Daniel sent us this prompt about baggage handling systems — those miles of conveyors and diverters behind the walls at every airport. He's fascinated by the fact that this whole infrastructure is necessarily invisible to passengers, and he's asking the right question: if barcode scanners fail at supermarket checkouts over something as trivial as a curved bottle of soda, how does a baggage routing system achieve near-perfect reliability when it's flinging duffel bags and hard-shell suitcases around at thousands per hour? He also wants to know the scale of these systems at hubs like JFK, how long they take to build, and what kind of error tolerances keep them from breaking down.
Four point five billion bags checked globally every year. And the vast majority reach the right plane, on time, without a human ever touching them after the check-in counter. That number alone should make anyone stop and think.
It's the kind of statistic where the miracle isn't that sometimes a bag goes to the wrong continent — it's that it almost never does.
And here's what makes this moment worth digging into. Post-pandemic air travel has rebounded, but baggage mishandling rates have actually dropped to historic lows. IATA — the International Air Transport Association — reported a mishandling rate of just zero point six nine percent in twenty twenty-five. That's fewer than seven bags per thousand. For context, a decade ago it hovered around one point five to two percent.
The invisible system got better while nobody was looking.
That's the thing about infrastructure that works — it becomes invisible twice over. First because it's physically hidden behind walls and under floors, and second because it's so reliable you never have reason to think about it. You only notice the baggage system when it fails. Which, statistically, it almost never does.
Which is a strange kind of engineering achievement, right? You spend years designing something, millions of dollars building it, and the highest compliment anyone can pay you is that they never knew you existed.
The silent standing ovation.
Let's pull back the curtain on this hidden world — starting with just how big these systems really are.
Denver International's system has over twenty miles of conveyor track. That's a single airport. JFK's Terminal Four alone handles twelve thousand bags an hour at peak. These aren't conveyor belts in the sense of the little rubber strip at the grocery store — these are industrial systems running through tunnels, under taxiways, between terminals.
Twenty miles of track and I still have to walk from gate B forty-seven to B forty-eight like a pilgrim.
You'd prefer they put a luggage conveyor in the passenger tunnel?
I'm just saying the bags are getting a better tour of the airport than I am.
The scale is genuinely hard to absorb. A system like Denver's takes three to five years just to design and build. And here's the part that should keep you up at night if you're an engineer — all maintenance happens in a two to three hour overnight window when flight volume drops. You can't shut down a baggage system at noon on a Tuesday.
You've got this sprawling physical beast running at Amazon fulfillment center throughput, hidden behind walls, maintained in the dead of night, and the whole thing depends on a little sticker with a barcode surviving the journey. The same barcode technology that gives up on my orange juice because the bottle has a slight curve.
That's exactly the tension Daniel's getting at. We've all stood at a self-checkout waving a bag of chips around like we're conducting an orchestra, trying to find the one magic angle where the scanner decides to cooperate. And that's in a calm, stationary environment with a human holding the item. Now picture a duffel bag tumbling onto a conveyor at twelve miles an hour, tag somewhere on the handle, possibly creased, possibly wet.
The mental image of the supermarket scanner's little red line just giving up on existence feels like it should be the whole story. But baggage systems are hitting better than ninety-nine point nine percent successful routing.
And that gap — between the supermarket experience and the airport reality — is where all the interesting engineering lives. It's not that they found one magic fix. It's layers. Multi-angle scanning arrays, RFID tags embedded in the paper label, recirculation loops for the bags that don't scan on the first pass, and humans stationed at manual resolution points as the last line of defense.
The answer to "how do they avoid scanning failures" is essentially: they assume failure will happen and build a system that absorbs it gracefully.
And that philosophy — designing for failure rather than pretending you can eliminate it — shows up in everything from the sensor layout to the mechanical diverters to the software that tracks every bag in real time. We're going to trace this from the physical scale of miles of track, through the scanning problem and all its solutions, and then into what happens when things actually break.
Let's start with the physical reality. Denver International's twenty-plus miles of conveyor track isn't just a number — it's a network that moves bags between three concourses, through underground tunnels, past explosive detection systems that are essentially CT scanners the size of small cars. A bag checked at the main terminal might travel nearly two miles before reaching a regional jet on Concourse A.
Two miles underground while I'm upstairs paying fourteen dollars for a sandwich.
The throughput is what makes the engineering punishing. JFK Terminal Four peaks at twelve thousand bags an hour. That's roughly three bags per second. Each one needs to be identified, security-screened, and routed to the correct gate — and the window between check-in and departure can be as tight as forty-five minutes.
You've got three seconds per bag, and if the scanner hiccups on one, the next two are already arriving.
That's the cascading failure risk. Which brings us to the scanning problem Daniel flagged. Supermarket barcodes fail on curved surfaces because the laser projects a single line and needs a clean, flat reflection back to the sensor. A soda bottle curves away from that line — the reflected light scatters. Baggage tags are printed flat on adhesive stock, which helps, but the bag itself is the problem. A duffel sags. A hard-shell has ridges. A ski bag is basically a nylon tube with straps everywhere. The tag could be facing down, partially obscured by a handle, creased from being slapped on at the counter.
Unlike my orange juice, nobody's standing there rotating the bag patiently until it beeps.
The solution is multi-angle scanning arrays. Picture a tunnel — the bag rides through on a conveyor, and inside that tunnel are six to twelve laser scanners mounted at different positions. Top, bottom, both sides, and several diagonal angles. Each scanner fires hundreds of pulses per second. The bag doesn't need to be oriented correctly because at least one scanner will catch a clean angle. If the tag is facing down against the belt, the upward-angled scanner underneath gets it. If it's on the side, the lateral scanners catch it.
It's not one perfect scanner. It's a firing squad where you only need one to hit.
That's the first layer of redundancy right there. The system isn't trying to eliminate scan failures — it's making sure no single failure matters. And here's where the barcode format itself matters. IATA Resolution seven forty standardizes the baggage tag globally. It's a ten-digit license plate number encoded in a one-dimensional barcode called Code one twenty-eight, which is extremely dense and readable at high speeds. That same number is also printed in plain text and embedded in an RFID chip inside the tag.
Wait — the paper tag has an RFID chip in it?
Most of them do now. The tag you get at the counter looks like a long paper sticker with adhesive backing, but sandwiched inside is a thin RFID inlay — basically a tiny antenna and a microchip. It's passive, meaning it has no battery. The reader emits a radio signal that powers the chip just long enough for it to transmit its data back. And because it's radio frequency, it doesn't need line of sight. The tag can be crumpled, folded, covered by a jacket strap — the reader still gets it.
The barcode is the backup, and the RFID is the primary.
IATA Resolution seven fifty-three, which took effect in twenty eighteen, mandated RFID tracking at key touchpoints — check-in, loading, transfer, arrival. By twenty twenty-five, about eighty-five percent of bags globally carry RFID tags. And the read rate difference is stark. Barcode-only systems achieve roughly ninety to ninety-five percent on first pass. RFID exceeds ninety-nine point five percent.
That's the difference between one in twenty bags needing a second look and one in two hundred.
When you're processing twelve thousand bags an hour, that's the difference between six hundred manual interventions and sixty. The economics of that gap are enormous — fewer staff at manual resolution stations, fewer missed connections, fewer compensation payouts for delayed luggage. The RFID infrastructure costs money upfront, but the operational savings pay for it within a few years.
Daniel's supermarket scanner problem is essentially solved by throwing more angles at it, and then made almost irrelevant by switching to a technology that doesn't need to see the tag at all.
Neither layer is perfect, which is why there's a third layer — recirculation. If a bag passes through the scanning tunnel and none of the twelve lasers get a clean read, and the RFID reader also fails — maybe the tag got torn off entirely, or it's an older tag without RFID — the system doesn't just shrug and send it to Cleveland.
The bag gets a do-over.
It enters what's called a divert loop. The conveyor splits, the bag takes a recirculation path that loops it back around for another pass through the scanning tunnel. Most problem bags get caught on the second or third pass. The angle changes slightly because the bag tumbled during the loop, a different scanner catches it, and the system moves on. The passenger never knows this happened.
Which is the whole point of invisible infrastructure. The failure happened, the system absorbed it, nobody noticed.
That recirculation loop is possible because these systems are designed with headroom. Denver's system runs at about sixty to seventy percent of peak capacity during normal operations. That slack is deliberate — it means a bag can take an extra lap without creating a traffic jam that cascades into dozens of delayed bags. The same principle applies to the mechanical side, which we'll get into, but the scanning architecture alone is a masterclass in graceful degradation.
The scanning layers catch almost everything. But what about when the mechanical parts fail? A diverter arm that's supposed to push a bag onto the Denver spur gets stuck. Or a motor burns out at peak load.
This is where the physical design mirrors the scanning philosophy. Conveyor belts have backup motors. The diverters — those mechanical arms that shunt bags onto the right spur — are built with fail-safe positions. If power cuts out, the diverter defaults to straight-through. The bag goes to a holding area, not onto the wrong plane.
Default to neutral. Don't make a decision if you can't be sure it's the right one.
Which is exactly the principle in aircraft autopilots and nuclear control rooms. When the system loses certainty, it hands off to a known safe state rather than guessing. In baggage terms, a bag in a holding area is a solvable problem. A bag on a flight to Singapore when it should be in Seattle is a much worse problem.
The cost of a false positive is wildly higher than the cost of a delay.
The system is designed around that asymmetry. The SCADA software — Supervisory Control and Data Acquisition — tracks every bag's position in real time. Each time a bag passes a scanning point, the system timestamps it and updates its location on a digital map of the conveyor network. If a bag misses its flight window, the SCADA doesn't just shrug. It reroutes to the next available connection, or sends it to a re-tagging station, or flags it for a handler.
The bag has a digital shadow moving alongside the physical one.
That shadow persists even when things go wrong. Which brings us to the human layer, because for all the automation, these systems are not designed to run without people. At JFK, each shift has fifty to a hundred baggage handlers monitoring the system — clearing jams, pulling bags that have gone around the divert loop too many times, manually entering tag numbers at resolution stations.
That's the part that surprises me. Fifty to a hundred people per shift sounds like a lot for a system that's supposed to be automated.
It's the resilience principle Daniel was getting at. The goal isn't to eliminate human intervention — it's to minimize it to the point where the humans only handle genuine exceptions. When a bag hits the manual resolution station, it's because multiple automated layers already failed. The tag is torn, the RFID is dead, three recirculation passes didn't work. A person reads the printed license plate number, types it in, and the bag is back in the system.
The human isn't the weak link. The human is the safety net under the safety net.
The design makes that net as small as possible without removing it entirely. There's a case study from London Heathrow that illustrates this perfectly. A bag with a damaged tag — partially torn, barcode unreadable, no RFID — went through the scanning tunnel three times, failed each pass, got diverted to a manual resolution station. A handler read the printed number, entered it, and the bag was back on track. Total delay: under four minutes. The passenger never knew.
Four minutes of invisible failure absorbed by a system that was built expecting exactly that kind of problem.
Compare that to an Amazon fulfillment center. Their Kiva robots achieve something like ninety-nine point nine nine percent accuracy on item picking. But when a mis-scan happens, the recovery is simpler — the item just takes another lap around the warehouse. No hard deadline. A baggage system has a flight departure gate. If that bag isn't on the plane in forty-five minutes, it's a missed connection, a compensation claim, a passenger standing at a carousel in another city watching everyone else's bags come out.
The hard deadline is what makes the engineering so constrained. Amazon can absorb a thirty-minute delay. The baggage system can't.
Which is why the entire architecture — multi-angle scanners, RFID, divert loops, fail-safe diverters, SCADA tracking, human resolution stations — is built around the idea that failure is inevitable, but catastrophic failure is preventable. Each layer catches what the previous layer missed. And the system runs at sixty to seventy percent capacity so there's always room for a bag to take an extra lap or two without triggering a cascade.
The throughput headroom isn't waste. It's shock absorption.
That's the design insight. In any high-reliability system — nuclear plants, aircraft autopilots, baggage handling — you don't aim for a perfect primary system. You aim for graceful degradation. When something breaks, the system gets slower, not wrong.
The passenger's experience of that grace is simply that their bag appeared where it was supposed to. The four-minute recirculation drama at Heathrow registers as nothing at all.
All of this adds up to a set of design principles that translate surprisingly well beyond airports. The first one is what we've been tracing — defense in depth. Multi-angle scanners, RFID, recirculation loops, human resolution stations. Four independent layers, any one of which can catch a failure.
None of them has to be perfect. They just have to be imperfect in different ways.
That's the key. If all your redundancy is identical — three copies of the same barcode scanner — they all fail on the same curved surface. But a laser scanner, an RFID reader, and a human reading printed text fail for completely different reasons. That's genuine independence.
The takeaway for someone building any system is: don't pour all your effort into making the primary layer flawless. Spend some of that effort on a second layer that fails differently.
The second principle is the capacity headroom. Running at sixty to seventy percent of peak isn't inefficiency — it's shock absorption. That slack is what lets a bag take three recirculation laps without jamming the whole system. In your own workflows, building in time buffers and backup processes is almost always more valuable than optimizing for maximum throughput.
The schedule with no gaps is the schedule that shatters the moment one thing goes wrong.
Which it will. As for what listeners can actually do with this — next time you check a bag, look at the tag. If it's a modern airline, that paper strip almost certainly has an RFID inlay sandwiched inside. And most airline apps now integrate RFID tracking data, so you can follow your bag's journey in something close to real time.
If your bag is delayed, the system hasn't given up on it. Between the recirculation loops, the SCADA tracking, and the manual resolution stations, there are multiple layers of recovery before a bag is truly lost. Most delayed bags are back in the system within twenty-four to forty-eight hours.
The panic at the carousel is understandable, but the engineering is quietly working the problem.
Next time you drop your bag at the counter, you'll know a little more about the journey it's about to take. But what's next for this hidden infrastructure? Because the scanning technology isn't standing still.
What are you seeing?
Computer vision is the obvious next frontier. If a bag's tag is completely destroyed — torn off, shredded, RFID chip smashed — the current system has to send it to a human. But some airports are already piloting camera systems that visually identify bags by color, shape, brand logo, even scuff patterns. The bag itself becomes the identifier.
The system learns what your bag looks like and tracks it that way.
Which raises the privacy question in a new form. Right now, the system knows a license plate number associated with a bag. It doesn't necessarily know whose bag. But a camera network that can recognize your specific Samsonite by its dent pattern — that's a different kind of tracking entirely.
It's not hard to imagine that data being useful to someone other than the baggage handler.
The bigger shift might come from passenger-side tracking. There's active research into using UWB — ultra-wideband — or Bluetooth beacons in your phone to track your bag in real time. You'd get a live map of exactly where your luggage is, from check-in to the cargo hold.
Which sounds great until you realize the airport also has that map. And now they know exactly where you are, exactly where your bag is, and exactly how those two paths relate.
The tradeoff is genuine. Better bag tracking probably means more surveillance. And tagging might eventually disappear entirely — your bag is just visually recognized and paired to your phone's location. No paper tag, no RFID, no barcode. But the infrastructure that makes that possible is also an infrastructure that never loses sight of you.
Something to sit with the next time you're watching your bag vanish through those rubber flaps.
Now: Hilbert's daily fun fact.
Hilbert: The vibrant yellow pigment used in eighteenth-century Ethiopian illuminated manuscripts — particularly those produced in the Gondarine period — was often derived from a local arsenic sulfide mineral called orpiment. But by the seventeen-eighties, some Gondarine scribes were experimenting with a substitute made from the dried and ground petals of a flowering shrub native to the Comoros Islands, which produced a nearly identical hue but faded to a soft ochre within a decade.
I don't know what to do with the phrase "soft ochre within a decade.
That was our producer, Hilbert Flumingtop. This has been My Weird Prompts. If you enjoyed this episode, tell someone about it — or better yet, leave us a review wherever you listen. We'll be back next week with whatever Daniel throws at us.
