pattern might be a magnetic domain, a charge trapped in a transistor, or a pit pressed into polycarbonate. The material it lives in determines how dense the storage can be, how long it lasts, how fast it can be read, and whether the format survives contact with a dusty shelf, a power outage, or a decade of neglect. We CT scanned five formats that shaped how computing stores information. The scans reveal what the outside doesn't: layer counts, mechanical tolerances, material boundaries, and the engineering decisions that separate a format that lasted two years from one that is still setting shipment records in 2025. Magnetic tape Tape looks like the wrong answer until you understand what it's optimizing for. A PET film substrate, coated in magnetic particles, written to by aligning domains with a read/write head. Sequential access only. No random reads. The mechanism is simple enough to have worked in mainrooms in the 1950s and in cassette-based home computers in the 1980s, where machines like the VIC-20 stored data as modulated audio tones at around 50 bytes per second. [embed: LTO tape scan] What kept tape alive while every other format on this list faded is the physics of the coating itself. Magnetic particle density has kept climbing. LTO-9 cartridges hold up to 45 terabytes compressed, and cartridges retain data for up to 30 years under the right conditions. The cartridge, once written and ejected, uses no power and remains offline, physically disconnected from any network. That air gap, which looked like a limitation in the era of always-on storage, turns out to be a significant security property in the era of ransomware. As IBM's Mark Lantz has noted, the technology hasn't been frozen in time. In 2024, LTO shipments hit a record 176.5 exabytes, 15.4% growth over 2023, the fourth consecutive year of records. Amazon, Google, and CERN rely on it for long-term archival. The material never stopped being right for the job. Floppy disks The floppy's design story is mostly about the enclosure catching up to the media. The original 8-inch disk, introduced in 1971, was physically floppy: a magnetic-coated polyester film disc in a paper sleeve, vulnerable to contamination and handling damage. By the late 1970s the 5.25-inch format had reduced size and added a stiffer sleeve. The 3.5-inch disk, which arrived in the 1980s and remained in common use into the early 2000s, finally solved the enclosure problem properly. [embed: 3.5-inch floppy disk scan] The CT scan shows what made the 3.5-inch format work: a rigid polycarbonate shell, a spring-loaded metal shutter that closes automatically when the disk is ejected, and a precise read-write aperture with minimal clearance between moving parts. The magnetic film inside is the same basic material as earlier formats. The engineering that extended the format's useful life was entirely in the housing. The disk stayed the same. The protection around it got better. Iomega's Zip and PocketZip formats attempted to extend this logic further, reaching 750 megabytes and 40 megabytes respectively by packing more surface area into a cartridge that handled like an oversized floppy. Alignment issues and media wear cut that effort short, and USB flash drives made the entire category obsolete within a few years. ROM cartridges ROM cartridges solved a different problem than magnetic storage: they needed to be read, not written to, and they needed to survive thousands of insertions into consumer hardware. Mask ROM chips, used in consoles like the Atari 2600 and the Nintendo Entertainment System, held program data burned in at fabrication. There were no moving parts, no magnetic coatings, nothing to degrade. [embed: ROM cartridge scan] The scan shows the architecture clearly: a ROM chip mounted directly to a PCB, clean solder joints, and precisely spaced edge connectors designed to mate with spring-loaded contacts in the console slot and maintain electrical contact through repeated use. The simplicity is the point. A cartridge that holds data in silicon and connects through a physical edge connector has almost nothing that can fail. The ritual of blowing into cartridges to fix loading failures is worth noting here. It was ineffective and often counterproductive: moisture from breath could corrode the gold or nickel-plated contacts over time. The actual failure mode was oxide buildup on the connector surfaces, which a cleaning kit addressed far better than a breath. The cartridge was more reliable than its reputation suggested. What failed was the connector interface, not the storage medium itself. Magneto-optical discs Magneto-optical storage is the most mechanically complex format on this list. Writing requires two simultaneous physical processes: a laser heats a spot on the ferromagnetic recording layer to its Curie point, at which magnetic coercivity drops to near zero, and a magnetic write head on the opposite side of the disc sets the domain orientation at that moment. Reading uses a lower-power laser and detects the polarization shift that occurs when polarized light reflects from a magnetized surface, an effect called Kerr rotation. [embed: magneto-optical disc scan] The 3.5-inch MO cartridge appeared in 1991 with capacities from 128 MB up to 1.3 GB, and the format became a trusted medium in enterprise archives for legal, medical, and media libraries. The CT scan shows why: a thick ferromagnetic recording layer enclosed in a robust polycarbonate cartridge, with a precisely toleranced guide system for laser alignment and careful spacing for the write head. The cartridge physically protects the media surface from dust and fingerprints. The write-verify cycle, which confirmed data integrity after every write, made it among the most reliable rewritable media available before affordable CD-R arrived. That reliability came at a cost in write speed, and once writable optical formats became cheap enough, the economic case for the complexity disappeared. Hard drives and solid-state storage Hard drives hold data on aluminum or glass platters coated in a cobalt alloy film. Read/write heads hover nanometers above the platter surface, floating on a thin cushion of air generated by the disc's rotation, and flip magnetic domains as the platters spin at 5,400 or 7,200 RPM. Multi-platter configurations multiply capacity by stacking surfaces. It is the most mechanically elaborate format in everyday use, and that mechanism is the source of its vulnerability: shock, vibration, and wear all eventually reach the heads. [embed: SSD scan] Solid-state drives replace that mechanism with NAND flash memory, where floating-gate transistors retain charge without power. The scan of a Lite-On SSD shows NAND packages arranged around a central controller, with dense surface-mount components and multilayer routing on a board with no moving parts. The controller manages wear leveling, data distribution, and error correction across the flash cells. NVMe models route data directly over PCIe, reaching transfer speeds beyond 7 GB/s. Five years of reliability data from Backblaze show that SSDs fail far less often than hard drives over time. The tradeoff is cost per terabyte, which remains substantially higher than either hard drive or tape at scale. What the formats share Looking across all five formats, the engineering logic is consistent. Each one reached its practical limits not when the storage mechanism failed but when the surrounding constraints changed: a faster, cheaper alternative arrived, or a use case the format wasn't designed for became dominant. Tape survived because the constraints it was designed for, density, offline durability, and cost at scale, turned out to matter more over time, not less. ROM cartridges survive in industrial and embedded applications for the same reason: nothing fails. MO discs disappeared when the reliability premium they offered stopped justifying their cost. The material inside each format encodes those tradeoffs as precisely as the data it stores.

From The Floor

Inside a 12‑Month Sprint from Concept to Factory‑Ready Product

I’m the co-Founder and Head of Hardware at Lumafield. I’ve spent my career in the electrical and manufacturing space, and now I’m leading the development of Triton, an automated in-line CT inspection solution, which we developed very quickly over the last year. The reality of manufacturing is that it’s always a little bit messier than people expect, especially when you’re iterating quickly. I’m an electrical engineer by training. When I started my studies at MIT, I thought I’d do computer science, but I realized I loved shipping physical things that touch the real world. That pulled me into roles at Sonos, Color Kinetics, a startup I co-founded called Digital Lumens, and later, the 3D printing company, Formlabs. The reality of manufacturing is that it’s always a little bit messier than people expect, especially when you’re iterating quickly. Each step taught me how ideas survive contact with factories and customers. With Triton, we heard loud and clear from customers that speed of inspection is critical. Automated in-line inspection and extremely fast CT was a development that came out of our deep engagement with customers. This article is about how our iterative, hands-on culture and staying close to customers let us build Triton quickly and reliably. Closing the Physical-to-Digital Loop In my last role at Formlabs, I worked on tools that moved designs from the digital realm into the physical world. What struck me was how few teams had tools to move in the other direction. It was almost impossible to bring detailed, actionable information from physical products back into the digital domain. With Lumafield, our vision was to make the physical world legible again. With Lumafield, our vision was to make the physical world legible again. X-ray CT gives teams immediate answers: Are parts true to intent? Are there cracks, pores, or dimensional flaws? Are components missing or assemblies incorrect? We set out to democratize that feedback and return it to the design process quickly, so products can be built more reliably. The Concept: Triton When we built our first hardware product here at Lumafield, Neptune, we were building a product that we ourselves, as designers of products, knew intuitively would be very valuable for the R&D and NPI process. We were delighted and surprised to see that right away, our customers took the Neptune tool and applied it to manufacturing problems that were a little bit outside the scope of what we had initially envisioned. Pretty shortly after we started shipping Neptunes, we started noodling on the idea of what continuous inspection for manufacturing could look like and how Lumafield could deliver it. In the same year that we shipped Neptune, we built a proof of concept of our idea for continuous inspection, which was essentially a robot arm fitted to a Neptune. We actually brought it to a trade show and showed it to real prospective customers. We started chipping away at that idea and refined it before we converged on the vision that ultimately became Triton. Triton’s design patent. How We Moved Fast: Prototype-Driven Iteration Many companies focus on nailing down the exact product requirements very early in the process and spend a lot of time upfront deciding what to build and how to build it before they even pick up a screwdriver. We do it differently. Our mission is to ship a prototype even though we know it might be missing features, get that out there in the field, and listen carefully. Prototype early, prototype often, and get your prototypes out there in the hands of real users as quickly as possible. We take an extremely iterative process of hardware development, and we ship our iterations. Prototype early, prototype often, and get your prototypes out there in the hands of real users as quickly as possible. Our different iterations of prototypes showed different approaches to part handling and conveyance, and with each generation we learned some important new lessons. I’m a strong believer in the iterative approach to managing risks and solving problems. Getting as quickly as possible into building some kind of proof of concept is really valuable. A good design is never completely done. We’re constantly shipping iterative improvements to our products, and we’ve even built that into our business model. The engineering crew collaborates to position a subsystem. Hands-On Culture, Hardware-as-a-Service (HaaS), and Shared Incentives Our culture is certainly not an accident. Our engineers are very hands-on. We have our own prototyping shop in addition to having our own factory. One of our strategic advantages is speed. This is enabled by our hardware‑engineering culture that has been pretty consistent from the beginning. Ultimately, that’s the superpower behind our ability to ship hardware quickly. Our engineers are there building the first versions of our product with their own hands, and they travel to the field to perform our initial deployments as well. Bundled together with the hardware delivery is ongoing support, so our company and our customers have a shared incentive that our hardware is reliable. They develop a holistic, full‑stack understanding, because a Lumafield product is not simply hardware; it's hardware, software, and inspection workflows all coming together as a system. Bundled together with the hardware delivery is ongoing support, so our company and our customers have a shared incentive that our hardware is reliable. As a leader, it’s important to simultaneously be very, very close to the engineering design and very, very close to our customers. Big smiles all around as the team hits a major assembly milestone on Triton. Speed as a Strategic Advantage We learned right away how important speed is for generating value for our customers. Across the board, we’ve heard loud and clear from customers that speed of inspection is critical. Ultra-Fast CT is a development that came out of our internal process‑development team. A core technical hurdle for us was part handling. One of the key requirements for an industrial manufacturing‑floor CT inspection machine is efficient, high‑speed part handling. Controlling our own manufacturing through vertical integration allows us to move very quickly on complex products. What you see in the Triton that we ship today is the result of many generations of refinement of that idea. We believe it’s the most robust possible way to move parts in and out of an X‑ray CT machine. Controlling our own manufacturing through vertical integration allows us to move very quickly on complex products. In addition, precision and reliability are both key parts of our product requirements for Triton. We test and validate every single claim we make. Why this Work Stays Fun I love visiting customer sites. The diversity of applications is immense: medical devices, batteries, plastic molding, aluminum die casting, automotive, sporting goods. We see everything from ten-cent pumps to premium surgical tools. The materials and business models vary, but the core problem is the same everywhere: getting what is on the screen to agree with what we ship in the real world. No matter what type of business we’re visiting, we find that we speak the same language, and we’re all solving the same problems at the end of the day.

Scan of the Month

Inside the Nintendo Switch 2 Joy-Cons

Nintendo has never competed on raw specs. Its consoles earn their following by rethinking how people play: the NES with its toy-like accessibility, the Wii with motion controls that pulled non-gamers off the couch, the original Switch with a form factor that collapsed the distinction between home and handheld gaming. The Switch 2 builds on that legacy with meaningfully better hardware: a new Nvidia chip, LPDDR5X memory, 256 GB of UFS 3.1 storage, an 8-inch 120 Hz HDR display. It is faster and sharper in every measurable way. We took a non-destructive look inside one of its most consequential components: the redesigned Joy-Con controller. The drift problem, still present Nintendo describes the Switch 2 Joy-Cons as redesigned from the ground up. That is structurally true. The rail-and-clip attachment system is gone, replaced by magnetic latching. There is mouse-style input, integrated voice and video chat, an updated look. But under the thumbstick, the fundamental design decision is unchanged. Stick drift is the defining complaint of the original Joy-Con. The character moves when you are not touching the controls. That is not a software glitch. It comes from physical degradation of the hardware: a potentiometer, where metal contacts slide across a resistive surface. The surface wears down over time, or accumulates dust, and begins producing false signals. The new Joy-Con uses the same sensing principle. Nintendo has already extended its Joy-Con repair program to cover Switch 2, which says something about their confidence in the fix. Our CT scan shows the internal layout clearly: the magnetic latch mechanism, the compact component arrangement, and the stick module at the center of it. The engineering is elegant in many respects. The potentiometer is still there. Why the alternative wasn't used Potentiometers dominate consumer controllers because they are inexpensive and straightforward to integrate. The failure mode is real but predictable, and the cost is low enough that replacement rather than prevention has been the industry's default response. The alternatives are better on durability but harder to implement. Hall effect sensors detect position through magnetic fields rather than physical contact, which means no wear and no drift. Third-party controllers from 8BitDo and GuliKit already use this approach, and the performance difference is measurable. TMR sensors, which use tunneling magnetoresistance, offer even greater sensitivity and stronger resistance to interference. The Switch 2 has a specific problem with Hall sensors: the Joy-Cons attach magnetically, and those mounting magnets would interfere with Hall-based position sensing. TMR could have worked around that constraint. But it would have required new suppliers, firmware changes, and a longer development cycle. Nintendo weighed those costs against the timeline and chose the known quantity. What the scan shows CT imaging does not render a verdict on that tradeoff. What it does is make the tradeoff visible. The Joy-Con is a well-integrated piece of hardware with a known failure mode at its center, and the scan shows both things simultaneously. Every hardware team works inside real constraints: timelines, supply chains, cost targets, technical dependencies. The Switch 2 Joy-Con reflects all of those. The drift risk reflects them too.

From The Floor

Malicious Hardware Hidden in Plain Sight

A USB cable is one of the most trusted objects in a workplace. It sits on desks, gets borrowed between colleagues, travels in laptop bags. Nobody thinks twice about plugging one in. That trust is exactly what the O.MG Cable is designed to exploit. Security researcher Mike Grover, known in the hardware security community as MG, spent years miniaturizing a complete attack platform into a cable that is physically indistinguishable from a standard USB cable. The implant stays dormant until activated remotely. When it is active, it can log keystrokes, inject payloads, or establish a bidirectional tunnel to a remote operator. The cable behaves as a normal USB 2.0 device in the meantime: 5V charging, 480 Mbps data transfer, nothing unusual on any log. The hardware doing all of that fits inside the connector housing at the end of the cable. We scanned one to see how. What the scan shows The CT scan reveals a microprocessor and antenna packed into the USB connector housing with almost no wasted space. The precision of the layout is immediately apparent. This is not a crude modification; the internal components are arranged with the same care you would expect from a consumer electronics manufacturer, because that is the only way to make the implant fit and still have the cable pass visual and functional inspection. The more revealing detail, and the one MG pointed out when we showed him the scan, is a secondary silicon component bonded to the underside of the primary processor. It is easy to miss. The two chips read as a single assembly at first glance, and the bond wires connecting them are approximately the diameter of a human hair. Those wires carry signals that would be invisible to any inspection method that cannot volumetrically resolve the internal structure of the connector. A 2D X-ray looking down at the assembly might show component outlines. It would not show this. The O.MG Cable is sold commercially for red team security testing, the legitimate practice of simulating adversary attacks against an organization's own infrastructure. Grover's stated goal is to demonstrate that hardware implants of this sophistication are possible and to give security teams a tool for testing their own detection capabilities. The cable currently sells for a fraction of what intelligence-grade equivalents cost: the NSA's COTTONMOUTH-I, a hardware implant with comparable functionality, was reported to cost $20,000 per unit. What it means for supply chain inspection The O.MG Cable is useful as a security research tool precisely because it makes the threat concrete. A cable like this can be introduced into a supply chain, left in a conference room, or handed to a target as a gift. Once it is plugged in, software-based security measures offer limited protection. The attack surface is below the operating system. CT scanning at the lab level does not solve this problem at scale on its own. A Neptune scan takes time and requires someone to specifically decide to inspect a cable. But Triton can automate high-volume inspection, and Mars can run 100% inline checks at production speed, which means the same visibility that makes the O.MG Cable's internals legible is available across an entire production run, not just a single unit pulled for investigation. The scan cannot tell you the cable is safe, but it can tell you whether the cable contains what a cable is supposed to contain. That distinction, between a pass on visual inspection and a verified internal structure, is where hardware security decisions actually get made.

From The Floor

Manufacturing in 2026: Less Disruption, More Discipline

Manufacturing enters 2026 under a heavy load of expectation. Industry headlines have been promising a wave of transformation for a while now: smart factories, reshoring booms, and autonomous production lines that will finally close the loop between design and reality.The truth is less cinematic and far more instructive. As Lumafield’s leadership sees it, this year will not bring sweeping reinvention. It will bring slow, deliberate progress and the inevitable friction that comes when soaring promises collide with physical limits.Manufacturing is now entering a phase of realism. The technologies shaping 2026 are significant, but so are the boundaries that define them. What's coming is not disruption but discipline.The slow burn of automation and AIFor years, analysts have treated AI as an epochal wave poised to wash over industry. The reality looks more like a rising tide than a flood.Andreas Bastian, Lumafield's Head of Product“Geopolitical and regulatory changes will likely outweigh the more incremental changes coming to market from new technology and product development,” says Andreas Bastian, Co-founder and Head of Product at Lumafield. “However, I expect to see steady adoption of automation in manufacturing, whether it’s more automated inspection technology, including CT, or document-processing applications. We will also see continued experimentation with LLM-based AI, but it will not fully displace anything at an industry level in 2026.”AI is already proving useful in narrow contexts such as inspection, quoting, and documentation, but it remains bounded by physical processes. The next twelve months will be less about revolution and more about integration that improves existing workflows.According to Dan Pipe-Mazo, Head of Engineering at Lumafield, “The barrier to customizing and building manufacturing software is significantly lower with AI tools such as Cursor, Claude, and ChatGPT. We’ll see more companies ditching traditional manufacturing software providers and rolling their own systems in-house, customized to their needs.”That shift is significant. AI will not yet replace human labor or factory control, but it will begin to replace the brittle, one-size-fits-all software that has dominated manufacturing technology for decades.Reshoring isn’t here just yetThe return of manufacturing to North America has been on everyone’s minds for at least the last year. The narrative sells optimism: new factories, secure supply chains, and national renewal. The timelines, however, don’t add up.Eduardo Torrealba, Lumafield's Co-founder and CEO“The uncomfortable truth is that the jobs are not coming back on the timelines people keep talking about,” says Eduardo Torrealba, Co-founder and CEO of Lumafield. “Manufacturing operates on multi-year cycles. If it takes five years to move a single production line from China to Vietnam, two countries that already know how to scale, then a one-year forecast isn’t realistic.”Even with historic incentives like the CHIPS Act, transformation remains slow. “TSMC announced its Arizona fab in 2020 and only began real volume production this October,” said Torrealba. “That’s five years, with extraordinary incentive. If $53 billion barely moves the needle on semiconductors, it will not solve reshoring everywhere else.”When it comes to reshoring in 2026, expect announcements, not achievements. The groundwork for regional diversification will continue, but progress will remain incremental, limited by logistics, skilled labor, and the deep interdependence of global supply chains.The real constraint is geopolitical and electricalDan Pipe-Mazo, Lumafield's Head of EngineeringEven as factories experiment with new automation and AI tools, the pressures ultimately shaping global production are coming from external forces. Policy and infrastructure are starting to define the limits of manufacturing flexibility. “Supply chain risk due to geopolitical conflict is very top of mind,” says Pipe-Mazo. “Things move much quicker in politics than supply chains, and any changes that restrict purchases from countries such as China entirely will be very disruptive.”Energy scarcity in particular will compound that tension. AI’s demand for computation is already straining power grids. “Power consumption and energy supply is the key bottleneck to AI data center buildout at the moment,” Pipe-Mazo adds. “We don’t have the grid capacity to meet demand. Electricity for data centers will become the new oil.” For that reason, efficiency may become a decisive factor.“I’m really fascinated by measures of information or intelligence per watt (IPW),” says Bastian, “and what new compute technologies, mathematical frameworks, compression techniques, and training strategies can open up.”Manufacturers planning large-scale digital initiatives such as AI-assisted process control, simulation, or data-heavy inspection will need to design around these constraints. The year ahead will be defined as much by what factories cannot run as by what they can automate.Clean data becomes imperativeEven amid global complexity, the most advanced manufacturers are quietly expanding their advantage.“Customers who are changing how they measure their processes are making deep discoveries about their products and processes that weren’t previously possible,” says Bastian.The most advanced manufacturers are moving beyond off-the-shelf tools and building proprietary datasets that give them a competitive edge. As Dan Pipe-Mazo notes, “They’re building datasets of their products using CT and surface capture in order to build their own AI models.” These custom datasets become the foundation for more accurate predictions, better process control, and ultimately differentiation in the AI race.This approach represents a more measurable kind of transformation. In 2026, the leaders will not be those with the flashiest tools, but those with the cleanest, best-labeled data.AI’s next phase: augmentation instead of replacementIf 2024 was the year of AI experimentation and 2025 the year of inflated expectation, then 2026 will be the year of pragmatic adoption.“We’re in a pragmatic portion of the hype cycle,” says Pipe-Mazo. “Reliable solutions aren’t quite there yet. We’ll find AI settling into a niche where it doesn’t replace work but instead handles the easier cases and helps teams be more efficient.”AI is unlikely to take over any manufacturing workflow this year, but that should not count against it. “None,” Pipe-Mazo answers when asked what processes AI will fully control by 2026. “Full takeover is not a great fit for manufacturing use cases that value determinism.”Bastian agrees. “A 2025 MIT study found 95 percent of AI pilots in organizations fail. This should not be interpreted as an indictment of the technology, but of our misunderstanding of its capabilities,” he says. “It would be a mistake to give up on the technology based on one bad experience. A single data point becomes stale within weeks given the pace the technology is moving at.”The message is clear. AI’s value lies in its partnership with human expertise, not its replacement.Leading companies who already manufacture their products in the U.S. such as Skydio are strengthening their team's expertise with intelligent quality systems including industrial CT scanning.The discipline of real progressBy the end of 2026, the headlines may look familiar: AI adoption rising, reshoring on the march, and greater automation. Behind those stories will be something bigger but more lasting, a new kind of operational discipline.Manufacturers are learning (or remembering) that even when it comes to AI, transformation is a marathon, not a sprint. It requires patience, measurement, and a deep understanding of the physical limits that advances in software cannot erase overnight.As Torrealba puts it, “Manufacturing doesn’t change in single-year increments. Yes, automation will advance, but the underlying industrial structure won’t shift nearly as quickly as the headlines imply.”

Materials World

Materials That Make or Break a Shoe

When I first started cutting shoes in half, I thought quality was just about what you get for your money. Do you get higher quality materials or not? But once I started learning more about barefoot shoes, I realized that there’s more to comfort than just materials. The shape of the last matters. Fit matters. There’s so much more to it than meets the eye.Quality is a multifaceted thing. Materials, aesthetic, and fit are indispensable elements of it, but the way you use the shoes is what puts quality to the test. You have to gather as much information as possible to find the right shoe for your needs. Do you have wide feet? Then you need a wide shoe. Are you doing heavy duty work? You need something durable enough for that. Winter? You need insulation and grip. The problem is that most people don’t even know what’s inside footwear unless you literally cut it in half.Myths about brands and countriesI run a YouTube channel called Rose Anvil, where I cut shoes and boots in half to learn about their construction. What started as just wanting to understand why some boots last decades and others fall apart in months has turned into hundreds of detailed teardowns. I’ve cut open more than 500 pairs, from luxury brands to bargain-bin imports, and what I’ve found along the way has changed how I think about quality, value, and the way things are made.“Over the last 500 pairs of shoes I’ve cut open, I’ve seen how older brands that get bought and sold over and over end up strip-mining all the value.”One of the biggest misconceptions of footwear manufacturing is that brands with long histories are automatically better. Over the last 500 pairs of shoes I’ve cut open, I’ve seen how older brands that get bought and sold over and over end up strip-mining all the value. They squeeze margins to pay off the business deal and the end consumer foots the bill.A craftsman stitches the upper of a leather boot, guiding each curve by hand to ensure strength, precision, and a perfect fit.Same with countries. Many people automatically assume that everything made in China is garbage. But look at Grant Stone. They produce some of the highest quality footwear in the world, and it’s made in China. On the other side, you’ve got the $10 boots you see on Temu, where you don’t even know what will show up. Is that China’s fault, or is that because people want to buy something for $5 that should cost $50?“The factory will build to whatever standard the brand sets.”It’s easy to blame where something’s made, but the truth usually comes down to choices made higher up. The factory will build to whatever standard the brand sets. When Dr. Martens moved most of their production out of England, people were right to complain. The quality dropped. But that wasn’t because the factories overseas couldn’t make them well. It was because the company told those factories to make cheaper boots to boost margins. They could have kept the same quality at a lower price, but they chose not to.Reading leather without cutting itMany people ask me how to assess the quality of leather. Here are a few signs you can look for. If you can see the raw edge, look at the fibers. Toward the top, closer to the smooth surface, the fibers should get really fine, almost solid. That means the grain is intact, which is the most sought-after part of leather. It looks smooth, but it’s also really strong because it binds all the looser suede fibers together.“As a general rule, the more real leather you see in footwear, the higher the quality.”Feel matters too. If leather feels foamy, dry, or like it could crack and fall apart, that’s obviously not a positive sign. If it feels almost wet to the touch or like it has conditioner or oil in it, that usually means they went the extra mile in finishing. As a general rule, the more real leather you see in footwear, the higher the quality. Rose Anvil x White’s Sundance Boot updates the classic Packer with a lighter build, high arch, and hand-sewn stitchdown craftsmanship made in Spokane, Washington.Pigskin is an interesting use case. It’s thinner than cowhide, but has a tighter fiber structure and big pores. Sometimes the pores even go all the way through. That makes it breathable and strong in thinner cuts. You can make an unlined pigskin boot that’s lightweight, durable, and breathable in a way you can’t with cowhide. The downside is the hides are smaller and supply is limited, so it’s ultimately harder to work with.Materials that fail and materials that lastEveryone has probably worn through the inside heel of a shoe. That’s one of the most common failure points. Higher-quality boots often have a suede or roughout leather cover inside the heel. It grips better, molds to your heel, and resists abrasion.Underfoot tells another story. For centuries people stood on leather midsoles. Leather shapes to your foot, absorbs sweat, and doesn’t smell as much. It’s also expensive and firmer than foams, which is why rubber and foams have taken over. Foams are cheaper, softer right away, but they don’t last as long.If you pull the insole and look underneath, you’ll see what you’re really standing on. Is it leather, fiberboard, or just thin fabric? That’s an excellent indicator of the lifespan you can expect from a shoe.Buying smarter with materials in mindDon’t trust the brand name alone. Feel the shoe for yourself. Look at it closely. Compare it to other shoes at the same price point. There is a minimum buyable product. If you’re buying footwear under $40, or boots under $100, something has to give. The quality will be poor enough that it likely will fall apart fast. In the long run, people often spend more replacing cheap footwear than they would by buying one decent pair. If you’re on a budget, you’re better off getting a used pair of higher-quality boots than a brand new $40 pair.“When cheap materials fail, you pay for it twice: once when the shoe breaks and again when you replace it.”People sometimes argue that spending more is just about fashion or image, but footwear really is a tool. Shoes are the equipment that get you from point A to point B, that keep you working, and that carry all of your body weight every day. When cheap materials fail, you pay for it twice: once when the shoe breaks and again when you replace it. Better materials may cost more upfront, but they save you money, time, and frustration.How it changed the way I buyGood shoes balance health, quality, story, and style. You wear them every day. Shoes are the one thing separating you from the ground. All of your weight rests on them. That means materials matter.“I want to support brands that still care about craft, not the ones that have been bought and sold until they’re just a shadow of what they were.”As a consumer, I’ve become picky with my choice of shoes. I want to support brands that still care about craft, not the ones that have been bought and sold until they’re just a shadow of what they were. I like finding footwear with a story behind it. It becomes a conversation piece, and it makes the experience of wearing them more enjoyable.This industrial CT scan of a Japanese M98 boot from World War II reveals its internal stitching, layers, and sole construction without needing to cut it in half.I’ve also changed for health reasons. Barefoot shoes, wide toe boxes, zero drop are all design choices that make you realize that shoes should actually be shaped like feet. Once you wear wider shoes for a while, putting on a narrow pair feels terrible.We spend all this time worrying about posture in chairs. Shoes deserve at least as much attention. They’re the foundation for our movement through the world.

Design to Reality

Not All USB-C Cables Are the Same

USB-C promised a universal standard: one connector for everything. One port for power, data, and video. One cable you could grab without thinking. What the standard actually delivered is a connector shape that looks identical across products with wildly different capabilities inside. The same plug that powers a laptop at 100W and transfers data at 40 Gbps can also be a cable that supports nothing beyond 480 Mbps USB 2.0, and you cannot tell from the outside which one you have. We scanned four USB-C cables at different price points to see what accounts for the difference. Adam Savage covered the results with us on Tested, and the response was significant enough that it's worth a closer look at the engineering. Apple Thunderbolt 4 Pro Cable The Apple Thunderbolt 4 Pro Cable retails for $129 for the 1.8m version. It supports Thunderbolt 3, Thunderbolt 4, and USB 4 data transfer up to 40 Gbps, DisplayPort video output, and charging up to 100W. The scan shows why. The connector has a stainless steel shell that is fully bonded to the plastic enclosure, not floating inside it. Where the cable meets the connector, a single-piece strain relief is crimped from eight directions, wrapping the entire base of the assembly. The connector tip has 24 pins, each mounted independently on a nine-layer printed circuit board. Filtering out the lower-density PCB substrate in Voyager reveals a dense network of blind and buried vias: blind vias connect outer layers to inner layers without passing all the way through the board, while buried vias connect only interior layers and are not visible from either surface. Both are manufacturing complexity added specifically to support high-speed signal routing in a connector with almost no available space. One detail on the board is easy to miss: a small detour on one of the traces, a deliberate length-matched wiggle that ensures it arrives at the same time as its paired trace. At 40 Gbps, timing differences of fractions of a nanosecond produce signal errors. The cable also carries three distinct wire types: coaxially shielded conductors for high-speed data, unshielded power wires, and two legacy USB 2.0 data wires for backward compatibility. Each wire and shield lands separately on the board. Amazon Basics USB-C cable The Amazon Basics USB-C to USB-C 2.0 Fast Charger Cable supports charging up to 60W and data transfer up to 480 Mbps. At a fraction of the Apple cable's price, it delivers a fraction of its capability, and the scan shows exactly where the tradeoffs were made. The connector has a metal jacket beneath its plastic enclosure, similar to the Thunderbolt cable, integrated with the strain relief. The strain relief itself is simpler: two arms that crimp together rather than the eight-direction wrap of the Apple design. Inside the connector tip are 12 pins, half the count of the Thunderbolt cable, and four pairs of those are jumped across rather than individually connected to the board. The plug shell is fully grounded and soldered to the end pins. For a cable that only needs to deliver power and USB 2.0 data, this is sufficient and efficiently made. NiceTQ USB-C cable The NiceTQ cable claimed data transfer up to 10 Gbps at $5.59. The scan tells a different story. There is no metal shielding beneath the plastic enclosure. The plug shell floats in overmolded plastic with no connection to the cable's ground wire. The strain relief is a rubber extension of the plastic enclosure, no metal reinforcement. Inside the connector tip are eight pins, but only four connect to the cable. The other four float in the overmolding with no electrical connection, unlike the Amazon Basics design where mirrored pins are jumped across to provide redundancy. If an active pin on the NiceTQ connector wears down or fails to make contact, nothing backs it up. The cable contains no circuit board at all. The pins connect directly to four stranded conductors, each in two layers of insulation. Four conductors and four active pins support USB 2.0 at 480 Mbps. 10 Gbps requires USB 3.1 Gen 2, which this cable cannot physically support regardless of what the listing claims. The cable had accumulated 29 one-star reviews on Amazon and was discontinued the day after we bought it. ATYFUER USB-C cable The ATYFUER cable, listed at $3.89 and advertised only as a charging cable with no data specs, turns out to be the most interesting comparison in the group. The connector has a full set of 24 pins on a circuit board, though only 12 are connected to traces. Despite being sold as a charging-only cable, its wiring supports USB 2.0 data transfer. Why does a cheap charging cable have 24 pins when only half are connected? One explanation: the same factory produces Thunderbolt cables and maintains a single connector design across its production line. Running the same tooling and assembly process for every cable, regardless of final specification, costs less than maintaining separate tooling for each tier. The result is a cable with more hardware than its function requires, sold at the bottom of the market. What the comparison shows USB-C is a connector standard, not a capability guarantee. The same plug geometry can house a nine-layer PCB with length-matched signal traces or four wires connecting directly to floating pins. Specifications printed on packaging describe what a cable is rated for; they do not describe what it contains. The scan closes that gap. For cable manufacturers, it also reveals something about how production economics drive design decisions at every price point: the Thunderbolt cable's complexity is justified by what it needs to do, the Amazon Basics cable is efficiently matched to its purpose, the NiceTQ cable cannot do what it claims, and the ATYFUER cable may be sharing tooling with something far more capable. A fair objection When this piece ran alongside Adam Savage's video on Tested, the most common response was a version of the same critique: comparing a Thunderbolt 4 cable to USB 2.0 cables is comparing different categories. It would be more useful to compare the Apple cable to an equivalently capable Thunderbolt 4 cable from Anker, Cable Matters, or Belkin at roughly half the price. That objection is correct. What this comparison does show is the full range of what the USB-C connector can contain, from a nine-layer PCB with length-matched signal traces to four wires landing directly on floating pins. That range is real and not obvious from the outside, which is the point. But it does not answer the question most people actually have: if you need Thunderbolt speeds, does the Apple cable justify its price over a certified alternative? We haven't scanned that comparison yet. It's the right next question.

Design to Reality

Nothing Ear 3 Has Nothing to Hide

Nothing built its reputation on showing you what's inside. The transparent casings, visible screws, and dot-matrix typography are all deliberate: the internal engineering is part of the product's identity in a way almost no other consumer electronics brand attempts. So it felt fitting to put the Nothing Ear 3 through a CT scanner and see what the transparency doesn't show. The Ear 3 is Nothing's most capable earbud to date, and at $179 it lands $70 below the AirPods Pro 3. The scans show why the comparison is worth taking seriously. Earpiece Each earbud contains a 12mm dynamic driver with a PMI diaphragm and TPU surround, tuned for a wide soundstage and strong bass response. The driver is the largest visible component in the earpiece scan, and its size relative to the overall housing is notable. A 55mAh lithium-ion battery sits alongside it. The top and side cross-sections of the battery show the characteristic spiral of a wound cell, the same basic construction used in batteries many times larger. Three directional MEMS microphones are distributed across each earbud. Two handle the hybrid ANC array, one facing outward to sample ambient noise and one inward to monitor what reaches the ear canal. The third handles voice pickup for calls. Alongside the microphones, a bone conduction voice pickup unit reads jaw vibrations directly, giving the call processing system a cleaner voice signal in noisy environments by bypassing airborne sound entirely. The antenna is visible in the scan running along the outer edge of the earpiece housing, positioned to maintain Bluetooth signal clearance away from the electronics. PCB The stem PCB scan shows the full electronics stack in a package smaller than your pinky. The Bluetooth SoC handles wireless connectivity and audio processing. A battery charge chip manages power delivery to the 55mAh cell. A touch sensor and touch chip on the lower stem handle gesture controls. Two circular elements at the base of the stem are the charging contacts. The component density on a board this small is more remarkable than anything the transparent shell reveals from the outside. Case The charging case carries a 500mAh battery, enough for multiple full recharges of both earbuds. The case scan also shows the Super Mic: a dual-MEMS microphone array built into the case itself, designed to function as a standalone microphone for calls when one earbud is removed. The two MEMS microphones are visible as distinct components separate from the charging circuitry. How it compares to the AirPods Pro 3 The AirPods Pro 3 is a different product with different priorities. Apple's H2 chip delivers what reviewers consistently describe as the best active noise cancellation currently available in consumer earbuds, along with Personalized Spatial Audio, an optical heart rate sensor, and hearing aid functionality. The IP57 water resistance on both buds and case is higher than the Ear 3's IP54. Battery life runs around 8 hours with ANC on versus the Ear 3's 5 to 6 hours. The tradeoff is the ecosystem and the codec ceiling. AirPods Pro 3 tops out at AAC with no hi-res audio support. The Ear 3 supports LDAC at up to 990kbps and 24-bit/96kHz on compatible devices, which matters on Android and for listeners who want hi-res playback. Most of what makes the AirPods Pro 3 exceptional is locked to iOS. Visibility beyond transparency Nothing's design philosophy is that visible engineering = honest engineering. What CT adds is the layer that even a transparent case cannot show: the spatial relationships between components, the construction of the battery cell, the microphone positions relative to the driver, the full electronics stack on a super thin PCB. Turns out there was more to see.

Recall Radar

Parts Under Pressure

Energy storage and fire hazardsTwo very different products—MaxKare electric blankets and DR Power lithium-ion battery packs—arrive at the same endpoint: fire and burn hazards. The routes of risk differ, between wiring and batteries, but the lesson is the same. Once energy storage or transfer is embedded in consumer goods, small errors don’t stay small. The blanket recall shows how challenging it can be to make warmth controllable in soft goods, where insulation, wiring, and thermostats have to coexist in a flexible package. The battery recall is another reminder that lithium-ion chemistry requires meticulous cell quality, robust protection circuits, and honest thermal margins. When any one of those slips, the failure is a safety event.Multi(dis)function devicesElectrical risk also hides in the seams of multifunction devices. Harbor Freight’s Predator 2000-watt power stations were wired such that the AC outlet is reversed in Emergency Power Supply mode. It’s an archetype of edge-mode complexity: products now juggle normal, backup, and “smart” states, and each state multiplies opportunities for a low-visibility error to become a high-consequence flaw. As devices add features, the boundary between safe and surprising can narrow to a single board trace or firmware path, and users have no way to see the difference until something goes wrong.‍Control system failures in toolsControl systems have also come under scrutiny in the workshop. Positec’s corded chainsaws and pole saws can keep running after the trigger is released, while DEWALT’s grinder flap discs can shed abrasive flaps mid-use. These products span different categories, but the user experience of the risk is similar. When tools don’t respond as expected during a split-second decision, there can be serious consequences. The hazard is the erosion of bedrock assumptions that “off means off” and “parts stay attached.” The humblest elements, like a switch mechanism, an adhesive bond, a retaining ring, become the difference between routine work and a laceration. Strain on structural integrityStructural integrity looms large around the home. Sanven Technology’s Vevor handrails can fail at the weld just when someone is leaning into them, turning an ordinary descent into a fall. Trane’s gas/electric packaged HVAC units, meanwhile, include a fuel gas valve that can open unexpectedly, introducing the possibility of a leak and ignition. One is a support you expect to bear body weight; the other is a control element you expect to meter fuel with clockwork predictability. The shared lesson is simple and old-fashioned: loads and valves are boring until they aren’t. When the structure you trust or the flow you contain goes sideways, everyday motions like taking the stairs or a heater cycling on become risk points.Equestrian equipment failuresEquine gear appears twice, and not by coincidence. Total Saddle Fit’s Western cinches can come apart, and Professional’s Choice bits can break, both leading directly to fall hazards. Horseback equipment sits at the junction of leatherwork, metallurgy, and dynamic loads. It also lives in a niche where brand reputation can stand in for standardized, cross-brand testing. Two recalls in the same cycle suggest a sector that would benefit from broader validation of critical components under real-world stress and contamination.‍Mobility and pressure-based risksMobility and momentum create their own vulnerabilities. Giant’s Momentum Vida E+ e-bikes have fork steerer tubes that can crack, break, or separate. E-bikes marry bicycle simplicity with added mass and speed, and that extra kinetic energy raises the stakes for parts that might have been “good enough” on lighter platforms. Meanwhile, a different kind of physics shows up in Walmart’s mass-market Ozark Trail 64-ounce bottles. Lids can eject forcefully after perishable or carbonated liquids are stored over time. That’s not electronics or metallurgy; it’s pressure and user behavior. Put the wrong drink in a sealed container, pressure builds, and the weakest link fails dramatically. Design and labeling have to account for how products are actually used, not just how they’re supposed to be used.TakeawaysThread these stories together and the through-line is clear. Safety isn’t a silo but a web of how elements like electrical design, mechanical robustness, control logic, and real-world behavior all intersect. Energy wants an escape route, moving parts want to keep moving, and loads will find weak points. Good design anticipates those tendencies, and good oversight catches the misses. This month's roundup is less a set of isolated failures than a set of reminders. Edges matter, basics matter, and the everyday moments when we lean, cut, heat, steer, or sip are where design choices finally tell on themselves.

Design to Reality

Pulling the Thread on Talenti’s Stubborn Lids

Talenti ice cream pints are notorious for how hard they can be to open. The subject of countless articles and forum posts, people have shared tips and tricks for reaching the sweet treat inside. Talenti itself has recognized the challenge their iconic plastic packaging can pose, releasing the Talenti Twist in May 2025 to aid with lid opening. That plastic shrinks when cold is widely identified as a reason the lids clutch the clear tubs so tightly, and letting the package warm up is known to help. But we wanted to take a closer look at the packaging to see if there were any aspects of the design that might explain what makes these tubs so exceptionally frustrating. Armed with our Neptune industrial X-ray CT scanner, we bought a couple of tubs of Talenti from our local grocery store to see what we could find. Scans of Talenti Alphonso Mango Sorbetto and Mediterranean Mint Gelato What our CT scans revealed We conducted 30-minute scans of the tubs in our Neptune, keeping the scans on the shorter side to prevent excessive melting of the gelato inside. CT scans provide a full 3D view of the interior of a product, allowing us to slice into our virtual model in any direction. This lets us take a look at how the threads of the lid and tub actually engage. Corner cutouts of the threads on the Mint and Mango Talenti tubs. We’ve scanned a lot of rigid plastic packaging at Lumafield, and at a first glance the threads on the tub appear sharper than what we usually see. A closer look lets us fully appreciate how narrow–and frankly rather pointy–the threads are. Close-ups of the threads engaging on the two jars. How Talenti compares to other packaging To validate this opinion, we scanned several other similarly-sized plastic containers. Talenti’s plastic packaging is unique among grocery store ice cream, and we were unable to source a similarly frozen comparison. But we procured and scanned a tub of refrigerated pickles, a jar of mayonnaise, and a container of grated parmesan cheese. All of these items have wide mouths, as opposed to bottles that taper into a narrow neck at the top. Assorted plastic tub packaging with close-ups of the threads. Note that the pickle tub was leaking, so brine is visible between the threads. The threads look significantly different. On the grated cheese and mayonnaise packaging, the threads are about the same thickness as the rest of the packaging wall, affording them some flexibility. The pickle tub threads are solid, like the Talenti threads, but much rounder. We can also observe that the Talenti threads are more deeply engaged, the point of the tub thread sitting relatively high on the cap thread. On the other packages, the threads appear to make softer contact. Some of that could come from plastic shrinkage due to freezing, but the sharper shape of the threads seems to make the problem worse. The shape of the problem: thread geometry The threads on the Talenti tub seem to more closely resemble what you’d find on a narrow-mouth rigid plastic container, like the water, ketchup, and soda bottles in the example below. CT cross-sections of a water bottle, ketchup, and soda cap assembly. Even when we compare Talenti to the closest comparison we found–the pickle tub and the soda bottle–the Talenti threads look pointier, for lack of a better word. While the pickle and soda container threads are blunted, the Talenti thread almost seems to be clawing itself into the lid. Thus, thread shape seems a plausible suspect for what makes these lids so stubborn. Close-ups of the thread engagements for the pickle tub, Talenti tub, and soda bottle. Over-torquing on the line Another possible cause is over-torque. Anecdotally, the lids felt easier to remove after the first opening. Back in 2022, Talenti acknowledged an issue in their production line that caused lids to be screwed on too tightly, which they subsequently solved. However, that may have been an extreme case, and the assembly of the lid onto the tub could still cause challenges. We scanned the Talenti tubs before and after opening to see whether that told us anything. Cross sections of the Mango Sorbetto, before and after the first lid opening. Like the new tubs, we scanned the opened tubs for 30 minutes as well. The gelato spent several hours in the freezer in between scans to restore its frozen state. The air bubbles in the sorbetto (and the chocolate chips in the mint) held their positions well enough that we could align on the same locations in the threads across scans. Digging deeper into the scans Examining a closely cropped segment of the thread engagement, it appears the threads are making less contact post-opening. This certainly an imperfect experiment dependent on our personal grip strengths. But we tightened the cap back on as best we could and the CT scan does show that the tub is sealed. A 3D detailed view of the thread engagements on the Mango Sorbetto, before and after opening/re-closing. When we examine the intersection as a 2D slice, we can measure a roughly 14% difference in contact amount between the new and previously opened containers. The scan is a bit rough, given the short duration and presence of ice, but the measurement does suggest that the line machine may be over-tightening the lid. It is also possible that the threads round out as the lid is used, or as the threads are damaged. One of the most popular Reddit recommendations for opening the gelato is to trace the inside of the lid with a butter knife. This could potentially dull the threads and account for why the hack is so effective. 2D slices of the intersection of the lid and tub threads. Why the lids remain viral We cannot say with certainty what makes Talenti jars so hard to open. Our sample size was limited, and we kept scans short to avoid melting, which added noise. What ultimately stands out most to us are the clear visible differences in thread design, compared with other similarly-sized plastic food packaging. It’s also notable that no other major ice cream brands use similar packaging. The other ice creams we encountered when looking for comparisons use either cardboard cartons, or, for larger containers, plastic tubs with lids that peel off sans threads. Most likely, it’s a confluence of factors—thread geometry, closure torque, and cold—that keeps this problem going viral. At a high level, it’s a reminder that small design choices compound into the user experience people remember. But, when the gelato is so good, we’ll wrestle with the lids anyways.

Materials World

Reading the Rock

Before any type of drilling, someone has to read the rock. That sounds straightforward until you consider what reading it requires: understanding how porous it is, how fluids move through it, how it behaves under stress, and whether the formations above it will hold. A core sample pulled from thousands of feet underground is the primary document. The question is how much you can learn from it without destroying it in the process. For most of the industry's history, the answer was: not much. Conventional core analysis required cleaning the sample, drying it, exposing it to chemicals, and flooding it with fluids to measure how they moved through the pore structure. Some tests pushed the rock to mechanical failure to understand how it would fracture. These methods worked, and they still form the backbone of what the industry calls special core analysis, but they remain slow and expensive while also consuming the very material they’re trying to characterize. A comprehensive study of a single reservoir can take years. A toolbox of techniques The analytical toolbox has expanded significantly over the past two decades. Helium porosimetry measures connected pore volume by expanding an inert gas into a chamber containing the dried sample, a reliable and relatively gentle method that leaves the rock physically intact even if the cleaning process alters its original fluid state. Mercury injection capillary pressure testing forces mercury into the pore system at progressively higher pressures to map pore-throat size distributions, a powerful technique for understanding how fluids will move and whether a formation can hold a column of hydrocarbons, but one that contaminates and effectively destroys the sample in the process. Thin section petrography, in which a sliver of rock is ground to translucency and examined under a microscope, reveals mineralogy, grain structure, and the history of how the rock was formed and altered over time. Scanning electron microscopy goes further, resolving pore structures and clay fabrics at scales far below what optical microscopy can reach. What CT adds to the picture What micro-CT adds to this picture is the third dimension. Rather than cutting, dissolving, or mechanically testing the sample, a CT scan passes X-rays through it from multiple angles and reconstructs the internal structure as a three-dimensional volume. Pores, grains, fractures, and cement are visible in their actual spatial relationships, at resolutions down to a few microns. Most importantly, the sample comes out of the scanner intact. That matters for several reasons. Core is finite and often scarce. A plug consumed in a mercury injection test cannot be imaged afterward to understand why the result came out the way it did. CT imaging early in an analysis program allows labs to select which samples are worth committing to destructive testing and which would only generate misleading data, because a fracture introduced during extraction had already compromised the structure. A geoscientist I spoke with at one of the largest oilfield services companies in the world described this as one of CT's most practical contributions: the ability to triage a sample set before anything is destroyed. A CT-derived 3D model of pore structure in a candidate reservoir rock, the same geometry that determines both hydrocarbon extraction potential and suitability for carbon sequestration. Running experiments on a digital twin The 3D model that comes out of a CT scan can also be used directly for simulation. Digital rock physics uses the segmented scan volume as the domain for computational fluid flow models, calculating how fluids would move through the actual pore network of that specific sample under various conditions. SLB now offers digital special core analysis as a complement to physical testing, running virtual experiments varying fluid viscosity, interfacial tension, and wettability scenarios that would take months to execute physically and can be completed in a fraction of the time. The limitations are real, and the field is candid about them. CT resolution today typically reaches a few microns, which means nano-scale porosity common in tight carbonates and shales remains effectively invisible and must be inferred from other measurements. The simulations assume idealized physics and do not yet fully capture the geochemical changes that happen over years of fluid injection or depletion. Digital rock analysis is most reliable when calibrated against physical measurements rather than run independently. Institutions including Stanford's Center for Carbon Storage and IFPEN, the French energy research institute, continue to work on closing these gaps, combining machine learning with CT imaging to predict rock properties faster and with less computational overhead. How the sequence has changed What has changed most visibly in practice is the sequencing of an analysis program. Labs now tend to run non-destructive imaging first, using CT to understand the sample population and identify the best candidates for physical testing, then commit a smaller number of carefully selected plugs to the destructive methods that generate the most critical dynamic data. That amounts to fewer tests, better tests, and more confidence in the results. CT reveals pre-existing fractures in this shale caprock sample, disqualifying it for mechanical testing, while confirming the dense, low-porosity structure that makes caprock an effective seal. Carbon sequestration The same logic applies to carbon capture and storage, where core analysis is structurally similar, but the stakes are inverted. Instead of characterizing a reservoir to extract what’s inside it, geoscientists need to characterize one to ensure it will hold what is injected. Porosity and permeability still matter, but so does the chemistry of how rock responds to long-term CO2 exposure, and whether the caprock above the storage formation will remain intact over timescales that extend well beyond a production cycle. Mercury injection and micro-CT are both being applied to seal characterization, and core flooding experiments with CO2-rich brine at reservoir temperature and pressure are generating the geochemical data that regulators and operators need before committing to a site. Recent research on deep saline aquifers has used CT-derived digital rock volumes combined with machine learning to estimate porosity and permeability for CO2 storage evaluation, illustrating how digital techniques can support site screening while limiting destructive tests. Geothermal energy Geothermal exploration adds another variation, and an instructive twist. In a geothermal formation, porosity is often a liability instead of an asset. Heat moves better through dense, intact rock, and the questions shift from fluid flow to thermal conductivity, fracture connectivity, and how the rock behaves under cyclic thermal stress. EU-funded research programs have explicitly extended digital rock physics from hydrocarbon reservoirs into geothermal applications, using advanced imaging and simulation to understand rock behavior under the coupled thermal, hydraulic, mechanical, and chemical conditions that govern heat extraction. Here, the same imaging methods are being used to assess the opposite properties. Lumafield's Voyager software quantifies pore size and distribution in core samples, generating the data that feeds reservoir simulations and CCUS assessments. Back to the rock Energy exploration has always been a discipline of inference. A core sample is a small physical fragment of a geological system that may extend for miles, extracted under conditions that alter the very properties being measured, and subjected to a sequence of tests that progressively consume it. The analytical methods that have developed around that problem are not a hierarchy with one technique at the top. They are a sequence, designed so that each method informs the next and the most irreversible steps are saved for last. CT fits into that sequence not because it replaces what came before, but because it tells you something about the rock before anything else does.

The Quality Gap

Reshoring's Hidden Constraint: Quality

The numbers on U.S. manufacturing investment are large and getting larger. The Reshoring Initiative counted 244,000 announced manufacturing jobs tied to reshoring and foreign direct investment in 2024 alone. Bain found that 81% of CEOs and COOs reported onshoring, near-shoring, or split-shoring plans that year, up from 63% just two years earlier. The capital is moving, and the facilities are being built. What the announcements tend to underspecify is what exactly companies are (re)building when they reshore. Physical infrastructure is the tractable part, as you can site a facility, install equipment, and run a production line on a predictable timeline. It’s much harder to rebuild the quality infrastructure and correlate institutional knowledge, and most reshoring plans underestimate this aspect. The bottleneck is judgment, not equipment After decades of offshoring, tacit manufacturing expertise in the United States is thinner than it once was. Toolmakers, metrology specialists, process engineers with deep experience in a specific process family, quality managers who have run the same line through a thousand failure modes: these people exist, but are hard to come by. Plants that have rebuilt their production infrastructure often find sophisticated automation and reasonably stable processes, but quality emerges as the bottleneck. The line may be healthy, but without somebody to read it, one can’t be sure. This creates a specific problem for inspection. Traditional quality tools work well when an expert operates them. A CMM reading, a dye penetrant test, an ultrasonic scan: each produces useful but partial data, and the synthesis required to turn those incomplete assessments into confident decisions about part condition is exactly the kind of judgment that takes years to develop. An aluminum casting that passes dimensional checks can still conceal porosity that becomes relevant downstream. An injection-molded housing can meet external tolerances while internal features drift with tool wear. Each of those failures demands a specialist's attention, which leads to a backlog for quality. The more specialists a program needs to function, the more exposed it is when they are unavailable, overloaded, or simply not yet hired. Though these two brackets, printed with different additive technologies, appear similar on the surface, CT scans combined with automated porosity analysis reveals dramatic differences in sphericity and distance from surface for internal voids. Encoding inspection knowledge The standard response to a quality expertise gap is to hire carefully, train aggressively, and accept that early ramp periods will be bumpy. Those are reasonable responses, but they address the symptom rather than the root of the problem. The structural question is whether inspection can be designed to carry expert judgment forward, rather than requiring that judgment to be present at every decision point. There is a meaningful difference between extending expert judgment and encoding it. Extending it means routing every difficult question to a qualified person. Encoding it means defining the question precisely enough that the answer does not require an expert to read it. One approach to this is inspection that examines a part's full internal geometry rather than sampling only its accessible surfaces. Industrial computed tomography, also called X-ray CT, works by rotating a part between an X-ray source and a detector, capturing projections at multiple angles, and reconstructing a three-dimensional model whose values correspond to material density. From that model, engineers can measure internal features, map porosity relative to load-bearing geometry, and track wall thickness at sections that have no external analog. Most importantly, the analysis can be codified. Thresholds, regions of interest, measurement sequences, pass/fail criteria can all become recipes that run on every subsequent part. A senior engineer writes it once, then the line runs it without her. At a new facility still building its quality organization, that leverage is far from marginal. What faster cycles unlocks When Nemak, a high-pressure die casting manufacturer, adopted CT for porosity evaluation, they saw a way to drastically condense their usual workflow. Before CT, characterizing porosity in a casting required lab sectioning, metallographic preparation, and a formal reporting cycle. The process involved 4 specialists, 11-12 hours of work, and a 2-3 day turnaround. With CT, a pair of engineers reached a complete porosity characterization in only 3 hours of a single day. The faster cycle was a welcome improvement, but the more consequential change was in who could act on the findings. A spatial porosity map aligned to CAD puts the engineering question in a form that a process engineer, a quality engineer, and a manufacturing engineer can evaluate simultaneously from the same file. Arguments about whether a finding is relevant become resolvable by measurement. At new facilities where relationships between functions are still forming, that shared evidentiary basis shortens conversations that would otherwise run for days. Corrective action gets implemented in the cycle where the problem was identified, instead of the one after it. CT scans of three aluminum castings show how porosity distribution predicts leakage: the two failing parts carry denser concentrations of irregular, low-sphericity pores along the leak path, while the non-leaker's porosity is sparse and scattered. Ramp periods and the feedback problem Early in a reshoring ramp, or any new production program, yields can be volatile while tools wear in, gauges settle, and operators develop fluency with machines they haven’t run before. The gap between when a parameter drifts and when an engineer detects it is where scrap accumulates. A team that can pull a sample, run a scan, and review results the same day closes that gap faster than a team sending samples to an outside lab and waiting. CT also gives engineers a way to measure drift that surface inspection can’t catch. Wall thickness distributions, porosity location relative to stress-bearing features, and dimensional change at internal geometries are all qualities that can shift before the part fails a functional test. Seeing this drift in progress allows process adjustments before yield losses compound, and creates a documented record that ties parameter changes to outcomes. Over multiple programs, that record becomes a body of institutional knowledge with a different character than accumulated individual experience. It can be searched, shared across sites, and reviewed by engineers who were not present when the decisions were made. Building the infrastructure for quality Reshoring is a capacity-building exercise, and the constraint on capacity is quality judgment applied consistently at scale. New domestic facilities face this acutely: the inspection expertise global customers expect is not uniformly available, supplier ecosystems are being reestablished from scratch, and tolerance for extended ramp periods is low. Inspection infrastructure that encodes expert judgment, compresses feedback cycles, and produces evidence that travels across organizational boundaries is so much more than just a feature of a reshoring program. Without it, you’re getting off on the wrong foot. Done well, inspection infrastructure becomes invisible. That’s not because it stops mattering, but because the problems it prevents never happen in the first place.

Recall Radar

Routine Uses, Real Risks

This month’s recalls cluster where products meet ordinary use, from small appliances that try to hold heat and pressure with too little margin to children’s items that expose compartments or shed parts under routine handling. Bikes and youth machines depend on bolts, bushings, and clearances that must carry real loads, and furniture and rails at home show how geometry decides whether heavy pieces stay put. On the road, software fixes shaped outcomes as much as parts, with updates correcting behavior without new hardware. The broader trend centers on everyday safety, with containment, stability, and current software separating routine use from real risk. Energy edge cases Portable power and small burners failed for the same reason, compact designs that leave little room for heat or pressure to go. INIU’s recall of about 210,000 power banks shows how dense cells and modest protection electronics can trap heat until it escapes the design. Walmart’s Ozark Trail tabletop butane stoves follow the same physics in metal, where a compact burner gives pressure few safe paths when something drifts. McLee Creations’ MyOnlyStyler hair dryer fails a different gate by omitting immersion protection, turning a routine bathroom drop into a shock hazard. In all three, safe behavior depends on giving heat, pressure, and current clear exits that still work after real handling. Clustered failures, common causes Two VEVOR-branded appliances imported by Sanven Technology were recalled this month, a pattern that often points to weak process controls somewhere in the supply chain or at final assembly. The ice crusher can enter a thermal event and ignite, which suggests heat generation and containment slipped outside safe bounds. The handheld steamer leaks or spits hot water, and its cap can come off during use, which shows sealing and retention that do not hold once the system warms and pressurizes. One recall can be a bad run, but clusters from one maker often reflect process failures rather than a single defect. High-speed debris Workshops and backyards told the same story about control. Grizzly Industrial’s heavy-duty planers can let the chip breaker touch the cutterhead, a small miss at the cutting gap that turns chips into high-speed projectiles. Primark’s hand water-balloon pumps fail from the opposite direction, where pressure rises inside a simple body without a defined outlet and the product can rupture. The common risk is unmanaged energy in close quarters, with operators standing exactly in its path. Household hardware Furniture and rails sit inches from people and carry real mass. Casaottima’s thirteen-drawer dressers violate the clothing storage standard and can tip if left unanchored. Two adult bed-rail campaigns, one from Vivohome and another from KingPavonini, point to gaps that can trap a person between rail and mattress. Ningbo Tianqi’s FUFU&GAGA Murphy beds add a reminder that a 215-pound frame can drop during assembly or disassembly if it is not restrained. The danger comes from size and proximity, where a small change in geometry can turn everyday movement into leverage against the body. Child-scale tolerances Children’s products clustered around small-parts access and basic stability. KTEBO’s writing tablet toys expose button cells when the compartment screw does not stay attached. CreateOn’s Pip-Cubes can release high-powered magnets if seams open, turning a drop into an internal puncture risk when magnets meet across tissue. HydroJug’s fourteen-ounce tumblers can shed handle rivets and add a choking hazard, and Primark’s Little Bear soother clips can lose a wooden button and expose a sharp screw. A separate thread ran through sleep products from multiple sellers, including Sofoliana and Glotika from Bosen US, Macardac, and Alinux from Winkids, where low sidewalls and oversized foot openings leave infants vulnerable and no stand prevents falls from elevated surfaces. These items live in kitchens, bathrooms, and nurseries with constant handling and cleaning, so safety comes from closures that stay captive and shapes that keep their limits after real wear. When play becomes serious When wheels start turning, small joints carry the load. Outdoor Master’s children’s and youth helmets fail positional stability and coverage, so a routine crash can land as head impact without the protection the standard requires. Luyuan’s youth ATVs miss the federal ATV suspension standard and add handlebars that can cut on impact. Different products share the same setting, a driveway or trail where forces rise quickly, and small lapses in retention or coverage can turn ordinary motion into injury. Software that drives hardware Vehicle recalls closed the month with a different kind of interface, one where code governs behavior as directly as hardware. Ram’s 2025 and 2026 pickups can lose the instrument cluster to a software fault, removing required brake and shift indications. KTM and Husqvarna’s 390 and 401 lines show a related control issue, with engines that can stall at low rpm until dealers update the controller. Waymo addressed an Automated Driving System behavior that could pass a stopped school bus, and the fleet fix was in place by mid-November. The hardware stayed in place while risk shifted with software version and configuration. Remediation comes with revised loads rather than new parts, which puts software control and validation at the center of road safety. Takeaways Together these recalls point to ordinary conditions rather than edge cases. Heat, pressure, and motion build inside compact products, and small gaps or loose fasteners decide whether that energy stays contained. Furniture and shop equipment show how mass and proximity turn minor errors into harm, while the vehicle campaigns show risk shifting with software even when hardware stays put. What matters is routine discipline in design and upkeep, with clear paths for heat and pressure, fasteners that stay captive, stability that resists tipping, and software that reflects a current, validated build. That is the line between ordinary use and avoidable harm.

Recall Radar

Safety Gaps That Keep Shipping

February's recall list is long and varied, but underneath the different product categories, the same problems keep appearing. Magnets turn up in toys not designed to contain them. Button cell batteries end up in housings that any child can open. Helmets reach consumers without passing the appropriate tests. If any of this sounds familiar, it should. These are the same hazard categories, the same regulatory violations, and largely the same root causes that have defined recall lists for months running. The thread running through nearly all these recalls is child safety. Some of that is expected, as products made for infants and kids carry an obvious obligation to protect them. But several of this month's recalls involve products where the hazard only materializes because a child got to it first. Children are curious, and they live in the same homes as these products. Designing without this in mind is how a lighter becomes an emergency room visit and a decorative light becomes a poison control call. The standard for child safety isn't just about what a product is intended for. It's about the environment it has to exist in. Small parts, big consequences Several recalls this month expose children to ingestion hazards through either loose high-powered magnets or accessible button cell batteries. Three separate magnetic chess game recalls, sold through Amazon under different seller names (Kaiwenshangpin, Zelbuck/n b plus, and Yiruikeji2024), cover roughly 6,800 units with an identical hazard. Loose magnets violate the mandatory toy standard, and when swallowed, they can attract across tissue and cause perforations, intestinal blockage, blood poisoning, and death. Three functionally identical products with the same defect reaching consumers independently points to a systemic gap in marketplace review. The Huaker Magnetic Balls and Rods Sets recall adds another angle, with small balls in a product intended for children under three, violating the small ball ban across about 780 units. On the battery side, SumDirect LED Mini Lights, JJGoo LED Balloon Lights, and Cubimana Island Storm Building Sets all share the same recall root cause. Button cell batteries sit in compartments a child can open without tools, and all three lack the warnings required under Reese's Law. The Prismatic 3D Prints Book Nooks recall is a quieter version of the same problem, with a spare lithium coin battery included loose and not in child-resistant packaging. Together these four recalls span roughly 10,700 units. The technical fix in each of these cases isn't complex. Compartment screws, packaging standards, and required labeling language are all well established. What's missing is verification that the production unit actually meets those requirements. Helmets that don’t hold Three helmet recalls this month cover a combined 43,840 units. The Todson Concord is an adult bicycle helmet; the SAMIT Youth Multi-Purpose and Semfri are marketed to children. All three fall short of mandatory bicycle helmet standards on impact attenuation, positional stability, and certification. A helmet that doesn't hold its position or absorb impact correctly isn't just a regulatory problem. It's a product that gives riders a false sense of protection, and may not do anything useful in a crash. Lighters on the loose Two lighter recalls cover a combined 202,270 units. The Somgem/Yomin Lighters, sold by Elepdv on Amazon, and Royal Oak Enterprises' multipurpose lighters both lack required child-resistant mechanisms, and the Royal Oak recall adds a labeling violation under the Federal Hazardous Substances Act. Child resistance requirements for lighters have been in place for decades. The mechanism is not novel or technically demanding. The Elepdv case is notable because these products also failed to meet pre-market submission requirements, meaning they skipped the process meant to catch exactly this problem before reaching shelves. Heat hazards at home Five recalls this month involve fire or burn hazards from household devices, and together they cover well over a million units. The Dupray Neat Steam Cleaner recall is the largest at roughly 651,000 US units, where a combination of overfilling, corrosion, and pressure valve malfunction can cause the boiler to rupture. The Airova Aroeve Air Purifier (roughly 191,390 units) and the PQL High Bay Linear LED fixtures (roughly 186,520 units) both present overheating risks, though through different mechanisms: the purifiers can ignite outright, while the LED fixtures fail when degrading retaining pins allow the LED board to come loose inside the housing. The Babysense Max View Baby Monitor (roughly 81,800 units) overheats or sparks during charging, a thermal management problem in a tight enclosure that echoes several of January's compact electronics recalls. Durability deficits The Weber grill brush recall is the month's largest at 3.2 million units. Metal wire bristles can detach, adhere to grill grates or food, and pose an ingestion and internal injury risk. Weber is replacing them with nylon bristle brushes, a material-level fix that removes the hazard entirely rather than patching around it. The Andersen Windows recall (roughly 91,000 units) involves opening control devices on 100 Series casement windows that can break or detach after impact, allowing the window to open freely and creating a fall hazard on upper floors. The Clark Associates Lancaster outdoor chair and barstool recall (roughly 158,000 units) is a more straightforward structural failure, with legs that bend or break under routine use. Both are products that see repeated mechanical stress in real-world conditions, and both represent the kind of failure that shows up when testing doesn't account for what the product actually goes through over time. Highway hazards February also brought a cluster of notable automotive recalls. Ford accounts for the majority of them.. The most serious involves certain 2024-2025 F-250 and F-350 trucks where a driveshaft friction weld can fail, causing sudden driveshaft separation and loss of drive power. A separate Ford recall covers a wide range of popular models (F-150, F-250 through F-600, Maverick, Ranger, Expedition, Navigator, and E-Transit) where the integrated trailer module can lose communication with the vehicle while towing, potentially cutting brake and turn signal lights or disabling trailer brakes entirely. Ford is also recalling certain 2020-2022 Explorer, Escape, Lincoln Aviator, and Lincoln Corsair vehicles for windshield wiper motors that can fail mid-use, and certain 2026-2027 E-350 and E-450 vans for a backup alarm connector that can detach and leave drivers without an audible reversing alert. Honda's recall of certain 2024 Acura ZDX and Prologue vehicles involves software errors that can cause the instrument panel and rearview camera display to fail, a direct compliance issue with federal rear visibility standards. International Motors recalled certain 2023-2027 IC Bus school buses for brake lines installed with incorrect hardware that can cause leaks. Land Rover recalled certain 2023-2026 Range Rover Sport and Range Rover vehicles for panoramic sunroof trim that can detach while driving. The three have little in common on the surface, but all involve failures in components that can have real consequences. Takeaways February's recalls are dominated by familiar failure modes recurring at scale. Loose magnets in toys, accessible batteries, helmets that skip certification, and lighters without child resistance are well-documented hazard categories with known solutions and long-established requirements. The volume of units involved suggests the problem isn't a lack of standards. It's a gap between what the standard requires and what gets verified before a product ships.

Recall Radar

Same Mistake, Same Results

April's list is defined by repetition. Six pressure washers were recalled for the same missing safety component, while three children's tower stool recalls were filed in the same week for structurally identical failures. A power bank was recalled for the second time after a first action that didn't reach everyone who needed it. When the same problem shows up across multiple companies in the same month, it usually indicates an underlying pattern. It means something in how these products get sourced, qualified, and brought to market consistently skipped the step that would have caught it. Shortcuts at the source Six electric pressure washer brands had recalls this month for the same reason. Sweetcrispy (2,300 units), Fengrong Tool (500 units), Agiiman (80 units), Patoolio Direct (2,066 units), BAYOTAK USA (360 units), and Le Hao Tool (3,100 units) all shipped products without a ground-fault circuit-interrupter, a component that has been required on outdoor electric equipment for decades and exists specifically to prevent electrocution when water and electricity meet. None of these recalls had reported injuries, which means the enforcement action preceded the harm in every case, and that is the best possible version of how this works. But the fact that six companies reached consumers with the same omission points to something structural, a sourcing and qualification environment where a well-established safety requirement can be skipped repeatedly without anyone catching it before the product ships. Collapsing children's chairs Three separate tower stool recalls landed in the same week, covering Wiifo (9,700 units), TOETOL HOME (3,000 units), and AMZCMJ DGD (130 units), for structural collapse or tip-over combined with openings large enough for a child's torso to pass through. Combined, the three generated 47 reported incidents and 21 injuries. It is likely not a coincidence that these actions were filed together. The TOETOL version had 11 injuries from 18 incidents, and both it and the Wiifo had been on the market for multiple years before the actions were filed. One recall in a product category tends to draw regulatory attention to similar products, and when the category is small enough that the same failure mode recurs across multiple suppliers, scrutiny tends to surface more than one problem at once. Burning batteries Three separate lithium-ion battery recalls this month range from inconvenient to fatal. The Casely power bank reannouncement is the most serious, with 429,200 units recalled a second time after continued incidents following an April 2025 action. Since the first recall, 28 additional incidents have been reported, including one fatality and one fire on a commercial aircraft. A 75-year-old woman in New Jersey suffered second and third degree burns in August 2024 and later died from complications, and the product that caused it was still in circulation because the first recall did not reach everyone who needed to be reached. The VEEKTOMX VT103 power bank recall covers 8,000 units with three fire reports and minor property damage, and the Osprey and S1 robotic pool vacuums recall covers 5,000 units with ten overheating incidents and nine instances of property damage, all occurring during or after charging. Across the three, the failure mode is consistent, but the contexts span a lap, a desktop, and a backyard pool. Low-quality batteries pose a serious hazard everywhere they are used. Lacerations and leaks The FitRx SmartBell adjustable dumbbell recall covers 50,000 units where weight plates can dislodge from the handle during use, with more than 115 reports and at least six injuries including broken toes and lacerations. The selection dial mechanism that determines which plates are locked to the handle appears not to hold its engagement under dynamic load. The Generac portable generator recall covers about 149,400 units where fuel can leak from the carburetor during first fill. The company's own guidance distinguishes between units that have previously been filled without incident and those that have not, which effectively makes the first fueling event a field test for whether the defect is present. For both these products, real-world use reveals gaps in design. Road risks April's automotive recalls span a wide variety of failure modes. Toyota is recalling certain 2025 RAV4 vehicles where front seat bracket welds may be improper, a structural failure in a component that exists to keep occupants in position during a crash. This recall only affected four vehicles, but the increased risk of injury in those few units is serious. Chrysler is recalling engines with two specific part numbers where internal debris can cause engine failure and loss of drive power, with the risk of an engine compartment fire in the worst cases. Altec Industries is recalling certain 2025 and 2026 Digger Derrick and Aerial Device vehicles where improperly installed front-end radar can interfere with the adaptive cruise control and active brake assist systems. The broadest action covers certain model year 2019 through 2024 Jaguar Land Rover vehicles across eight nameplates, where a DC-DC converter microchip fault can cause loss of 12-volt system charging and potentially complete loss of drive power and exterior lighting, a remedy for which is still under development. Across all four, the failure is in a system the driver depends on without thinking about it, and the consequences arrive without warning. Deployment disconnects The Aligned Medical Solutions recall covers two angio pack kit configurations containing a Medline Namic angiographic rotating adapter syringe where the rotating adapter can unwind during use, creating the potential for a loose or complete disconnection between the syringe and manifold during an interventional procedure. The affected kits were distributed between October 2024 and November 2025. The consequences of that disconnection include biohazard exposure, blood loss, infection, and air embolism. Three instances of inadvertent air injection into patients have been reported to Medline, along with one biohazard exposure incident, none resulting in death. Takeaways Several of this month's recalls share the same basic shape. Multiple companies shipped products with the same deficiency through the same channels, and the problem only became visible when enough of them accumulated to draw a response. The tower stool filings and the six concurrent GFCI omissions are the clearest examples, but the pattern extends into the automotive and medical recalls as well, where individual components failed under conditions that were entirely predictable. The difference is scale and visibility. A category-wide qualification gap produces a cluster of simultaneous recalls. A single component that wasn't validated against real use conditions produces one. But the underlying logic is the same, and in both cases something about how the product would actually be used wasn't part of how it was verified.

Materials World

Seashell Architecture

Marine snails don't find their shells, they make them. As they grow, they secrete calcium carbonate to build protective exoskeletons that scale with their bodies. These shells serve as both armor and housing, evolving into highly specialized forms that reflect each species' movement, feeding strategy, and environment. The exterior of a shell hints at its purpose. The internal architecture, how the structure actually works, is sealed off from view. CT scanning changes that. Through nondestructive imaging in full 3D, at resolutions fine enough to reveal growth patterns, material distribution, and the spatial logic of forms evolved over millions of years, we can examine these structures without touching them. The scans below cover six different shell types, each a distinct structural answer to the same basic challenge: how to live and grow inside a mineral shell. Moon snail shell The shark eye, a type of moon snail, builds a smooth, rounded shell with a low spire and wide domed base. It is built for movement. These snails live just beneath the sand's surface, gliding through sediment as they hunt buried clams. The shell minimizes drag, offers concealment, and provides uniform protection without excess weight. The CT scan shows how the internal structure supports that function. The coil is wide and shallow, wrapping around itself in a low spiral rather than rising upward. There are no internal partitions or reinforcing structures, only continuous curvature that distributes pressure evenly across the surface. The entire shell acts like a dome under compression, built to resist force from every angle. The engineering logic is quiet and efficient: nothing added that isn't needed, nothing missing that is. Turrid shell This slender shell belongs to a turrid, a family of marine predators. Its tall spire, narrow profile, and fine axial ridges give it a streamlined shape suited to life in shifting sands and gravel, where maneuverability matters more than bulk. The turrid doesn't rely on heavy armor. It relies on being fast and light. The array of CT visualizations shows the shell from multiple angles: the external surface, a full cross-section revealing density variation, a lower-opacity volume view, a transparent render of the coiling structure, and an interior view of the hollow spiral. Internally, the scan reveals a tight, uninterrupted spiral with no sealed chambers or reinforcing struts. The structure is a continuous coil that maximizes internal volume while minimizing material use. Density mapping shows that mass is concentrated at the apex, where the earliest growth formed a solid core, and in the outermost wall. Everything between is open, hollow, and shaped for efficiency. Material is placed where it counts and nowhere else. Cone shell At first glance, a cone shell appears almost mathematical: clean, conical, unornamented, with a polished surface that conceals its role as a predatory tool. Members of the Conidae family, including this specimen, are hunters that deliver venom through a harpoon-like radula. The shell's streamlined shape supports that behavior, dense and smooth, built to move through water or sand with minimal resistance. CT reveals an interior that adds an unexpected structural logic to the spare exterior. The outer walls are considerably thicker than the interior walls, which are paper thin and delicately bound to a central columella that tapers dramatically from base to apex. In cross-section, these internal braces form a tree-like pattern that distributes force efficiently across the shell's body. The architecture is subtle, lightweight, and strong without announcing itself. Turritella shell The turritella exemplifies structural repetition: each whorl builds directly on the last in a precise logarithmic progression. The spiral rises in stacked increments like a helical column. From the outside it looks regular; inside it is even more so. The CT series exposes the internal logic directly: a cutaway revealing tightly packed whorls, a density-weighted view of volume distribution, a transparent rendering of the continuous spiral, and an interior view highlighting the sealed chambers. CT shows how growth is recorded in each successive twist of the helix, with older whorls tapering toward the apex and culminating in a dense point. Unlike the other shells in this group, the turritella lacks a thick central columella. The structure relies more on the external wall for strength across growth cycles, a pure corkscrew form with the load carried at the periphery. Murex shell The branched murex is heavily armored. Thick ridges carry hollow spines that extend outward as defensive structures. The spines are hollow, which preserves their reach while saving weight, a tradeoff visible in the CT scan that would be impossible to confirm from the outside. The interior shows something equally interesting. At the core is the columella, the spiral's load-bearing axis, reinforced with pronounced internal ridges. These ridges are functional for the animal: as the snail grows, it grips the columella with its body, anchoring against the torque of movement or external force. The ridges provide traction, like a screw thread, allowing the snail to retract deeper into the shell and brace itself. CT makes both facts legible simultaneously: how the shell was built, and how the animal used it. Conch shell Even in its early stages, this horse conch shows the structural logic of a large-scale design. The spire is tall and flared, and axial ridges add stiffness and surface complexity. The growth pattern is less uniform than the turritella's, adapted to the demands of scale rather than optimized for repetition. It is a blueprint that allows the largest horse conchs to reach two feet in length without losing mechanical efficiency. The embedded Voyager viewer lets you move the revolving slice slider and examine cross-sections of the conch shell anchored to its central columella, following how internal structures spiral and shift with each degree of rotation. The shell's curved partitions follow the spiral without sealing off earlier chambers, adding reinforcement without accumulating material. The tradeoff between size and strength is managed by distributing support continuously along the length of the shell rather than concentrating it at any single point. Measuring shell structure CT not only reveals the internal geometry of a shell, it allows us to measure it. Voyager's wall thickness analysis maps exactly how material is distributed and what that distribution reflects about the structural demands of each species' life. In the turrid shell, the thinnest region is the outermost lip of the aperture: the newest part of the shell, freshly grown and not yet reinforced. The columella is the thickest. In the turritella, the thickest points are the joints on the outer wall between successive rungs of the spiral, where the loads from each growth cycle converge. The most precise measurement in this group comes from the moon snail. The scan shows a shallow circular depression near the apex, consistent with the drilling behavior of predatory octopuses, which bore through shells to reach the snail inside. Using Voyager's probe tool, we measured the thinnest point of that depression at 0.098 mm. Without CT, that measurement would be nearly impossible to take without damaging the shell further. Form and function Each of these shells solves the same problem differently. Some minimize material. Others distribute it precisely. Some achieve near-perfect geometric regularity; others adapt their structure to the demands of growth and scale. What they share is that the logic connecting form to function was invisible until imaging made it accessible. For engineers and biologists alike, that access is the point: natural structures that have been optimized over geological time, now measurable in three dimensions, in the same object, without cutting anything open.

Materials World

Speaking in Steel and Sapphire: MING’s 20.01 Series 5

In watchmaking, every material shows up in the final product. You see it in the way light moves across a dial, feel it in the weight of the case, and hear it in the click of a pusher. With our S5 chronograph, the materials arose from years of persistent trial and error, testing what works both technically and aesthetically. MING is a small independent brand. We design and develop everything in-house in Kuala Lumpur, then work with specialist partners in Switzerland and elsewhere to build watches in limited batches. The S5 is our flagship chronograph. It’s a maximalist piece meant to push every part of the design, movement, and materials as far as we could take them. I’m not a trained watchmaker. My background is physics, finance, and photography. But eight years in this business has taught me one thing: you don’t survive by doing what everyone else is doing. Materials are part of that challenge. Over the years, we’ve worked with titanium, tantalum, magnesium alloy, and more. Each one changes how a watch looks, feels, and wears. The S5 is a distillation of that experience. Inside the Movement A movement is mostly brass; it’s stable, easy to machine, and fine detail holds well. Steel takes on higher-wear jobs. Bearings are synthetic ruby, which are both hard and consistent. Some parts need specialized alloys. The hairspring is Nivarox for temperature stability. The balance wheel is made from a similar alloy to hold its moment of inertia. For the S5, we started with a complex chronograph from Agenhor and reworked it. We reshaped the bridges and refinished the surfaces. Earlier models had a gradient of DLC coatings across the layers. They’re small details, but they add depth. The Case Normally a watch case is built like a can: the movement sits inside a middle ring, the bezel goes on top, and the case back closes from the bottom. The S5 is different. Because I wanted integrated lugs with hollows, the case had to be split in an unconventional way. The movement sits inside a donut, the lugs attach from the sides, and the bezel and case back come in from the top and bottom. The seams are hidden in 90-degree orthogonal divisions, so you don’t actually see where it comes together. Four pillars integrated with the bezel hold the whole thing in place, which means the watch uses about half the number of screws you’d normally need. In fact, it stays together without screws at all. They’re only there to lock it down at the end. Weight and Performance Weight in a watch is a choice. You can make it much lighter than it looks, or much denser. That contrast between what you expect and what you feel makes an impression. But it’s not just perception. A mechanical watch runs 24/7, beating three to five times a second. Tolerances are so tight that if one jewel in the escapement is off by 10 microns, the whole thing stops. Materials have to survive that workload, stay dimensionally stable, resist corrosion, and be safe against skin. Every alloy and coating is chosen because it can do that and still deliver the feel we want the moment you pick it up. The Dial The S5 dial starts as a single billet of titanium. We laser and micro-machine it to create fins and facets, then apply a blue PVD coating. After that, it’s lasered again to bring out additional detail. The result is a layered surface that shifts depending on the light. We knew we wanted contrast in the different finishes, colors, and textures to control how light hits the dial. Some of it is determined by what a material will accept. Titanium takes certain surface treatments that steel won’t. Sapphire, on the other hand, can be etched in layers. We prototype, assemble, and decide by eye. Engineering White Lume Standard lume colors are fixed by the dopants in strontium aluminate: blue, green, yellow, orange, and pink. True white needs mixing across the spectrum, but each dopant has its own brightness and decay rate. The blend that looks white to the eye is uneven—heavily weighted toward some colors, reduced in others—and has to stay white as it fades. We kept adjusting until it worked. That’s how we achieved the S5’s Polar White lume. What’s Next We have more than a hundred projects in the works. Some use metals never seen in watchmaking. Others involve multi-phase color-shifting coatings, seamless transitions between metal and sapphire, interference patterns, and animated effects. All of it is about how materials behave and how to make them do something new. For me, materials are so much more than just the medium. They’re at the heart of watchmaking as a pursuit. The S5 speaks in steel, titanium, sapphire, and light. Every part of it is chosen for what it brings to that conversation.

Recall Radar

Stored Energy Meets Soft Spots

September’s recalls offer a clear snapshot of predictable weaknesses in manufacturing, from overheating in compact electronics to pressure escaping at weak seams and mechanisms that fail to stay locked. Most trace to thin thermal margins, stress concentrators, and tolerance drift that better inspection and supplier change control could have caught before scale turned them into headlines.The monthly cameo: power banksPower banks are our recurring headline, and this month HaloLock from ESR fills that slot with 24,000 units recalled in the United States and 9,900 recalled in Canada. High energy density in a pocketable enclosure leaves little room for forgiveness when resistance rises with age or when thermal paths vary from build to build. A pack that looks uneventful on day one can run hot months later under the same use conditions, which is why this category keeps appearing.Electronics that overheatSmall power devices show how little margin many designs keep. Deale International’s Altafit smartwatch charging pad, about 2,900 units, overheats on a nightstand; Viewrail’s LED modules, 2,720 units, overheat and melt inside architectural channels. Even at low wattage, driver stages, rectifiers, layout, clearances, and cutoff placement decide whether heat soaks the assembly or is shed. These products operate in insulated spaces with poor airflow, so packaging and spacing matter more than their size suggests.The same pattern scales up in other everyday appliances and fixtures. Epoca’s Paris Hilton mini beauty fridge, about 110,000 units, relies on a switch that can short and run hot. RYOBI’s pressure washers, about 764,000 units in the United States and 16,000 in Canada, use capacitors that can overheat and burst. Air Vent’s attic fan motors, about 2.9 million units, include a safety cutoff that can short in the very environment it should protect. None of this requires exotic theory: contact resistance climbs as surfaces pit, electrolytics face ripple they cannot shed, and heat accumulates inside enclosures that favor appearance over airflow. When margins shrink, temperature does the rest.Stored energy looking for a seamMakita’s cordless grease gun hoses, at about 62,927 units, can develop pinholes that eject a cutting jet; Drinkmate’s 1-liter PET carbonation bottles, at about 106,200 units in the United States and 5,000 in Canada, can fail under pressure with fragments and sharp pressure waves. Pressure systems rarely fail in the middle of a perfect wall. They fail at bends, crimps, knit lines, thread roots, and transitions where stress concentrates and materials age. Add pulsation, cleaning chemicals, heat, and UV, and a hairline imperfection becomes a release valve with teeth.Mechanisms with no middle groundLoad-bearing gear and security hardware live on yes-or-no performance, not “almost.” DT Swiss carbon wheels, about 6,000 units in the United States and 150 in Canada, can lose structural integrity and trigger a crash. Werner’s Multi-Max Pro ladders, 122,250 units, have locks that can jam instead of fully engaging. C.A.M.P.’s Nimbus Lock carabiners, 12,600 units in the United States and 2,100 in Canada, can break their automatic closing function. StopBox USA’s AR-15 Chamber Lock Pro, only about 300 units, can be forcibly removed. The counts range from hundreds to tens of thousands, yet the rule is the same. A latch, gate, pawl, or bond must deliver the same state every time under wear and variation. If geometry permits a partial condition that feels complete, users will find it at height or speed.Small closures turn routine motions into hazardsSynergy Housewares’ Wolfgang Puck petite tea kettles bring a humble but consequential miss, with about 40,000 units affected. An infuser lid that can drop mid-pour converts a simple motion into a scald risk. It is not glamorous, but retention features, thread profiles, and assembly torque matter. A spring force that looks acceptable on a new stainless lid may weaken with thermal cycling and scale. A working design should tolerate wear, heat, and mishandling without turning into a scald risk.Why the numbers matterScale amplifies quiet misses. Air Vent’s 2.9 million attic motors, RYOBI’s three-quarter-million pressure washers, and Werner’s six-figure ladder population show how a single part choice or tolerance window propagates across product families and seasons. Mid-sized recalls like Drinkmate’s bottles and Makita’s hoses point at specific stress concentrators and aging effects that repeat wherever similar materials and geometries appear. Smaller counts, from Altafit’s pads to StopBox’s locks, still mark high-consequence categories where failure modes are unforgiving.The manufacturing story underneathAcross these families you can draw a clear thread. Thermal margin, derating, creepage, and clearance decide whether switches, capacitors, drivers, and cutoffs stay cool enough to be boring. Stress concentrators at bends, crimps, knit lines, and thread roots decide whether pressure remains contained. Mechanisms meant to be binary live or die on geometry, spring metallurgy, surface finish, and total tolerance through stamping, molding, bonding, and coating. None of this is speculative about any one brand; it is the physics that governs all of them.September’s docket is a reminder that defects rarely hide in the grand gesture. They hide in contact resistance that grew with wear, in a capacitor’s ripple rating, in the braid that thinned at a crimp, in a pawl that allows a false seat, in a lid that gave up a few newtons after heat and scale. Build generous margins into those details and the headlines shrink. Ignore them and the calendar fills, with another power bank joining the list right on time.

From The Floor

The Missing Middle in Battery Manufacturing

When I talk with small businesses, startups, and big OEMs working on batteries, I hear the same story. On one side, there are labs making tens of cells by hand in glove boxes, with careful notes and plenty of craft. That effort can signal feasibility but rarely convinces a VP or investor who needs scalability and repeatability. On the other side, there are lines turning out tens of thousands of cells a day, locked to one product and one recipe. They won’t pause so a small program can run a few hundred cells of five slight variants. The useful ground in the middle is missing. Teams need a way to make hundreds of consistent cells quickly, trace exactly what went into them, and repeat runs after tweaking a single variable. Right now that means bending someone else’s operation out of shape and learning with thin evidence. The team at the Carolina Institute for Battery Innovation (CIBI) at the University of South Carolina (USC) discovered this through open discussions with people across the battery landscape as well as surveys of possible users and other universities and contract facilities. Some universities cover parts of the need but are oversubscribed. Contract plants excel once you know exactly what you want, but they aren’t built for the early, messy questions. This leads to projects leaping from coin cells to full production assumptions, then stalling when real-cell behavior diverges. By the time teams chase large-scale capacity, they’ve often already made costly decisions on ambiguous data. Cylindrical cells set for small pack integration and testing in Prof. Austin Downey’s lab at the University of South Carolina. What engineers actually need When a team asks for a pilot run, they don’t want (or need) a mini gigafactory. They want to keep most variables steady, change only one knob, and walk away with data they can defend in a room where people can say no. They want their engineers on the floor next to ours so the learning isn’t third-hand. They want cells and datasets to travel together so that six months later, a voltage plateau or rise in impedance still maps to a specific weld head, slurry lot, or dry-room dew point. Trust comes from pragmatic traceability. No one needs a full battery passport for early R&D. They need to record which coating lot ran on which web, the calendaring target, the weld-head parameters, and the formation protocol. That’s enough metadata to explain performance, but not so much that analysis becomes the job. Format choice matters. Five- to ten-amp-hour pouch cells expose thermal and mechanical realities without the risk profile of very large cells. Tabs, seals, and stack pressure behave like they do in production, but a mistake won’t shut down the line for a month. From there, many programs step to 18650 or 21700: big enough to matter, small enough to manage, and aligned with industry direction. Dr. Shichen Sun in Prof. Kevin Huang’s lab analyzing new characterization data. Victoria Colon-Laborde and Kyra Barton in Prof. Goli Jalilvand’s lab inspecting punched components for making coin cells for proof-of-concept tests with a partner. Ms. Belinda Agamah in Prof. William Mustain’s lab punching electrodes for Si-anode battery testing. Constraints that shape the work University lines can log data around the clock, but when you’re changing recipes daily, a full MES only slows you down. As we are building the capabilities in CIBI, we are mindful to stay at the pre-A/A level of technology validation, not mass-production PPAP, so our quality system matches our intent. We will barcode every cell and record only the critical variables like coating lot, calendaring target, weld parameters, formation profile. That way, we will always know what changed without drowning in metadata. Metrology earns its place: if a measurement doesn’t alter a decision, we won’t collect it. That lean traceability drives our turn to X-ray CT. Rather than scan every cell, we ask early experiments where CT delivers the biggest payoff—post-stack alignment, weld-head integrity, or post-formation defects—and then deploy it as a decision lever. CT reveals subsurface voids, misalignment, or coating delamination that electrical tests alone can’t catch. By targeting scans where they expose hidden failure modes, we turn images into same-day adjustments and avoid metrology for its own sake. Good data plumbing ties it all together: every plot still maps to a coating lot, web ID, or formation profile without detective work. Where teams stumble Most early projects succeed or fail on materials integration. A partner ships a new powder or separator. You mix, coat, dry, calendar, and slit. Everything looks fine—until a weld or seal reveals a sensitivity invisible in coin cells. A powder that behaves well in small batches may fail when exposed to production solvents, drying curves, or pressure. A separator tolerant of one weld energy may buckle at another. Small process changes can break lifetime. Running mixing → coating → drying → calendaring → slitting → assembly → welding → sealing → formation as a coherent sequence surfaces those sensitivities quickly and in context. If a dataset flags a high tab-weld void rate correlating to rising formation resistance, you adjust the weld head that day—not debate slides weeks later. Data trips teams up too. Spreadsheets multiply, copies diverge, and folders labeled “final” fill with versions. Months later, no one can reproduce the figure that justified the next phase. Design for the human who reviews the data: decide what to collect, make it mandatory, and keep it close to the work. The aim isn’t pretty dashboards; it’s a clear decision someone can trust. Assembling samples for structural battery testing in Prof. Ralph White’s lab. A practical pilot, not a mini-factory There is real demand for a trusted middle ground. Some academics excel at mechanism studies and new materials—work that happens in my research lab and those around USC, which I respect deeply. But, as we are building CIBI, we are drawn to what happens in real cells under real operating conditions, and to building teams and infrastructure so more people can do that work well. We are standing up a pilot line in the Southeast United States for that middle ground. We’ll begin with 5–10 Ah pouch cells (operational in Q4 2026) because that’s where demand is strongest and where we learn fastest without splitting focus. Tabs, seals, and thermal behavior reveal themselves early enough to guide changes with manageable risk. About 6 months later, we will bring production of 21700 cylindrical cells online. That sequence reflects input from safety consortia, internal experience, and government R&D programs. 21700 hits a Goldilocks zone—large enough for meaningful energy-density and thermal studies, small enough to avoid the safety and capital burdens of very large formats. Wide view of students and postdoc’s in Prof. Goli Jalilvand’s lab. Includes pouch cell assembly, using the in-situ Raman and preparing samples for various testing. Three Engagement Patterns Make to recipe. You provide a bill of materials and process windows. We will assemble, form, and test to spec, then return finished cells and a clean dataset. Think of it as “make 300 cells with this conductive cathode additive exactly as written.” The value lies not in the machine list, but in data and cells that agree and can be defended. Design of experiments. We codevelop a matrix—say, anode additive vs. multiple electrolytes—execute builds, handle formation and testing, and deliver results in a format your analysis stack ingests. The aim is clarity: if an additive works only at certain solvent ratios, you see it on one page, not buried across files. Longer R&D partnerships. Over a semester we plan three or four build cycles, layer in physical characterization and failure analysis, and coauthor decisions that turn ideas into IP. This is where CT, cross-sections, and simple mechanical tests converge. A weld fix shows up in resistance data and images, not just in meeting minutes. For data stewardship, we formalize capture, curation, and handoff so results stay useful across iterations. Engagements run days to weeks. Engineers work side by side in the assembly process. Our target remains pre-A and A cells for technology validation. Every cell gets a barcode. Lots and critical settings are tracked. We define capture, curation, and handoff up front so that months later, explanations still link performance back to process variables. If this middle ground sounds familiar, bring your engineers and open questions. We’ll bring a line built for learning and the discipline to keep evidence clean. What motivates the CIBI faculty and staff is being the trusted partner that moves ideas out of the glove box and into real cells. It is gratifying to see a company come in with a hunch and leave with evidence they can act on. Rendering by designers at Little Diversified Architectural Consulting in North Charleston, SC of Phase 2 of the Carolina Institute of Battery Innovation (CIBI), planned for 2028. CIBI’s vision is to become a collaborative ecosystem with university and industry researchers, with 100,000 sq ft. of dedicated battery space, pilot manufacturing for pouch, cylindrical, and prismatic formats, a world-class battery safety lab, and a large battery cycling/performance lab.

Design to Reality

The Pink Tax: Are Men's and Women's Razors Actually Different?

The razor industry has been segmented by gender for more than a century. Gillette introduced the safety razor for men in 1901 and launched the Milady Decolletée, the first razor marketed specifically to women, in 1915. The market bifurcated and has mostly stayed that way, and the price gap between men's and women's products has drawn enough scrutiny to earn its own name: the pink tax. The term describes the pattern where products marketed toward women cost more than comparable products marketed toward men. We CT scanned two of Gillette's current flagship cartridge razors, the Venus Platinum Extra Smooth and the Fusion5, to understand whether the design differences between them justify the price difference, and to trace what engineering choices follow from the different jobs these razors are asked to do. A facial razor and a body razor are solving different mechanical problems The distinction matters before you look at a single spec. Men's cartridge razors are designed primarily for the face, where hair is short, coarse, and densely packed. Beard hair has roughly the tensile strength of copper wire. A facial razor needs to cut efficiently on short, controlled strokes, hold a consistent angle on the jaw and neck, and handle dense material in a small area. Women's razors are body razors. They need to cover far more surface area, in more varied orientations, over curved geometries that change dramatically between a shin, a knee, and an underarm. The hair is finer and longer. The strokes are longer. The grip shifts constantly. These are different mechanical briefs, and the hardware reflects that. What the scans show Both razors use five blades on the same basic cartridge platform. Gillette markets the Fusion5 blades as slightly more widely spaced for better rinsing and clog prevention, but our scans showed no measurable difference in blade spacing between the two cartridges. The more meaningful differences lie in the pivot, the lubricating strip, and the handle. The most consequential difference is the cartridge pivot. The Fusion5 uses a stiffer fore-and-aft pivot optimized for the short, controlled strokes of facial shaving, where you want the blade to track predictably under moderate pressure. The Venus pivot introduces considerably more compliance into the cartridge latch and surrounding plastic, allowing the head to follow convex and concave body curves rather than holding a fixed plane. The scans show this clearly: the Venus attachment interface flexes in ways the Fusion5 does not. That flexibility trades the locked-in precision of a facial razor for forgiveness over larger, less predictable surfaces. The lubricating strips differ in a related way. Both use water-activated strips that deposit a film of hydrophilic polymers and moisturizers ahead of the blades. The Venus strip is wider across the cartridge and deeper front-to-back, giving it more surface contact per stroke. That matters on a leg, where a single pass covers far more skin than a pass across the cheek. The Fusion5 strip is functional but smaller, tuned for a face where the strip-to-blade ratio is less critical. The handles follow from the same logic. The Fusion5 handle is slimmer and weighted for fingertip control, suited to short precise strokes where angle matters. The Venus handle is wider, with pronounced curves and softer grip zones designed to be held in multiple orientations: overhand, underhand, pinch grip. A razor you use on your ankle needs to work in a completely different hand position than a razor you use on your face. What the price difference actually reflects A 2015 report by the New York City Department of Consumer Affairs found that women pay an average of 13% more than men for equivalent personal-care products. And a 2018 study by the US Government Accountability Office confirmed that women's razors and replacement cartridges are often priced higher than men's. Gillette has attributed price variation to differences in technology, materials, manufacturing, and promotional costs. Current retail pricing tells a more complicated story. Venus Platinum Extra Smooth replacement cartridges run roughly $4.75 to $6.00 per cartridge in four-packs. Fusion5 replacements run roughly $4.25 to $5.25. The gap exists, but it is narrower than the pink tax framing typically implies, and pack size and promotions close most of it. Starter kit pricing is broadly similar. The scans support the view that these are genuinely different products built for different mechanical problems. The Venus cartridge has a more flexible pivot, a wider lubricating strip, and a larger oval head. The Fusion5 has a stiffer pivot, a precision trimmer on the back of the cartridge for detail work around the nose and sideburns, and a geometry optimized for facial angles. Whether those differences fully account for any remaining price gap is a harder question. What the hardware shows is that the engineering differences are real, the functional tradeoffs are deliberate, and the two razors are less interchangeable than their shared five-blade platform makes them appear.

The Quality Gap

The Price of Trust: Behind the Takata Recall

Sometime around the year 2000, an engineer at Takata's inflator facility in LaGrange, Georgia, watched an airbag inflator blow apart during a fire test. He heard it from a floor away. Minutes later, a technician came running upstairs, splattered with blood. Her colleague had just been hurt by the device he was supposed to be improving. Within days, the engineer had learned some new vocabulary. What had happened was not, in Takata's official language, an explosion. Their preferred term was energetic disassembly, a prime piece of 1984-style Newspeak. He was told not to use the other word.That instruction is where this story begins. Not because the language itself caused any of the deaths that followed, but because it reveals how an organization decides which information is allowed to travel and which is not. Encompassing 67 million inflators, 19 automakers, 34 brands, 28 confirmed U.S. deaths, the Takata airbag recall strikes most people as a textbook case of corporate greed. But that reduces it to a just-so story. Underneath the simple moral tale we find something more instructive and (for anyone who builds products at scale) more uncomfortable: a failure of information architecture so complete that by the time the right people knew the truth, the wrong decision had already found its way into tens of millions of cars, many of which are still on the road today.Saving a few dollarsTakata had been making seat belts since the early 1950s. It entered the airbag business in the late 1980s and, by the mid-1990s, was looking for a way to compete with the industry's dominant propellant, tetrazole. Tetrazole worked, but it was also expensive. Through a series of acquisitions, Takata had obtained patents from a rocket research facility in Moses Lake, Washington, and its VP of inflator development saw an opportunity in ammonium nitrate, a compound used in fertilizer and solid rocket boosters. It was significantly cheaper to produce and conveniently already the subject of considerable internal chemistry work.The problem with ammonium nitrate is well-known and extensively documented. It’s sensitive to moisture and temperature cycling, and even mild environmental swings can cause condensation to form inside a propellant wafer, causing its structure to break down over time. Autoliv, Takata’s Swedish-American competitor that dominates the market, had tested the design and walked away from it. According to Autoliv's head chemist Robert Taylor, the compound destroyed their test fixture: "We just said, 'No, we can't do it.'" The danger was documented in Takata's own patent filings, which described PSAN-containing propellants exhibiting "significant aggressive behavior with regard to ballistic properties" under thermal and humidity stress. A propellant expert at TRW, which briefly used ammonium nitrate before abandoning it, put it plainly in a 2003 internal document: there were “well-known issues,” creating “conditions that stimulate an explosive response.”This Lumafield CT scan of a Takata airbag inflator recovered from a Volkswagen shows the internal propellant chamber in cross-section. The granular ring of PSAN propellant wafers is clearly visible surrounding the central initiator housing.Takata moved forward with it anyway. Linda Rink, then a senior scientist at Autoliv, recalled that Takata's inflators were roughly 30% cheaper than her company's, a price differential large enough that General Motors asked Autoliv to match it or risk losing the contract. Autoliv refused, and GM switched to Takata. Others followed, and within a few years, PSAN inflators were moving through the supply chains of nearly every major automaker on earth.The write-off on this decision, per unit, was a few dollars. The total liability would eventually reach $9 billion.Takata’s cover-upBy around 2000, the problems were visible to anyone working close enough to the product. Engineers at the LaGrange facility had documented that the PSAN propellant was not behaving predictably. When propellant wafers absorb moisture across temperature cycles (something that happens naturally over the life of any car) the wafer structure degrades and fractures. Fractured wafers have irregular surface area, which produces irregular burn rate. Irregular burn rate generates unpredictable pressure, and inside a steel housing, unpredictable pressure means the housing can rupture and send metal shards flying everywhere.Kevin Fitzgerald, who joined Takata in 1999 as Engineering Manager at LaGrange and eventually rose to VP of Inflator Engineering and Processing, commissioned an internal report catalogued as ERL-144, Propellant Integrity Evaluation, that concluded an official effort needed to be undertaken to address the defect. The report was handed back. In a detailed personal account published through Recall Awareness, Fitzgerald describes 15 years of sustained internal resistance before he resigned in November 2014, the same day the FBI opened its criminal investigation, and turned over to federal investigators the two reports he had personally commissioned. He describes those documents simply as "all that were needed."https://www.youtube.com/watch?v=mOem13H6rMsTakata employees altered and omitted failing test data in reports sent to automakers, deleted rupture results from validation reports, and in at least one documented case, assigned new barcodes to failed test units so they could not be tracked. The falsification, according to DOJ's criminal information and plea materials, was systematic. The lab producing accurate findings was shut down. A desiccant fix, adding moisture-absorbing packets to the inflator housing, was proposed and implemented, but according to Fitzgerald's account, the head of propellant development allegedly discarded desiccant samples recovered from failed units during an investigation, allowing the absence of desiccant to be used as evidence of manufacturing error rather than design flaw. The band-aid, in other words, was also being falsified.Three Takata executives, Shinichi Tanaka, Hideo Nakajima, and Tsuneo Chikaraishi, were indicted in January 2017 on wire fraud charges. Prosecutors described a scheme spanning more than 15 years, with the executives knowingly approving falsified reports to automakers from around 2000 onward. All three are in Japan. There is no extradition treaty covering this class of case, and Japan has not chosen to prosecute them domestically. None of them has faced trial to this day.Honda and TakataThe automakers buying from Takata had no independent window into any of this. Their view of the inflator program was exactly what Takata chose to show them, and no company had more at stake in that arrangement than Honda. Honda had been buying seat belts from Takata since the mid-1980s, when Japanese automakers were building transplant factories in the U.S. and bringing trusted domestic suppliers with them. The relationship deepened into something closer than a typical procurement arrangement. Honda held approximately 1.2% of Takata's shares, a small stake but, in the framework of Japanese keiretsu relationships, a symbolically meaningful one. When Takata entered the airbag business, Honda was its launch customer for the PSAN inflator program and, by far, its largest account. Beginning with the 2001 model year, Honda installed Takata ammonium nitrate airbags in the Accord and Civic, two of the consistently top-selling vehicles in the United States. The financial weight of that relationship, and the cultural obligation that came with it, made the program exceedingly difficult to question and nearly impossible to stop.The keiretsu model was supposed to provide exactly what was absent here: mutual visibility into quality practices, shared values, and an OEM's ability to guide a supplier's strategic direction. What it provided, in practice, was trust. And trust, extended to an organization that had decided at the highest levels not to honor it, is what created the conditions for a quality disaster of unprecedented scale.The only visibility Honda had into Takata was the validation report, which they appear to have taken at face value. By 2004, Honda had received reports of inflator ruptures in its own vehicles. By 2009, it requested new inflators from Takata without notifying NHTSA. In January 2015, the U.S. Department of Transportation fined Honda $70 million for failing to report 1,729 death and injury claims along with related warranty data (not limited to faulty airbags) between 2003 and 2014. Honda's failure to report followed the same instinct that had kept Takata's own engineers quiet for years, playing out one tier up the supply chain.NHTSA hits a wallNHTSA opened its first inquiry into Takata airbags in late 2009, following the death of 18-year-old Ashley Parham in a minor fender bender in a high school parking lot in Oklahoma. Parham's carotid artery was severed by shrapnel from the inflator in her 2001 Honda Accord. The agency's investigation focused initially only on Honda and did not turn its full attention to Takata as the source until 2014, five years and several deaths later.When NHTSA did act, the primary lever it had was a $14,000-per-day fine for failure to cooperate with information requests. That number, set against a supplier doing billions in annual revenue, tells you something about the proportion between the agency's authority and the scale of its mandate. In November 2015, after years of expanding recalls and a rising death toll, the DOT levied a $200 million civil penalty against Takata, still the largest in NHTSA's 54-year history.During this period, according to reporting by Automotive News, Takata retained a panel of former U.S. Transportation Secretaries and former NHTSA administrators as consultants. A proposal reportedly reaching Takata's chairman focused on recruiting allies in Washington instead of accelerating repairs. Takata knew its time was running out, so they pulled out all the stops to prolong the inevitable. This revolving door is central to understanding why the recall moved so slowly for so long.The issue that matters more for consumers and will have broad implications for as long as it stands hinges on NHTSA’s mandate. NHTSA can order a recall, but it can’t compel an owner to bring their car in. In a country where roughly one in three recalled vehicles across all categories goes unrepaired, even for the most serious defects, the regulator was trying to reconstruct from the outside what the people inside the supply chain had known for years, and then enforce a solution to a problem it lacked the authority to fully address.Recall reckoningThe recall has continued to mount ever since. Priority groups were established from early on, sequenced by inflator age and geographic exposure to heat and humidity. This was the best NHSTA could do within a recall model that has no framework for coordinating a defect involving tens of millions of units across nearly two dozen manufacturers simultaneously. The Alpha inflators, an early subset of PSAN units installed in 2001 to 2003 Honda and Acura vehicles, were eventually found to have a rupture probability of approximately 50% on deployment. Honda issued a Do Not Drive warning for those vehicles in February 2023, more than twenty years after the propellant decision that caused it.All in all, the recall touches 67 million inflators in the U.S. across the entire industry: 19 automakers and 34 brands, from Honda Civics to Ferrari 458s to early Tesla Roadsters. Takata pleaded guilty in January 2017 to wire fraud and agreed to pay $1 billion in criminal penalties, $850 million to reimburse automakers, $125 million to victims, and $25 million as a federal fine. Six months later, under $9 billion in total liabilities, Takata filed for bankruptcy, was acquired by Key Safety Systems for $1.6 billion, and was rebranded as Joyson Safety Systems.Takata is no more, but it’s still too soon to close the book on the recall. In February 2026, Stellantis issued a Do Not Drive warning for 225,000 additional vehicles, Chryslers, Dodges, Jeeps, some built as recently as 2016. NHTSA reports that approximately 98% of recalled Takata vehicles have been repaired as of early 2026, the best completion rate the agency has achieved for any recall in its history. This is a testament to valiant (and creative) efforts including door-to-door outreach, Facebook campaigns, and graphic public notices that went well beyond anything NHTSA had attempted before. But at 67 million vehicles, 98% still leaves approximately 1.3 million unrepaired cars on U.S. roads.Costs of not knowingThe manufacturing supply chain runs on trust by necessity. No OEM can verify everything its suppliers make, and no regulator can see inside every component that moves through commerce. The question is not whether to extend trust but what happens when trust is the only thing standing between a known defect and the people it could eventually kill. In Takata's case, the answer unfolded over 20 years and claimed 28 lives.The more fundamental problem the recall exposed is not that Takata lied, though it did, systematically and for a long time. It’s that the systems surrounding Takata, the equity relationships, the validation protocols, the regulatory frameworks, were all designed to process information that suppliers voluntarily produce. None of them was designed to generate ground truth knowledge of what is truly happening at the component level. When a supplier decides to withhold or conceal, those systems prove powerless. They’re not broken. They just weren’t built for that.Quality, in the end, is an information problem. Organizations that figure out how to get accurate data about their products and supply chains, independent of what anyone chooses to report, will be the ones that catch the next Takata before it becomes a tragedy. The ones that don't will be waiting, as Honda waited, for someone with knowledge of what’s happening to find the courage to share it.

Recall Radar

Tolerance Tested

This month’s recalls cut across a cross-section of everyday use. Spanning everything from batteries and bikes to children’s products and vehicles, the volume of recalls this month makes patterns easier to see. Read as a whole, the list reveals familiar failure modes under ordinary conditions where thermal margin disappears in tight packaging, interfaces shoulder more load than the spec anticipates, and small components drift out of tolerance and undermine system performance. Packed cells, thin margins November’s battery recalls span devices of every size. Belkin pulled portable power banks and wireless charging stands for pack overheating, about 83,500 units in the United States and 2,385 in Canada. Knog recalled bicycle lights for the same hazard at roughly 3,790 units. Overheating can start with variation in cell quality, defects introduced during cell or pack assembly, weak battery management systems, or integration choices that leave the pack without a clear thermal path. These products live in pockets, on nightstands, and on handlebars, so a single fault can turn into fire or toxic smoke in occupied space. Mechanisms on the move Mechanical limits surfaced in everyday gear, and the modes were not subtle. Lezyne bicycle floor pumps, about 7,500 in the United States and 680 in Canada, can launch the canister from its base under pressure, putting users and bystanders in the flight path. Spartan riding mowers leave room for steering arm dampers to be installed the wrong way on about 650 machines, a setup that produces bounce and loss of control. STIHL’s BR 800 backpack blowers add a rotating failure with fan wheels that can break apart on about 47,800 units and cause lacerations. The throughline is a single dependency at the point of action where a modest interface decides whether force is guided or let loose. Saddles under stress Peloton’s Original Series Bike+ returned to a familiar stress path. The seat post assembly can fracture under load, and about 833,000 bikes are included. In 2023, Peloton recalled the original Bike for the same basic failure mode, a larger campaign that covered roughly 2.2 million units. Taken together, the two actions show how repeated, off-axis riding loads concentrate at the saddle and post, and how small margins at that interface can scale into very large field populations. Tolerances for tiny users Child-focused recalls centered on restraint, retention, and access in ordinary play. AliExpress’s convertible strollers violated the stroller standard when the restraint system can fail. Only about 15 units are involved, yet child safety is judged by quality, not scale. Little Partners’ Grow ’N Stow folding learning tower reported platforms that can collapse across about 9,780 units. Konges Sløjd’s three-wheeled scooters can lose the left front wheel on roughly 50 units and turn balance into a sudden drop. Two toys raise ingestion and choking hazards, including Bettina doll sets that expose button cells on about 380 sets and Inkari plush alpaca toys for under-three use with eyes that can detach on about 64,000 units. Across these cases, safety depends on restraints, platforms, wheels, and small parts that remain secure through real handling rather than only in ideal tests. Strain on simple things Several household goods looked simple but carried real hazards. F&F Fine Wines recalled Kirkland Signature Valdobbiadene Prosecco DOCG for bottles that can break or shatter, about 941,400 bottles. H-E-B’s 12-pack Destination Holiday Glow bracelets include a green stick that can leak an irritant liquid, about 6,600 packs. Kroger’s Halloween-themed skeleton candles include flammable ornaments, about 3,680 units. Risk tracks variation, not complexity. Glass under pressure demands uniform walls, sealed chemistries depend on clean continuous bonds, and anything near a flame must use materials that do not ignite. Small failures, big consequences On the automotive side, recent recall campaigns focus on fuel containment, traction batteries, and braking electronics. Hyundai’s 2020–2023 Sonata and Kia’s 2021–2024 K5 cite a damaged check valve that can let air into the fuel tank, allow it to expand against hot exhaust components, and ultimately melt. Ford’s 2020–2024 Escape PHEV and 2021–2024 Lincoln Corsair PHEV list manufacturing defects in high-voltage cells that can short and fail. A separate action for certain 2025 Maverick and Escape vehicles points to an electric brake booster ECU cover that can overheat the circuit board and disable ABS, electronic stability control, traction control, or brake assist. Small parts carry big consequences on the road, and a single faulty valve, cell, or cover can put everyone nearby at risk. Takeaways November underscores that reliability lives at interfaces. Thermal margin around cells and chargers determines how quietly products age. Small joints carry outsized safety loads, from coaster hubs and pump canisters to seat posts and fan wheels. In children’s products, batteries and small parts must remain inaccessible after real wear, not just on day one. In vehicles and large systems, modest components shape outcomes for many people at once. The mix this month also shows that scale does not track complexity. A simple interface repeated across a product family can create a very large field count, while a small run in child gear still matters because the harm is concentrated. The pattern is consistent across categories and it points to the same place: the ordinary connections that decide how products behave when they leave the bench.

Recall Radar

Too Hot, Too Sharp, Too Loose

The broad range of product recalls this month serves as a useful snapshot of where manufacturing processes and quality control can fall short. From blade exposure to electrical fire risks, many of the issues could have been caught through more comprehensive inspection or rigorous design validation.Batteries in the spotlightSeveral recalls involved lithium-ion batteries, pointing to a persistent challenge as these power sources continue to proliferate across consumer devices.FENGQS recalled 100 e-bikes sold on Amazon due to batteries that can overheat and ignite. Consumers are being instructed to mark and photograph the battery before disposal, and to work with local hazardous waste facilities to ensure safe handling. VIVI also recalled 24,000 e-bike batteries, offering free replacements.More troubling is a recall from Transpro, involving 700 electric scooters that were sold with unauthorized UL certification labels. In this case, overheating batteries were linked to $200,000 in reported property damage. Consumers are being offered a refund or a free replacement, but must disable and dispose of the scooter in accordance with hazardous waste regulations.Smaller devices aren’t immune either. Tomauri's iStore power banks, with over 14,000 units across the US and Canada, were recalled due to a similar overheating risk. Even bug zappers were affected. iMirror recalled nearly 30,000 units of its electric swatters due to rechargeable batteries that can catch fire after extended use.While none of these recalls are new in type, their frequency reflects how battery integration continues to be a complex and high-risk area for manufacturers. Thermal management, quality control of battery cells, and supplier verification remain critical.Mechanical hazards still commonSome failures are more physical than chemical. Leatherman recalled 17,000 Charge Plus and Charge Plus TTi multi-tools due to a knife blade that doesn’t fully retract. RIDGID recalled over 70,000 framing nailers for a misfiring defect, where nails could discharge when only the trigger is pulled, bypassing the intended two-step safety feature. And in the kitchen, IKEA’s garlic press was pulled from shelves after reports of metal pieces detaching, with over 54,000 units affected across North America.These kinds of issues typically stem from minor design decisions that don’t account for real-world usage. They’re also relatively straightforward to prevent with sufficient validation and testing, including mechanical fatigue analysis and tolerance checks.Products for children and familiesTwo recalls this month centered on magnetic ingestion hazards, a category where the risks are well-known and potentially life-threatening.Tegu’s magnetic floating stackers were recalled because the magnets can loosen and detach. Pura Scents issued a larger recall—over 850,000 units—of smart diffuser covers for a similar risk. In both cases, the danger isn’t the size of the product, but the strength and accessibility of the internal magnets. When swallowed, they can attract across tissue and cause serious internal damage.These incidents continue to reinforce the importance of secure component housing and long-term wear testing, especially in products that may end up in the hands of children.Structural failures and overpressure incidentsStructural integrity also featured in this month’s list. Apollo electric scooters (790 units) were recalled due to a potential crack in the stem weld, which could lead to a fall. CasaClean steamers were pulled for leaking boiling water, while Winston Products recalled 3.6 million “burst-proof” expandable garden hoses for doing just the opposite, with a potential for bursting that creates both impact and hearing hazards.All of these point to common stress failure modes, like poor welds, insufficient thermal management, or misaligned mechanical tolerances. These are areas where non-destructive internal inspection techniques like CT scanning can significantly reduce the chances of an issue reaching customers.TakeawaysThis month’s roundup underscores the diversity of risks that can emerge in products after shipment, and how even small design or production missteps can have large downstream consequences. Mechanical issues like misfires and leaks remain common, and battery-related failures continue to demand attention, especially as energy density increases and safety tolerances shrink.These recalls serve as a reminder that reliability isn’t just about avoiding mistakes; it’s about building processes that surface them early, when fixes are still cheap and private. The cost of missing that window shows up later, in returns, reputation, and sometimes, real harm.

Scan of the Month

We CT Scanned a Bag of Chips and 3D Printed the Results

This project did not begin with a serious engineering question. It began with a bag of Doritos, a Neptune CT scanner, and the reasonable suspicion that chip geometry is more interesting than it looks. We were right, and we went further than we probably needed to. Here is what we did: we CT scanned bags of Cheetos, Doritos, and Ruffles, rendered them in full detail, 3D printed individual chips from the scan data, and painted them by hand until they were indistinguishable from the real thing at arm's length. The whole process took weeks. The chips were not edible at the end. Why these three Cheetos, Doritos, and Ruffles each present a distinct geometric challenge. Ruffles have a repeating ridge pattern that varies subtly across each chip. Doritos are roughly triangular but with a surface texture of seasoning powder that gives them a granular, irregular topography. Cheetos are the hardest: the puffed corn structure is chaotic, full of voids and irregular protrusions, and the craggy orange exterior bears no resemblance to any engineered geometry. CT is the only practical way to capture all of that nondestructively and in full three dimensions. We also scanned the full bags. Bags of chips are famously mostly air, which the scans confirm, but the CT data also captures every crumb and fragment at the bottom of the bag with the same fidelity as the intact chips above them. From scan to model Each scan produced a detailed digital reconstruction of the chip's internal and external structure. We exported the scan data as STL files directly from Voyager and brought them into Blender to add color and surface texture, building out a complete digital replica of each chip down to the seasoning distribution. You can download the STL files and explore the scans yourself through the links below. 3D printing We printed using a stereolithography resin printer, chosen specifically because SLA excels at the kind of fine surface detail the scans captured. We selected one chip from each variety, exported its STL, and ran the prints. Fifteen hours of print time, three isopropyl alcohol baths, and a final cure at 70°C later, we had three plastic chips that were geometrically exact replicas of their originals. Then one of our interns painted them. Starting from a white base coat, they worked up the color and texture of each chip by hand. The Dorito took the longest. The Cheeto looked disturbingly accurate. Explore the scans The Voyager links below let you rotate and examine each scan in full. You'll need a free account to get started. Once you're in, use the bookmarks on the left to navigate and the range map on the right to adjust the visualization. The STL files are available to download if you want to print your own.

Design to Reality

What Are Counterfeit Batteries?

18650 lithium-ion batteries are among the most common in the world. Their wound design and standard form factor makes them fast and economical to produce at scale, with around 5 billion being manufactured every year. But the popularity and standardization of these cells creates an opportunity for bad actors to capitalize on the quality efforts of reputable manufacturers.Counterfeit 18650 cells move through fast-moving marketplaces where listings turn over quickly and a convincing external wrap substitutes for provenance. 18650s generally all look the same on the outside, given their defined can size of 18 mm by 65 mm. The truth, however, lies within, where the jelly roll reveals dangerous shortcuts.Counterfeit Technique 1: ImpersonationThe clearest counterfeiting comes from direct impersonation. Some 18650 products copy a specific, reputable model in name, appearance, and positioning. In our Battery Quality Report, we examined cells from Amazon vendor SOOCOOL that closely mimicked the appearance of Samsung’s 30Q 18650 product. The cells were wrapped in a strikingly similar pink color, and featured similar printed warnings advising against use in vaping devices. The listing title “Authentic 30QP Rechargeable 3.7 V 18650 Battery Flat Top, Real 3000mAh(2PCS, with Free Plastic Case),” also clearly leaned into the Samsung 30Q identity, and they were priced at a premium, costing more than the actual Samsung batteries but available on the more convenient platform of Amazon. This particular listing has since been removed, but countless others can still be found.The Amazon listing for the counterfeit 30Q cells from SOOCOOL. The listing has since been removed.Initially, the batteries also seem to perform similarly to the real Samsungs, their measured capacities landing in the same neighborhoods. However, our industrial X-ray CT scans told a different story. Anode overhang, which should sit near 0.5 mm to maintain a safety margin as the cell ages, came up short in several cells, with one sample even showing negative anode overhang. Insufficient anode overhang is a defect that can meaningfully contribute to uneven lithium plating, which in turn can result in performance problems and even internal shorts down the line. The edge alignment of the SOOCOOL cells also deviated wildly. Anode and cathode layers should be as straight as possible in a wound lithium-ion battery, and the wavy patterns of the SOOCOOL batteries indicate poor manufacturing process controls. The quality concerns that the CT scans surface confirm these cells are counterfeits.The color of the counterfeit cell is an obvious imitation of the actual Samsung.Counterfeit Technique 2: MisrepresentationLow-cost cells that print impossible specifications engage in a different kind of misrepresentation. Two brands in the set scanned for the Battery Quality Report, Benkia and Maxiaeon, had “9900 mAh” printed on their wraps, a physically impossible figure for the form factor. The measured capacity for the both brands sat around 1200 mAh, and the scans showed why. Large portions of the interior had little or no active material. These cells do not impersonate a particular brand, but they inflate performance claims beyond plausibility. Packs designed around those claims will deliver less energy and age unpredictably because the interior cannot support the promise on the label.Comparison of an authentic cell, a counterfeit cell, and a misrepresented low-cost cell.A Gray Area: RewrappingSeparate from these two groups are rewrapped batteries. In most cases they begin as legitimate OEM cells that have been relabeled for distribution under a new brand. Many perform as expected because the core cell was designed and built by a competent manufacturer. But risk comes from uncertainty about the exact source. Rewraps can include surplus lots, cells that failed OEM QA, or units that were disassembled from larger salvaged battery packs. The new wrap hides that history, so the question becomes not whether rewraps are inherently poor, but whether any given batch matches the underlying standard. In our scans, several rewraps generally mirrored mainstream OEM geometry, though the anode overhang distributions were not as tight. One of the three rewrap brands we scanned was also markedly worse than the other two, illustrating the variability rewrapping can hide, though even the worst rewrap cells were still much higher quality than the low-cost/counterfeit cells. Population behavior separates these groups more cleanly than any single measurement. The SOOCOOL set scattered more broadly and stood well apart from genuine Samsung samples, even though some individual cells looked acceptable in isolation. The spec-inflated low-cost brands had the widest distributions, indicating exceptionally poor process controls. Rewraps generally clustered tighter, with outliers that reflected the uncertainty of source rather than a systematic departure from good practice. Distribution of median and minimum anode overhang measurements per cell.Can you trust the batteries you’re buying?Validating 18650 supply is essential because these cells are then buried into larger electronic assemblies. With numerous potentially problematic cells clustered in hard-to-access areas of devices, failures become hard to identify and address. Unfortunately current oversight of lithium-ion batteries is insufficient, focusing on hazardous material content, shipping standards, and coin-cell child safety. All are important, but battery quality is going largely untraced. The enormous size of the market makes it easy to hide gray-market activity. Questionable vendors can vanish and reappear under new listings when problems arise, as we saw firsthand.Counterfeits and dubious claims will continue to circulate as long as batteries remain ubiquitous and profitable, and they flourish in the gaps left by limited oversight. That reality makes it all the more important to recognize that every can is not created equal. When billions of cells enter the market each year, some will carry the design discipline of a reputable manufacturer and others the shortcuts of a dishonest one. Understanding the difference is not just about calling out bad actors, but about seeing how trust in the battery supply is built (or eroded) one cell at a time.

The Quality Gap

What Counterfeit Apple Products Look Like on the Inside

Counterfeit electronics have gotten good enough that the outside tells you almost nothing. The logo is right, the packaging is close, the weight feels approximately correct. Some fakes even function well enough to pass a quick test. The problems are internal, and they range from reduced performance to genuine safety risk. We CT scanned two specimens of counterfeit AirPods Pro alongside a genuine pair, and a counterfeit MagSafe 2 power adapter alongside Apple's original. What the scans reveal is not just corner-cutting. In a few cases it is active deception, engineering designed specifically to mislead rather than to perform. AirPods Pro: batteries The genuine AirPods Pro house custom button cell batteries in each earbud, sized and shaped to fit the available space precisely. Both counterfeit specimens use lithium-ion pouch cells instead: rectangular pouches crammed into circular cavities rather than engineered to fit them. Pouch cells are not inherently unsafe, but the fit matters. A battery that is not properly constrained within its enclosure is more vulnerable to deformation under stress, and a deformed lithium-ion cell is a thermal event waiting for a trigger. AirPods Pro: circuitry The genuine AirPods use a combination of rigid and flexible printed circuit boards, packed densely to make use of every available millimeter. The counterfeit versions use simpler electronics assembled from off-the-shelf components. The practical consequence is visible in the scan: fewer microphones, less control circuitry, and less room for the processing that determines sound quality and noise cancellation. The counterfeits look like AirPods. They do not work like them. AirPods Pro: build quality Two findings here are worth dwelling on. One of the two counterfeit specimens has no wireless charging coils at all, meaning it cannot charge wirelessly despite appearing identical to a product that can. The other has coils but lacks the magnets that align the genuine case with an Apple Watch charger. The more revealing finding is the weights. Both counterfeit specimens contain small internal masses with no functional purpose. They are there to make the product feel heavier, to replicate the heft of genuine AirPods in the hand. That is not a manufacturing shortcut. It is a deliberate design decision to deceive the buyer at the moment of first handling, before any functional test can be run. MagSafe 2 power adapter: power management The stakes are higher with a charger than with earbuds. A counterfeit AirPod underperforms. A counterfeit charger can damage the device it is connected to or start a fire. The genuine Apple 85W MagSafe 2 adapter has a sophisticated internal power management system with filtering components that condition and regulate the power delivered to the laptop. The counterfeit's circuitry is substantially simpler and lacks those filtering stages. That simplification affects both performance and safety: unfiltered power delivery puts stress on the connected device, and a charger that cannot regulate its own output properly tends to run hotter. MagSafe 2 power adapter: thermal management The genuine charger uses a relatively thin heat sink that wraps around most of the transformer, distributing heat across a larger surface. The counterfeit uses a heavier but geometrically simpler heat sink that covers less of the transformer. Apple's design requires more manufacturing steps to produce and provides more even heat distribution. The counterfeit's design is more prone to hot spots, a problem compounded by the fact that its less sophisticated transformer generates more heat to begin with. MagSafe 2 power adapter: the fake ground This is the most serious finding in the comparison. The genuine MagSafe 2 adapter has a grounded pin that connects to grounding circuitry inside the charger, accessible via an optional three-prong plug. The counterfeit has a pin in the same position that looks identical from the outside. The scan shows that the pin is not connected to anything inside. It is cosmetic. A user who plugs this charger into a grounded outlet believing it to be grounded is not grounded. The charger is performing the appearance of a safety feature it does not have. What the scans show Performance differences between genuine and counterfeit products are expected. The internal weights in the AirPods and the fake ground pin in the charger are something different: engineering effort directed specifically at deceiving the buyer rather than serving them. CT makes that visible in a way that no surface inspection can. The outside of these products was designed to pass scrutiny. The inside was not.

The Quality Gap

What Food Manufacturers Can't See

A nylon staple is about the same density as an M&M. That similarity is enough to defeat a metal detector, confuse a 2D X-ray, and pass through a standard visual inspection. If one ends up in a bag of candy, the first person to find it is probably a consumer. That is the foreign material problem in food manufacturing, and it is more common than most people outside the industry realize. Production environments are imperfect. Packaging lines introduce debris. Equipment wears and sheds fragments. The question is not whether contamination is possible but whether your inspection process is capable of catching it before it ships. The limits of conventional detection Metal detectors catch metal. Magnets catch ferrous metal. Traditional 2D X-ray catches dense foreign objects against low-density backgrounds, which works well for a bone fragment in a chicken breast and less well for a plastic shard in a chocolate bar. The failure mode in each case is the same: the method is looking for a contrast that may not be there. When a contaminant's density is close enough to the product surrounding it, it disappears into the image. Industrial CT scanning addresses this with three-dimensional density data rather than a two-dimensional projection. Instead of casting a shadow and asking whether something looks wrong, CT builds a volumetric model of the product and lets you examine any cross-section from any angle. A nylon staple embedded in candy is visible not because it casts a distinct shadow but because its density and position are fully resolved in three dimensions. Porosity and what it reveals Foreign material is the dramatic failure mode, but it is not the only one that CT makes visible. Porosity analysis, the quantification of internal voids and air pockets, gives food manufacturers a different class of information about product consistency. In a candy bar, small air pockets formed during production affect texture and, in some cases, shelf life. In a baked good, internal void structure can predict structural integrity under distribution stress. In a sealed package, void analysis can reveal delamination in multi-layer materials before the seal fails in the field. None of these are visible from the outside. All of them are measurable in a CT scan. The practical value is not just catching defects after they occur. It is understanding where in the production process they originate, which allows manufacturers to address the root cause rather than inspect their way around it. Seal integrity and packaging Packaging is where food quality decisions compound. A product that leaves the line in perfect condition can arrive compromised if the seal fails, the material delaminates, or fill levels are inconsistent enough to create headspace that accelerates spoilage. CT gives a nondestructive view into all of these: gap detection in heat seals, structural inconsistencies in multi-layer packaging materials, and fill-level verification across units. The alternative is destructive testing on sampled units, which confirms that some products were acceptable at the moment of testing and says relatively little about the rest of the batch. Recall calculation Consumer trust in food brands is built slowly but lost quickly. A recall is much more than a logistics problem. It communicates loud and clear to consumers that the manufacturer's quality system failed to catch something that should have been caught. The reputational cost outlasts the operational one. CT-based inspection does not make recalls impossible. It makes them less likely by moving detection earlier in the process, where the cost of catching a defect is a rejected unit rather than a market withdrawal. For food manufacturers whose brand equity depends on consistency and safety, that shift in where detection happens is the real value.

Scan of the Month

What Medical Connectors Have to Get Right

A medical connector looks simple from the outside. It joins one tube to another, allows fluid to flow, and holds everything in place. But the engineering required to make that happen reliably, at scale, in a sterile environment, under repeated use, is anything but simple. CT imaging lets us examine these components at the scale where the real design decisions live. Here is a look inside six connector types, with attention to how each works and where each can fail. Luer lock The luer lock is the most common connector standard in medical devices. It joins two components with a quarter-turn twist that engages a threaded collar, creating a leak-free seal for fluid or gas transfer. Our scan shows the external fitting and the interlocking thread geometry in cross-section, including the flat disk that prevents backflow and dripping when the connection is broken. Also visible: a thread that has fractured under mechanical stress. That crack would be undetectable from the outside. Using Voyager's revolving slice plane, engineers can examine the joint from any angle and track crack propagation before it becomes a failure. T connector A T connector splits a single fluid path into two. The scan shows the connector's external geometry and, cropped to the midplane, its internal channel structure. For a device like this, dimensional accuracy is not a tolerance band to hit during validation and forget. It is a continuous requirement across every unit produced. CT offers a nondestructive path to verify internal dimensions on assembled parts without cutting them open or rendering them unusable. Double hemostasis valve Y connector This connector handles several functions at once: splitting flow across a Y-body, controlling backflow through hemostasis valves at the lower ports, providing secure attachment at the top via a rotating male luer lock, and maintaining a leak-proof seal through a series of O-rings throughout. The scan makes all of it legible simultaneously — the Y-body geometry, the valve positions, the seal locations, the sideport access point. For a component with this many interdependent features and no tolerance for assembly error, CT-based verification is the only practical way to confirm that everything seated correctly without destroying the part. Luer activated valves Luer activated valves open only when connected to a matching luer fitting and close automatically on disconnection, preventing both contamination and leakage. The mechanism depends on precise coordination between an internal valve stem, a seal, and accordion-shaped spacers that control the opening and closing sequence. Our cross-sectional scans show each component differentiated by density and mapped to distinct colors in Voyager's attenuation range visualization. Two valves are shown here, and they work differently from each other: one uses the mating connector to push the silicone sleeve into a wider chamber where it can open; the other uses a sliding carriage to keep the fluid path clear while a silicone spring closes the opposite end on disconnection. The silicone pre-pleating differs between them for the same reason. Both cost less than a dollar. The engineering inside is not commensurate with that price. 3-way stopcock A 3-way stopcock routes fluid or gas through three ports from a single source, with a lever that selects which pathway is open at any given moment. The scan shows the internal channel geometry and the lever mechanism. Voyager's porosity analysis goes further: it quantifies the size and distribution of voids within the lever itself, surfacing the subsurface consequences of the injection molding process. Voids in a stopcock lever are not cosmetic. They are structural weaknesses waiting for a stress event. Aseptic connector An aseptic connector creates a sterile link between two fluid systems without requiring traditional sterilization of the connection point. That function places an unusually high burden on dimensional fidelity: the connector has to perform exactly as designed, every time, because any deviation from specification creates a contamination pathway. The CAD comparison workflow in Voyager makes that verification direct. Import a CAD file, extract a mesh from the scanned part, align them automatically, and a deviation heatmap shows where the physical part diverges from design intent. The process is fast enough to be routine and precise enough to catch what other methods miss. Connecting the connectors These six connectors represent different mechanisms and different risk profiles, but they share a common problem: the features that determine whether they work correctly are internal, assembled, and invisible to surface inspection. Thread integrity, valve seating, seal compression, channel geometry, material voids—none of these can be confirmed by looking at the outside of the part. CT makes them visible without destroying the part, which means engineers can verify assembly, catch defects, and understand failure modes on the same unit that would otherwise go into service.

Design to Reality

What QMSR Means for Medical Device Product Lifecycle Management

The FDA's February 2026 transition to the Quality Management System Regulation, or QMSR, has captured the attention of the entire medical device industry, and rightly so. For many teams, particularly those already operating under ISO 13485, the practical implications are less dramatic than the headlines suggest. For others, especially companies that have only ever sold into the US market, the shift requires real preparation. Understanding what changes and where the gaps tend to appear is the most useful place to start. What QMSR and ISO 13485 change in practice For companies already certified to ISO 13485, the move to QMSR is not a major disruption. If you are selling globally or into Europe, you have been integrating risk management through the product lifecycle already, and in some respects the alignment is a relief: one standard to track rather than two that occasionally pull in different directions. The companies with more work ahead are those that have operated exclusively under the FDA's previous Quality System Regulation without ISO certification. For them, the most important immediate step is a thorough gap analysis: reviewing their current quality management system and standard operating procedures against the new requirements, and identifying where their documentation and processes fall short. The area that tends to surface the most gaps is risk management, specifically the expectation that risk thinking is woven through the entire product lifecycle, not treated as a discrete activity at specific checkpoints. Risk-based thinking across the product lifecycle Risk, in the regulatory sense, is a broader concept than most people assume when they first encounter it. It is not just about whether a device might break. It encompasses everything that can affect product quality from the moment manufacturing is complete through the full lifecycle of the device in the field: transit, sterilization, storage, handling by clinical staff, interaction with the patient. Any variable that is not understood and controlled contributes to overall product risk. In practice, manufacturers plan and manage this through a structured process. There is typically an overarching risk management file, with subsidiary plans and files covering areas such as design, production, and usability. Each of those feeds into a cumulative risk profile. The goal is to understand the variables and outside forces that may impact patient safety and product quality, and the process ensures that no high risks go unmitigated, and that the aggregate of lower-level risks does not reach a point where the product should not be released. As risk cannot be eliminated entirely, this understanding and mitigation is critical to successful and safe product launches. Risk does not start or end at the manufacturing floor. Under QMSR, teams are expected to trace and mitigate it across every stage shown here, from the first design decision to the last field return. The process follows a straightforward logic: what is the initial risk, what is the response to it, and what is the residual risk after that response? Regulators are not looking for perfection. They are looking for rigor: evidence that teams have thought carefully about risk at every stage, documented their reasoning, and put controls in place that are proportionate to what they found. The audit gaps most teams miss For companies preparing for their first audit under QMSR, the gap that appears most often is a disconnect between how risk management is understood internally and what regulators expect to see demonstrated, and it rarely shows up as a missing document. The best teams treat risk management as more than just a compliance activity, more than checking a box as part of a design history file. The regulatory expectation aligns with that thinking: risk management should be embedded in the way the business operates rather than being bolted onto it. That means having documented risk assessments at the appropriate stages of the product lifecycle, ensuring that the assumptions driving those assessments are interrogated rather than carried forward uncritically, and maintaining the traceability to show how risk was tracked and addressed as the product evolved. ISO 14971, the standard for QMS risk management, is a useful reference point here. Checking your risk management file against the standard, and ensuring your documentation reflects the right assessments at the right lifecycle stages, is a concrete and practical way to close the most common gaps before an audit. The evidence that regulators increasingly expect Risk analysis is only as credible as the evidence supporting it. Two categories of evidence have become increasingly central to how quality teams build that credibility. The first is failure mode characterization. Testing products to failure, understanding how and why they fail, and documenting those failure modes rigorously is foundational. This is not new, but the expectation of depth and traceability has increased. Knowing your failure modes well also creates a significant advantage later: if a device is returned from the field, you are starting from a position of knowledge rather than having to reconstruct the root cause from scratch. The second is real-time statistical process control (SPC), brought as close to the source of manufacturing as possible. No manufacturing process produces identical results every time. There are too many variables. But robust insights and trending data on your inspection equipment, maintained continuously rather than sampled at checkpoints, give you the ability to see problems forming before they result in nonconforming product. That is the direction quality teams should be moving: from reactive inspection to predictive process monitoring. More data, or better data, generated with the same level of investment, translates directly into greater confidence in your risk analysis. That confidence is what regulators are ultimately looking for. Verifying that what you build matches what you designed One of the more persistent challenges in medical device manufacturing is the gap between design intent and production reality. A device that performs well in development can behave differently at scale, and the variation is often subtle enough that checkpoint-based inspection misses it. The most effective approach is identifying the features that are truly critical to quality, understanding what deviations in those features actually mean for product performance, and then building inspection and verification around them rather than around a general sampling protocol. This allows quality teams to generate richer, more targeted data, reduce the number of parts destroyed in functional testing, and maintain a clearer line of sight between manufacturing variation and product risk. CT scan of a luer lock connector showing wall thickness analysis (left) and cross-sectional dimensioning of the threaded mating interface (right). The measurements confirm whether critical geometry falls within tolerance before the connection ever reaches the field. When something goes wrong in the field Field failures, when they occur, require rapid and credible root cause analysis. The most valuable data for that process typically comes from returned devices. If a device can be retrieved and analyzed in the lab, a team that has thoroughly documented its known failure modes is in a fundamentally different position than one that has not. Rather than re-investigating failure modes from first principles, they can compare what they are seeing against a library of documented failures and understand quickly whether this is a known mode, a variation of one, or something new. This is also where machine learning is beginning to have a concrete and meaningful impact. Training a model on documented failure modes from the design process allows post-market surveillance teams to work with better data, flag anomalies faster, and identify edge cases that may not have been anticipated during development. That kind of capability, connecting design-phase knowledge to field performance data, is one of the more practical and immediate applications of AI in medical device quality today. Building a quality system that holds up under scrutiny For earlier-stage companies and startups, the QMSR requirements can feel like a significant compliance burden. The more useful framing is to think of them as a design problem: how do you build risk-based thinking into the way your team works rather than layering it on top of existing processes as a documentation exercise? The most important principle is thoughtful interpretation of the regulation for your specific context. There is no single correct implementation. What matters is that your approach is coherent, that it genuinely reflects how you manage risk, and that it produces the traceability to demonstrate that. Build it into your business processes, not just your QMS. Establish checkpoints throughout the product development lifecycle where the assumptions driving your risk analysis are revisited and stress-tested. And keep the end goal in view: a product that surgeons and clinicians can rely on, with a post-market record that reflects the rigor of what went into building it. The next five to ten years The next five to ten years in medical device development will be shaped significantly by the integration of new tools into quality and risk management workflows. Machine learning, AI-assisted analysis, and digital twins are beginning to make it possible to understand products faster and with more precision than was previously achievable. The value of those tools, in a regulatory context, is that they can accelerate the process of understanding what level of risk control is needed to produce a device that is both safe and reliably manufacturable. The underlying expectation from regulators is not changing: demonstrate that you understand your product, that you understand the risks, and that your controls are proportionate and traceable. What is changing is the sophistication of the tools available to meet that expectation. The companies that learn to use them well will be better positioned not just for audits, but for the harder and more important goal of putting better devices into clinical use.

Materials World

What Roasting Does to a Coffee Bean, Seen From the Inside

Coffee roasting is a manufacturing process. Beans enter as raw agricultural material with a precise chemical composition and exit as something engineered. They achieve a specific density, a specific porosity, and (most importantly) a specific flavor profile built from controlled heat applied over a specific stretch of time. What happens inside the bean during that transformation is not directly visible to the roaster, and for most of the industry's history, it has been understood largely by inference, by sound, by smell, and by color. CT imaging changes that. We scanned four coffee beans at successive roast stages, from unroasted green through French roast, using a Neptune configured for the resolution required to resolve fine internal features. What we found was the material logic underpinning the visual record of a familiar process. The green bean In its unroasted state, a coffee bean is a seed: dense, high in moisture, and structurally intact. The outer silver skin has been removed during processing, but a fine membrane of chaff remains nested in the central cleft. Voyager's measurement tools put it at 15 microns thick, about one-sixth the diameter of a human hair. The embryo, which could have developed into a new plant, appears as a darker region within the endosperm on the scan. On a density map where white is highest density and black is lowest, the green bean reads light throughout, consistent with high moisture content. The complex sugars and acids concentrated here are the precursors to everything that will happen next. Light roast: first crack Roasting begins to alter the bean's structure at around 385°F (196°C). The water inside the bean vaporizes, pressure builds, and the structure releases in what roasters call first crack, an audible pop that signals the start of qualitative transformation. Our light roast scan shows the first formation of cracks in the endosperm and the beginning of porosity as the cellular matrix opens up. The bean has lost at least 10% of its weight. The chaff, now 20 microns thick, has expanded slightly. The all-important Maillard reaction is now underway, building the first flavor compounds. This is the earliest point at which the bean can be ground and brewed. Medium roast: the Full City At approximately 435°F (224°C), the bean reaches what roasters call the Full City roast. The CT scan shows a substantially more porous internal structure as the cellular matrix continues to break down. Surface oils are beginning to develop. The chaff has thickened to 25 microns. On the density map, the bean reads darker, reflecting the lower density of an increasingly expanded structure. The flavor chemistry is more developed here: caramelization has progressed, and the balance between acidity and sweetness is at its peak. Dark roast: second crack and beyond Second crack begins around 445°F (230°C), and French roast pushes to approximately 465°F (241°C). By this stage, the bean has shed roughly 25% of its original weight. The scan shows a structure that is highly expanded, extremely porous, and brittle throughout. Oils have migrated to the surface, which is why dark roast beans have a visible sheen. The chaff has puffed dramatically, ranging from 40 microns to 0.3 mm in thickness, and has roasted to the same darkness as the surrounding bean, making it invisible in ground coffee. Acidity has largely been eliminated. What remains is intensity, bitterness, and the particular smokiness that comes from pushing a material close to its limits. The roaster The beans for this study came from our friends at Four Barrel Coffee, in the Mission District of San Francisco. The roast-master there, Ryan Woodrow, works on a 1957 Probat UG15 drum roaster, a cast-iron machine that applies convection, conduction, and radiant heat simultaneously. Ryan showed us how he monitors temperature probes throughout each roast, adjusting gas flow to hit specific endpoints defined by time, temperature, and color, with aroma checks at every stage. After roasting, the beans move immediately to a cooling tray to stop the process in its tracks. The Probat's age and construction enable fine-grained manual control, the kind that lets an experienced roaster express the specific character of a specific bean from a specific origin. Craft roasters like Four Barrel rarely push beyond medium or second crack for this reason. Past that point, the bitterness of the roast begins to overwhelm the origin complexity that justifies sourcing carefully in the first place. If you keep going, carbonization and then fire are the only destinations left. What the scans add Roast level has always been assessed from the outside: bean color, surface texture, the sound of the cracks, the smell of the exhaust. Those sensory cues are real, and experienced roasters read them well. But they are surface signals. CT scans make the internal state legible: the specific progression of porosity, the structural changes in the cellular matrix, the behavior of the chaff at each stage. For a roaster trying to understand why a particular profile produces a particular cup, or why two batches that looked identical off the drum tasted different, that internal view is a different class of evidence. Triton or Neptune can assess roast uniformity across individual beans and across batches, measure chaff integrity, and detect structural inconsistencies that surface inspection cannot catch. For a craft operation, that is a tool for refining their process. For a larger operation, it’s one way to ensure quality stays consistent at scale.

The Quality Gap

What’s Hiding Inside Haribo’s Power Bank and Headphones?

Before it disappeared from Amazon, the Haribo 20,000 mAh power bank spread quickly through backpacking communities. It offered high capacity at a low weight, and its specs made it attractive to ultralight hikers who count every gram. And it was cute to boot; Haribo branding adorns the front and a gummy bear is attached to the charging cable. For a few months it looked like a great deal with broad appeal. Then the listing was pulled, with Amazon telling customers there was a “potential safety or quality issue.” DC Hong Kong Global, the manufacturer of Haribo-branded electronics under license from the gummy bear company, didn't publish a statement about the issue. We CT scanned the power bank along with a couple of the brand’s earbuds. Every product we scanned shows battery defects that increase the likelihood of failure and possibly fires. Why backpackers loved itA power bank sits at the intersection of energy storage, reliability, and field conditions. Backpackers rely on GPS, weather forecasts, and emergency communication. A battery failure can end a trip or create a dangerous situation. When a lightweight pack promises substantial capacity and has a great price, it's too much to resist.The Haribo power bank weighs roughly 286g and has a capacity of 20,000 mAh. That ratio drew attention because it implied efficient cell packaging. Our scans show that the structure inside the pouch has bigger problems than sheer capacity.Overview of the Haribo 20,000 mAh power bankWhat we found inside the power bankThe power bank contains two pouch cells with poor edge alignment. The electrodes don’t form a uniform block. Instead, the layers bend and rise in a wavy pattern. This pattern reflects uneven winding during assembly. It also indicates that the electrodes were not stabilized before sealing. Poor edge alignment can lead to uneven lithium plating, which in turn can cause dendrite formation. This increases the risk of shortened battery life, failure, and, depending on where the dendrites form, the worst-case scenario of thermal runaway. The misaligned layers also indicate poor process controls generally, and one has to question the potential risks embedded in the other steps in the battery manufacturing process. Uneven edge alignment in the Haribo 20,000mAh power bank Deeper structural faults in earbudsThe bluetooth earbuds use smaller cells, but the defects inside them are more severe.The noise-canceling earbud has torn cathode layers near the top of the stack. The tears leave exposed, irregular edges that can shift or shed material during cycling. When present, they signal significant problems in electrode cutting or handling, a clear indicator of a lack of quality control. The two Haribo bluetooth earbudsThe same earbud shows inconsistent anode overhang. In some slices, the anode layer stretches far beyond the cathode. In others, it nearly disappears. An even overhang provides room for the anode to expand during charging. When it varies this widely, the cell cannot maintain balance under normal use.The lower-cost, basic set of earbuds also suffer from poor anode overhang. We measured areas where the anode overhang was 0.072 mm, far below the industry standard of about 0.5 mm. Sufficient anode overhang is seen as a safety-critical feature to prevent irregular lithium plating and dendrite formation. The lack of overhang we observe in the earbuds meaningfully increases the likelihood that these batteries could fail prematurely or even catch fire. The charging cases have the same issues. Their cells show poor alignment and insufficient overhang. These problems are systematic rather than isolated, indicating that all the cells across the Haribo-branded product line may come from the same low-quality supplierOn left: battery from the noise canceling earbud. On the right: battery from the low-cost earbud.Why these defects matterPouch cells live inside compact enclosures with little space for deformation. A poorly-wound cell with uneven layers is at an increased risk of both performance and safety failures. Earbuds and headphones operate in even smaller envelopes. Their cells have tight enclosures and no ventilation path. A torn electrode or improper overhang removes the safety margin that prevents localized heating.In both cases, internal geometry dictates reliability. The aesthetics and advertised capacity cannot compensate for unstable stacking.What the scans tell usThe Haribo products show the same pattern: attractive specifications that hide weak manufacturing controls. They raise the likelihood of failure in environments where users expect consistency. Industrial CT makes that gap visible. Once you see the internal stack, the disappearance from Amazon makes sense.

Materials World

What’s Inside a Battery?

Most of us carry batteries around every day without thinking about what’s inside them. Phones, laptops, and cars all rely on their power. From the outside they seem simple; inside they’re made of carefully chosen materials that have to work together to provide the right amount of power. Battery basics The main pieces of a battery are the anode, the cathode, a separator, and an electrolyte. Cathode materials have shifted over the years. They used to be lithium cobalt oxide or lithium manganese oxide, though those are mostly phased out. Today you’ll see acronyms like LFP (lithium iron phosphate) and NMC (nickel, manganese, cobalt). Each mix has different strengths: some give higher energy density, others last longer, or are a bit more stable and safer. Anodes are usually graphite spread onto copper foil. The electrolyte is what lets lithium ions move between the anode and cathode. It is often a liquid solvent, which can be flammable; there are also versions that use a polymer instead. Those can improve safety and sometimes performance. Separators are thin polymers that keep the anode and cathode from touching. Solid-state batteries go further by using solid electrolytes, though they are still in development. Making a battery is not just stacking parts together. Powders like graphite or lithium compounds are mixed into a binder and coated onto thin foils. How evenly that coating spreads affects how it dries and how it compresses later. The tricky part is that every material behaves differently. You cannot simply swap one formulation for another and expect the process to work; each one needs its own set of parameters, which means testing, adjusting, and running trials until it works consistently. Competing priorities Every choice in battery design comes with trade-offs. Cost is always a factor; if a battery is too expensive, it will not make it to market. Safety is another non-negotiable. Failures can lead to recalls or worse. Beyond those, it depends on the application. Some batteries need the highest possible energy density, like those in electric cars. Others need long cycle life, like batteries for grid storage. Engineers weigh performance factors such as energy, lifespan, safety, and affordability against what the end product requires. Supply chain and recycling Batteries rely on critical minerals such as lithium, cobalt, nickel, and graphite; each is sourced from different parts of the world. That makes supply chains important. Tariffs, import sources, and political shifts all play a role in availability and cost. Recycling can help, and new processes are being developed, but it remains complicated. Each battery chemistry uses a different mix of metals, which makes separating them difficult. Recycling techniques depend on the physical and chemical properties of the powders and foils, so a process that works for one chemistry will not necessarily work for another. There is also the economic side. Recycling only makes sense if the materials are valuable enough to recover. Using less expensive materials can reduce dependence on scarce resources; it can also make recycling less worthwhile. Finding defects It only takes a small flaw inside a battery to cause an issue. Historically, anode overhang, where the anode extends too far past the cathode, has been a primary concern. Other defects include poor tab welds, wrinkles, or tears in electrodes. One of the main issues today is foreign particles. Tiny bits of metal can break off during welding or cutting, or fall in during assembly; those particles can trigger failures. [Embedded scan] To catch these problems, you need ways to look inside the sealed battery. CT scanning is one option, and it can detect very small particles, even down to a few microns. Ultrasound can also work, though with lower resolution. Destructive testing is another method, but it is slow and not practical for every battery that comes off the line. Balance From the outside, a battery looks simple. Inside, it is a complex mix of materials, processes, and trade-offs. Every decision, whether it is choosing LFP over NMC or tweaking how powders are coated, has ripple effects on performance, cost, safety, and recyclability. There is no perfect recipe; it is always about finding the right balance for the job at hand.

Scan of the Month

What's Inside a Contactless Credit Card

A contactless credit card, opened up A standard credit card is about 0.76 millimeters thick. Inside that space, bonded between rigid outer layers of PVC under heat and pressure, is a complete wireless communication system: a copper antenna coil, a dual-interface EMV chip, and the connections between them. None of it is visible from the outside. The scan shows the whole thing. The antenna The most immediately striking feature in the wide scan is the antenna loop. It runs along three sides of the card's perimeter before routing inward to connect with the chip module at the right side of the card. A second, smaller coil structure is visible in the upper left quadrant. The antenna operates at 13.56 MHz, the standard frequency for NFC payment transactions, and its geometry matters: the number of turns, the wire gauge, and the total coil area determine the inductance of the circuit and how reliably it couples with a reader. Because the antenna spans most of the card's area, the effective pickup zone is the entire card, not just the region near the chip. The chip The close-up scan of the chip module shows the inductive coil in detail: multiple concentric rectangular turns wound tightly around the chip package. The chip itself sits at the center, its internal components visible at varying densities, with bonded contact points and interconnects distinguishable within the package. This is a dual-interface EMV chip, meaning it handles both contactless transactions through the antenna and contact transactions through the gold pads on the card's face. The two interfaces share a single chip. How the transaction works When the card enters the field of a reader, the antenna picks up the electromagnetic signal and transfers power inductively to the chip. No battery, no direct contact. The chip generates a one-time cryptographic code, packages it with the transaction data, and transmits it back through the antenna. The reader passes that data to the card network, which runs fraud screening and routes the authorization request to the card issuer. The whole sequence takes a fraction of a second. The manufacturing challenge Getting this assembly into a card that survives years of bending, flexing, and contact with the bottom of a bag requires careful lamination. The antenna and chip inlay are bonded onto a substrate, then sandwiched between layers of PVC with different stiffness: a rigid outer layer and a softer inner layer that flows slightly under heat and pressure to encapsulate the assembly without distorting the antenna geometry. Distortion changes the electrical properties of the coil and can cause the card to fail. The antenna's corners and bridging points are the most vulnerable, which is why card manufacturers treat lamination control as a quality problem, not just a production step.

Design to Reality

What's Inside the World's Fastest Marathon Shoes

In 2017, Nike released the Vaporfly, the first commercial running shoe with a carbon-fiber plate embedded in the midsole. Before it, only 19 women had ever run a marathon in under two hours and twenty minutes. In 2023 alone, 26 did. The performance shift was significant enough that World Athletics introduced regulations in 2020 limiting sole thickness and requiring competition footwear to be commercially available before use in elite races. Every major brand has since developed its own version of the super shoe. To understand what separates the leading models, Lumafield worked with the Financial Times visual storytelling team to CT scan two of them: the Nike Alphafly 3 and the Adidas Adizero Adios Pro Evo 1. The FT used the scans as the evidential foundation for a reported investigation into the technology driving record-breaking marathon performance. What follows is a detailed account of what the scans revealed. [embed: Nike Alphafly 3 scan] [embed: Adidas Adizero Adios Pro Evo 1 scan] Why CT scanning A running shoe is a layered object. The components that matter most for performance, the carbon plate, the foam geometry, the air chambers, sit inside a midsole that is sealed and opaque. You can cut a shoe in half, which reviewers do regularly, but you can only cut it one way, you destroy the shoe in the process, and you see a single cross-section. CT scanning rotates the X-ray source around the object, reconstructing a full three-dimensional density map from hundreds of two-dimensional images. The result is a model you can slice in any plane, isolate by material density, and examine without touching the object. For a story about what is actually inside these shoes, and whether the manufactured product matches the design intent, that is the only way to see the whole picture. The problem both shoes are solving Every foot strike in a long-distance race is an energy transaction. Force travels down through the shoe, and some fraction of that energy returns to the runner as forward momentum rather than dissipating as heat. Over 26.2 miles, the cumulative difference between a shoe that returns measurably more energy and one that doesn't is visible in the finishing time. The engineering challenge is that the three variables runners care about, energy return, cushioning, and weight, pull against each other. More foam means more cushioning and potentially more energy return, but also more weight. A stiffer plate improves propulsion but changes how the shoe interacts with different foot-strike patterns. Every design decision involves a tradeoff, and the two shoes have resolved those tradeoffs in opposite directions. The Nike Alphafly 3: a system of components Nike's answer is accumulation. The Alphafly 3 is built around three technologies working together: a full-length carbon-fiber Flyplate, two Air Zoom pods embedded in the forefoot, and a continuous ZoomX foam midsole connecting heel to forefoot. The stack height is 40mm at the heel and 32mm at the forefoot. The shoe weighs 218 grams. [image: Alphafly 3 CT cross-section, full midsole] The scans show how those components actually relate spatially inside the assembled shoe, something neither marketing materials nor cutaway illustrations can provide. The carbon plate runs the full length of the shoe at a consistent depth within the foam, curved rather than flat. That curvature matters: a flat plate would shift the point of force application toward the toe, increasing calf strain. The curved plate keeps it under the ball of the foot and directs propulsion forward. The two Air Zoom pods sit in the forefoot above the plate, enclosed in foam on all sides. Inside the pods, the scans resolve fine internal fibers that divide the air volume into smaller cells. Those fibers limit radial expansion under load, which increases the pressure the pod sustains and improves the energy it returns per compression cycle. This detail is invisible from the outside and does not appear in any published cutaway of the shoe. The continuous midsole, a change from the Alphafly 2 which had a gap between the heel and forefoot sections, shows in the scan as a single uninterrupted foam body from heel to toe, with the plate and pods embedded within it. The geometry confirms what Nike describes as a smoother heel-to-toe transition. [image: Alphafly 3 forefoot detail, Air Zoom pod internal fiber structure] The Adidas Pro Evo 1: reduction as engineering Adidas took the opposite approach. Where Nike added components, Adidas removed them. The Pro Evo 1 weighs 140 grams, 78 grams less than the Alphafly 3, and much of that reduction comes from eliminating material rather than substituting lighter versions of it. Instead of a full-length carbon plate, Adidas embedded eleven carbon-composite rods, each 5.5mm thick, running longitudinally through the midsole and mapped to the metatarsals. The intent is to target propulsion under specific joints, including the big toe, rather than distributing it across the full foot. Two pieces of Lightstrike Pro foam are bonded around the rods. The outsole is not a separate rubber component but a liquid rubber coating applied directly to the foam surface, removing approximately 20 grams compared to a conventional outsole. The upper is a single-layer transparent mesh with no sock liner. [image: Pro Evo 1 CT cross-section, full midsole showing rod array] The scans confirmed the rod placement and bonded foam geometry. They also found something Adidas had not publicized. What the scans found in the Adidas rods Several of the carbon-composite rods contain internal air pockets, visible in the density mapping as voids within what should be solid carbon composite. [image: Pro Evo 1 carbon rod detail, CT cross-section showing voids] The FT put this finding to Adidas directly. The company described the voids as minor irregularities that can occur during the manufacturing process without affecting the performance or durability of the rods, and said it was not aware of any rods breaking in competition. That response is plausible. Carbon composite fabrication at complex geometries routinely produces small internal voids, and the rods have performed at the highest level of marathon competition without reported failures. But the finding points to something worth understanding about how performance footwear gets made. The Pro Evo 1 is largely assembled by hand, a consequence of the precision required to position the rods correctly. Adidas has acknowledged the shoe does not generate direct profit at its $500 retail price. It functions as an engineering platform: what gets learned about rod placement, foam bonding, and wear behavior at the elite level feeds into higher-volume models. A scan that surfaces manufacturing variation in that platform is exactly the kind of evidence that drives the next iteration. CT produces quantitative density data rather than surface images, which means it can show the difference between a solid rod and one with a void, between a foam layer bonded as designed and one with a gap, between a component that matches specification and one that doesn't. That is what it contributed to this investigation, and what it contributes to the manufacturing process more broadly. What the engineering difference reveals The two shoes represent different theories about where the performance gains in marathon footwear come from. Nike's answer is a tuned system of interacting components, each refined over multiple generations, designed to work across a range of foot-strike patterns and runner profiles. Adidas's answer is that weight is the primary variable and that the structural function of a plate can be distributed through discrete rods positioned anatomically, eliminating material rather than optimizing it. Both approaches have produced marathon world records. What the scans add is a precise account of what is actually inside each shoe, what the manufacturing process delivered relative to the design intent, and where the gap between those two things is visible. The distance between design intent and manufactured reality is where quality is won or lost, and where visibility matters most.

The Quality Gap

What Went Wrong Inside These Recalled Power Banks?

Lithium-ion batteries power the many electronic devices that we rely on every day, from EVs to smartphones and laptops. They’re so prevalent, that the average American owns nine lithium battery-powered devices.1 But they carry real risks when quality lapses occur. Overheating and fire hazards can cause property damage, injuries, or worse. With millions of these batteries in circulation, even a single defect can have a ripple effect of consequences for consumer safety and brand reputation. Recently, Anker recalled over one million PowerCore 10000 power banks, model A1263, produced between 2016 and 2019 and sold through 2022.2,3 Anker has provided a general warning that the lithium-ion battery can overheat, but they have yet to share the exact reason for the recall. Armed with our Neptune Industrial CT Scanner and five A1263 power banks from Lumafield team members, we set out to see if we could identify the source of this recall. Could we identify the defects with CT scanning? And could CT inspection during development or manufacturing have prevented the faulty power banks from shipping in the first place?Scanning SetupWe procured five potentially-affected power banks from various Lumafield colleagues, which we labelled PB1, PB2, PB3, PB4, and PB5. We then ran the serial numbers against Anker’s recall form, and determined that three of our power banks were impacted, while two were not.The power banks were scanned using Lumafield’s Microfocus Neptune. This configuration’s small X-ray spot size makes it the ideal tool for resolving fine details in electronics and batteries.4Use the embedded Voyager window below to explore the CT scan of the recalled power bank. Rotate and examine it from any angle.First Look: Battery CellsThe first element we compared across the power banks were the battery cells themselves. A lithium-ion battery consists of two separated electrodes. The anode, usually graphite, stores lithium ions during charging. The cathode, typically a layered metal oxide, releases and accepts lithium ions as they move through the electrolyte during charge and discharge, enabling reversible energy storage.There are a few common defects in the battery manufacturing process that can be easily spotted in a CT scan. For example, in lithium-ion batteries, it’s important to ensure that the anode has a sufficient overhang above the cathode, preventing the lithium plating that can lead to dendrite formation.5 Dendrites can subsequently result in degraded performance and short circuits, which can cause the worst-case scenarios of thermal runaway. CT scanning can also be used for Foreign Object Detection (FOD) within batteries, as particle contamination can lead to reduced performance and potentially short circuits.The A1263 power banks each contain three 18650 lithium-ion battery cells inside. It was quickly apparent that batteries from at least two different suppliers had been used within the affected power banks. The cells in PB1 (recalled), PB2 (recalled), PB4 (not recalled), and PB5 (not recalled) seemed similar, but the cells in PB3 (recalled) had a few key differences. First, the batteries in PB3 have a mandrel to help prevent core collapse, a type of deformation in lithium-ion batteries with “jelly roll” construction where the innermost layers can sag into the center.6 Mandrels can also serve as a pathway for gas escape in case of overpressurization. Some suppliers add a center tube to strengthen the core and prevent this from happening. Though none of the batteries in the five power banks we scanned seem to show core collapse, only PB3 has reinforcement material to combat this potential issue.On the left cell, from PB3, reinforcement foil appears as bright, dense material around the battery’s center. This is noticeably absent from the right cell, from PB1. The other discrepancy in the PB3 batteries is the number and style of vent openings. PB1, PB2, PB4, and PB5 have four smaller openings at the positive terminal, while the PB3 cells have three larger ones. This further suggests that the batteries in PB3 come from a different source.In the left image, we observe 3 vent openings from a cell in PB3. In the right image we see the four vent openings common across the cells in PB1, PB2, PB4, and PB5. If the recall is affecting units made with 18650 battery cells from multiple suppliers, that suggests the root cause of the recall stems from elsewhere in the power bank. We next focused on the PCB and assembly of the board with the cells.Assembly Discrepancies Looking at the assembly of the A1263 power banks, one difference between the recalled and non-recalled devices stood out immediately. All five power banks used tab bus bars to connect the positive and negative terminals of the 18650 cells. However, the non-recalled PB4 and PB5 devices used conventional insulated wires to make the positive and negative connections to the PCB, while the recalled PB1, PB2, and PB3 power banks used flat tab wire for the entirety of the connection.On the left image of PB4, thin insulated wires are clearly visible. On the right image of PB2, wide thin tabs are used for the connections instead.‍A closer look at the connections between the cells and PCB in PB4 (left) and PB2 (right).Looking more closely at the connections of PB1, PB2, and PB3, we see some process variation that could be a potential cause for concern. The shape of the negative tab differs dramatically, and in PB1 it appears slightly twisted. From left to right, the negative tab crossing above the positive tab - PB1, PB2, PB3.We can measure the distance to quantify how dramatically the gap between the positive and negative bus bars varies across the three units. In PB1, that distance is only 0.52 mm, compared to 1.12 mm in PB2 and 1.58 mm in PB3. The short distance and visible deformation in PB3 suggests the possibility of the negative bus bar shorting out against the positive bus in some scenarios.From left to right, measurements of the space between the negative and positive tabs for PB1, PB2, and PB3.Evolving Power Bank DesignIn addition to the recalled power bank, we scanned Anker’s current PowerCore 10000 model, the Anker 313. This model was first released in January 2023. The exterior of the power bank has changed significantly. At 5.87 in by 2.68 in by 0.55 in,7 compared to the A1263’s dimensions of 3.62 in by 2.36 in by 0.87 in,8 the 10K product has become longer and wider, but also noticeably thinner. Looking inside, we find this slimming has been accomplished by replacing the three 18650 cylindrical cells with a single lithium-ion pouch cell. With 90-95% packaging efficiency, pouch cells are maximally compact relative to other battery cell designs and can be customized to accommodate different industrial design requirements.9 The single-cell configuration of the Anker 313 simplifies the assembly required. We also see that significant changes have been made to the embedded electronics, as is to be expected given both the new product architecture and the general advances in PCB manufacturing since the A1263 was manufactured.CT scan of the updated Anker PowerCore 10000, showing the single lithium-ion pouch cell and a slice of its PCBA.ConclusionCT scanning the A1263 Anker PowerCore 10000 power banks has provided a small window into the supply chain complexity of these pervasive, ostensibly simple products. This SKU was manufactured over a period of 3 years and 10 months, and sold for over six and a half years. From our limited analysis of just five power banks, a small subset of the millions of devices sold, we have identified that throughout the A1263 production period at least two battery cell versions and two battery connector designs were used. A larger sample size would likely reveal further variations. Managing a high-volume battery supply chain is inherently challenging, and ensuring quality throughout each stage is essential. Only Anker can know for sure why these popular consumer devices have warranted a recall three years after they were last sold. However, though we don't know exactly what flaw triggered this massive recall, it illustrates just how costly quality issues can be. The volumes involved in these types of consumer products can be staggering, and it can be extraordinarily difficult to track quality issues all the way upstream. Since we began this exploration, Anker has issued a new recall for five additional power bank models: Anker Power Bank models A1257 and A1647, Anker MagGo Power Bank model A1652, and Anker Zolo Power Bank models A1681 and A1689.10 Unlike the A1263 Anker PowerCore 10000 power banks, these are more recent products. Anker has shared that the recall was caused by potential issues with one of their battery suppliers. Their English announcement credits the more rigorous quality processes they implemented earlier this year with catching the cause for concern,11 while announcements in China by Anker and competitor ROMOSS Technology disclosed that the cell supplier in question made an undisclosed change in the raw materials used, which could lead to insulation degradation.12 Anker has recalled 710,00 units from those models in China, and the global numbers have yet to be disclosed. Looking just at the recall of A1263 Anker PowerCore 10000 power banks sold between 2016 and 2022, about 1,158,000 units have been affected. Impacted users will receive either a replacement unit or a $30 Anker Store gift card, implying a financial exposure of over $34 million, not including other relevant recall-related costs. Beyond the immediate economic impact, battery recalls can cause permanent reputational damage that proves impossible to quantify in the long term. Anker has underscored their commitment to better quality assurance processes and signed an agreement with a new battery vendor13 as they work to regain the trust of their customers.Industrial CT inspection offers a powerful tool that can help ensure the safety of the many battery products that surround us. It can be leveraged during the design and development stages to validate designs and brought into the manufacturing ramp-up to support First Article Inspection and validate assembly processes. In production, it can non-destructively verify quality, both upstream and downstream. And after parts have shipped, CT scanning can be used to support faster failure analysis as issues arise in the field. Ultimately, ensuring the reliability of lithium-ion battery products is becoming more critical than ever, as these devices become increasingly embedded in our daily lives. As CT inspection becomes more accessible and easier to adopt, it can serve as a safeguard of both consumer safety and manufacturers’ bottom lines.

The Quality Gap

What You Don’t Measure Will Cost You

When Lumafield commissioned a survey on the Cost of Quality earlier this year, I went in with a rough hypothesis: manufacturers know quality costs money, they probably undercount it, and they're working to close the gap. A year of writing Recall Radar will do that to you. The evidence of quality failures reaching consumers is consistent enough that you start to assume the organizations upstream are at least aware of the problem. The survey results confirmed some of what I expected, but the scale of what's missing surprised me. We surveyed 210 quality decision-makers across the US and Canada through independent research firm NewtonX, spanning aerospace and defense, automotive, consumer electronics, CPG, and medical devices. While the methodology was straightforward, the results were not. Let’s start with a figure to contextualize what follows: 42% of respondents estimate that quality-related costs exceed 5% of their revenue. Another 40% place that figure between 2% and 5%. These are substantial numbers on their own, and the academic literature has long suggested the true cost lands even higher, somewhere between 5% and 15% of revenue for most manufacturers. But the finding that gave me pause was this: 58% of respondents believe at least a quarter of their true quality costs go unaccounted for. They're spending a lot, and they think they're still missing most of the picture. The Cost of Quality framework has been around since the 1950s, when Joseph Juran and Armand Feigenbaum formalized it as a way to give executives a dollar-denominated view of what quality is actually worth. It includes prevention, appraisal, and failure costs. The idea is hardly new, and yet only 35% of the organizations we surveyed said they use a comprehensive COQ model. The majority track some components, mostly failures like scrap and warranty, but prevention and appraisal costs are substantially underanalyzed. Just 24% of organizations formally consider prevention costs at all. This matters because prevention and appraisal are where the leverage is. Costs incurred after a defect has left the factory are categorically more expensive than costs incurred to prevent it from forming in the first place. Every recall I cover in the Recall Radar column is a failure cost. Most of them trace to a process or design issue that was knowable earlier. The survey data suggests that most organizations aren't actively measuring the value of upstream interventions, which makes it nearly impossible to make the ROI case for investing in tools for closing the gap. The barriers are real and well-understood by the people living with them. Two-thirds of respondents cite data scattered across teams and systems as the primary obstacle. Nearly as many point to difficulty quantifying certain quality costs. These aren't organizational failures so much as structural ones. Quality-related data flows through almost every function, across quality, operations, manufacturing, supply chain, engineering, customer support, and finance, but the responsibility for consolidating it typically falls to the quality team alone. What doesn't get consolidated doesn't get counted. I found the tariff data unexpectedly telling. Sixty-two percent of respondents say recent trade changes have made quality management harder. The mechanism matters here: 54% report that suppliers have downgraded materials to offset tariff costs, and 27% say suppliers have misrepresented country of origin. These aren't abstract supply chain risks. They are substitutions and misrepresentations that introduce variability into products, often without the buyer's knowledge. In a measurement environment already characterized by significant blind spots, that's a compounding problem. The supplier finding ran through every industry we surveyed. Forty-one percent of respondents identified supplier-related quality issues as the single category where a 25% reduction would have the greatest operational impact. That was the top answer across aerospace, automotive, CPG, consumer electronics, and medical devices, without exception. The VP of Procurement from one of the CPG companies we interviewed put it plainly: "I always say if we buy cheap, we buy expensive. Because nobody wants to deal with a recall or with quality incidents." If there's a finding in this report that I think deserves more attention than it's likely to get, it's the inspection labor picture. Seventy-seven percent of organizations still rely on manual visual inspection as part of their process. In aerospace and defense, 30% of respondents spend more than 40 hours per month on inspection and verification activities alone, nearly double the study average. That's an enormous resource commitment concentrated in a method that is inherently variable, fatigue-dependent, and difficult to scale. AI adoption is growing, 80% of respondents claim to be using AI-based tools in some form, but the in-depth interviews painted a more nuanced picture. Most of that activity is early-stage. The barrier to deeper utilization is not in the technology so much as in the organizational change required to make it work. One quote from a consumer electronics engineering program manager stayed with me: "The cost to implement the technology is not that big. The major expense is training and changing mindsets and behavior." That struck me as honest in a way that most technology adoption narratives are not. The bottleneck isn't access to tools, but rather the human and organizational infrastructure that determines whether or not manufacturers end up using them. The broader conclusion I take from this data is that the COQ measurement gap is not primarily a technical problem. The data exists in most organizations. What's missing is the architecture to pull it together, the organizational will to count things that don't make anyone look good in the short term, and the time to do the work when teams are already managing active quality issues. The Director of Quality from one of the aerospace companies we interviewed offered the clearest framing I found in the whole dataset: "There are so many things that quality can do up front that are value-added. They're the cost of quality. Versus after you're in production and starting to make mistakes, then that's the cost of poor quality. Focus on the cost of quality, because it pays off in the end. The cost of poor quality is just taking away from your margin." What you don't measure, you can't manage. And while it might not kill your business, it will cost you. The only question is how much. Our survey suggests that most manufacturers already know this. The gap is between knowing it and doing something about it. The full Lumafield Cost of Quality Report is available here. You can also take the survey for yourself to see how your organization stacks up against industry benchmarks.

From The Floor

Your Toner Cartridge Is Lying to You

You've been there. You walk to the office printer to pick up a document and find the low toner warning blinking at you. You shake the cartridge, slide it back in, and get another fifty pages out of it. It works every time. But why? We scanned a "empty" toner cartridge alongside a brand new one to find out. What the scans showed was not what we expected, and it raises a real question about how these products are designed. What the scan found A brand new toner cartridge, fresh from the box, has toner occupying about 20% of the hopper volume. The cartridge the printer had flagged as empty still had roughly 15% of the hopper volume filled. The entire usable operating range of this design is a 5% change in volumetric fill. The other 80% of the cartridge is air. That gap is not an accident or a manufacturing shortcoming. It is a deliberate design decision, and the reasons behind it are more interesting than they first appear. Why the hopper is mostly empty Toner is a finely granulated thermoplastic powder, typically polyester or styrene-acrylate resin with carbon black for color and silica coated onto each particle to control flow and prevent clumping. Individual particles run between 5 and 12 micrometers in diameter, about one-tenth the width of a human hair. At that scale, powder behavior is finicky. Pack it too densely and the internal augers and paddles that move toner toward the developer roller can seize, or the powder bridges across the hopper and stops flowing entirely. The generous headspace keeps the toner fluidized. The motion of the cartridge during normal use, combined with internal agitators, keeps the powder evenly distributed and consistently charged so that print density stays uniform from the first page to the last. The extra volume also lets manufacturers use the same outer shell for standard and high-yield versions of the same cartridge, with different fill weights on the same automated line. How the sensor works, and why shaking fools it Toner level sensing varies by manufacturer and model, but the most common approaches are optical and page-count based. Optical sensors use a light beam and a flag or window inside the hopper: when toner blocks the beam, the printer reads it as present; when the beam gets through, it reads low or empty. Page-count chips take a different approach, counting pages printed and assuming a standard coverage percentage per page, then asserting empty once a calculated consumption threshold is reached, regardless of what is actually left in the hopper. Shaking works against optical sensors because toner settles and becomes unevenly distributed over time. The region the sensor is watching can read empty while the rest of the hopper still has powder. Redistributing the toner by shaking temporarily restores coverage over the sensing region, clearing or delaying the warning. Against a page-count chip, shaking does nothing to the counter. But it still moves toner back toward the pickup area, which is why print quality often recovers briefly even when the chip-based warning persists. What this reveals The 5% operating window is the part worth sitting with. A cartridge that reports itself full is 80% air. A cartridge that reports itself empty has 15% of its volume still filled with usable powder. The sensor is not measuring what you think it is measuring, and the warning it generates is not the signal you think it is. That is not necessarily a criticism of the design. The headspace serves real engineering purposes, and consistent print quality across the life of the cartridge is a legitimate goal. But the CT scan makes visible something that the product's exterior is designed to conceal: the gap between what a device reports and what it actually contains is a design choice, and understanding that choice changes how you use the product. The shake works. Now you know why.

In this Article:

Small parts, big consequences

Helmets that don’t hold

Lighters on the loose

Heat hazards at home

Durability deficits

Highway hazards

Takeaways