https://amongst-other-things.com/ Marginalia, read aloud. Short, deeply-researched audio histories — aviation and the crashes that rewrote design, the engineering behind flight and spaceflight, and the early American West — told as stories, not summaries. en-us Mon, 13 Jul 2026 00:00:00 +0000 Amongst Other Things Marginalia, read aloud. Short, deeply-researched audio histories — aviation and the crashes that rewrote design, the engineering behind flight and spaceflight, and the early American West — told as stories, not summaries. episodic false Amongst Other Things REPLACE_WITH_YOUR_EMAIL@example.com The finale of Follow the Watt. For seven episodes we walked one watt across one Iowa campus — through the interconnection queue, the substation, the batteries and generators, the copper, the chip, the cold plate, the cooling tower. Every stage, we said, can fail. Tonight we do the math that promises they won't all fail at once. And the first thing the math tells you is brutal: chain eight subsystems that are each ninety-nine point nine percent reliable, and the whole is worse than any one of them — about seventy hours of downtime a year. The building exists to fight that multiplication. Mon, 13 Jul 2026 00:00:00 +0000 technology/concepts/061-reliability-tiers-redundancy 3803 false Episode seven of Follow the Watt, and the second half of the heat story. Last time we got the heat off the chip and into warm water at forty-five degrees — still trapped inside the building. Tonight it finally leaves. And the way it leaves is almost insulting in its simplicity: a hundred-plus megawatts of heat exits a hyperscale AI campus the same way a wet towel cools the back of your neck. Evaporation. We walk the facility water loop end to end — CRAHs in the air hall, the chillers you switch off whenever Iowa lets you, cooling towers versus dry coolers versus hybrids, and the genuine water-versus-electricity trade hiding behind a WUE of zero-point-three-zero. Mon, 13 Jul 2026 00:00:00 +0000 technology/concepts/060-facility-cooling-heat-rejection 3283 false Episode six of Follow the Watt, and the first of the heat half. Five episodes delivered one watt to a square centimeter of silicon; last time it became heat and had nowhere to go. Tonight the chase begins — getting that heat off the chip before the chip cooks itself, at a heat flux that, per square centimeter, rivals the inside of a nuclear reactor, with hotspots reaching toward a rocket nozzle. The counter-intuitive claim this hour is built to prove: air didn't fail from neglect. It hit a physical ceiling no engineering cleverness could move, and the whole industry is now running, not walking, toward the only tool physics left standing — water. Mon, 13 Jul 2026 00:00:00 +0000 technology/concepts/059-chip-and-rack-cooling 3164 false Episode five of Follow the Watt, and the pivot the whole series turns on. Four episodes got the watt to the rack at four hundred fifteen volts; tonight it goes down the last stretch — through the power supply, through a voltage regulator that steps forty-eight volts down to under one, and into a square centimeter of silicon that eats fifteen hundred amps at less than a volt. The counter-intuitive claim this hour is built to prove: the last centimeter of copper is the hardest engineering on the entire campus. The transmission tap, the transformers, the forty-four diesels — all easy by comparison to holding fifteen hundred amps within twenty millivolts across a thirteen-micro-ohm path. Mon, 13 Jul 2026 00:00:00 +0000 technology/concepts/058-rack-and-chip-power 2975 false Episode four of Follow the Watt. The watt has survived the ten-second gap and picked up its bodyguard detail — now it has to go to work, down the last four hundred meters of copper from the hall's four-hundred-eighty-volt bus to two cords dropping into a rack. This episode makes that last stretch legible: the unit substation, the overhead busway the industry ripped out of the floor and hung from the ceiling, the giant power strip called a PDU, three-phase power and the one equation worth slowing down for, and the single most load-bearing idea in the hour — why the little breaker has to die first. Mon, 13 Jul 2026 00:00:00 +0000 technology/concepts/057-power-distribution-to-the-rack 3247 false Episode three of Follow the Watt. The watt is on campus at thirty-four-five, and the grid feeding it is worth only three nines — which means it is going to drop. This episode is the ten-second gap between a grid blink and a headline, and the machinery built to fill it: double-conversion power supplies, a room full of lithium, forty-four three-megawatt diesels, and the transfer switches that hand the load from one guardian to the next. We build each one from scratch, stress the series thread hardest here — the watt never stops — and answer the question you're already forming: why not just make the batteries enormous and skip the fragile engines? Mon, 13 Jul 2026 00:00:00 +0000 technology/concepts/056-ups-batteries-generators 3332 false Episode two of Follow the Watt. Last time we won the interconnection queue and left a single watt waiting at the fence at three hundred forty-five thousand volts. Today it crosses — and we discover the hardest part was never the queue. It was the passive steel box in the middle: a transformer with essentially no moving parts, no spare on any shelf, and a wait that now runs three to four years. We build the campus substation from scratch — why three hundred forty-five kilovolts, what a transformer physically is, why Prairie Junction runs two of them, the millisecond reflexes that clear a fault before you can blink, the ring bus that heals around any gap, and the beer-and-foam trick behind power factor. Mon, 13 Jul 2026 00:00:00 +0000 technology/concepts/055-grid-interconnect-substation 3462 false Episode one of Follow the Watt — an eight-part deep dive that tracks a single watt of electricity across one hyperscale data-center campus, from the utility queue where it's fought for, to the chip where it becomes heat, and back out as warm Iowa air. We start where the machine actually starts: not with a chip, but with a question nobody used to ask — can you even get the power? The head of Nvidia says every future data center will be power-limited, and this episode is about why the man who sells the chips is telling you the chips aren't the problem anymore. Mon, 13 Jul 2026 00:00:00 +0000 technology/concepts/054-datacenter-load-and-siting 3274 false One hundred percent reliability is the wrong target — on purpose. The best reliability engineers on earth spend real effort making sure their systems aren't too reliable, because the last nine is invisible to users and exponentially expensive to buy. This episode is about the machine that budgets failure instead of chasing perfection: the steady-state feedback loop that runs in production forever, deciding when normal has stopped being normal and whether it's worth waking a human at three in the morning. Mon, 13 Jul 2026 00:00:00 +0000 technology/concepts/053-observability-prod-feedback 3557 false The biggest IT outage in history — eight and a half million machines blue-screened worldwide in a single morning — wasn't a hack, and it wasn't really a code bug. It was a delivery failure. CrowdStrike ran two release pipelines, and only one of them had a governor. We take that puzzle apart and rebuild the whole discipline of progressive delivery out of it: feature flags as the actuator, automated canary analysis as the sensor, rings and percentages as the ramp schedule, and the governor that pulls the dial back the moment the signal goes bad. Release stops being an event you jump off a cliff into, and becomes a control system with a hand on it. Mon, 13 Jul 2026 00:00:00 +0000 technology/concepts/052-progressive-delivery 4072 false At Google's scale, a hundred perfectly-correct end-to-end tests, each ninety-nine percent reliable, come up all-green only about thirty-seven percent of the time. That single piece of arithmetic breaks the intuition that testing is a correctness question. At two billion lines of code and forty thousand commits a day, testing becomes something stranger — a resource-allocation problem and a trust problem at once, governed by three questions. Which tests do you run when you can't run them all? When a test goes red, do you believe it? And when everything is green — do you believe that either? Sun, 12 Jul 2026 00:00:00 +0000 technology/concepts/043-testing-in-ci-at-scale 3211 false In nineteen seventy-six, Stuart Feldman wrote Make over a weekend because a colleague had just wasted a morning debugging code that was already fixed — he simply hadn't recompiled. This episode opens the black box under last time's ten-minute build: the dependency graph every build tool secretly walks, the hermetic sandboxing that lets you trust a cached result you didn't compute, and the remote caches that let a whole company skip work a single teammate already did once. Sun, 12 Jul 2026 00:00:00 +0000 technology/concepts/042-build-systems-monorepos 4384 false A single un-copied file made Knight Capital insolvent in forty-five minutes. This episode isn't about that disaster — it's about the machine that was supposed to prevent it. We walk mechanism by mechanism through modern software delivery: continuous integration and the ten-minute build, the fail-fast deployment pipeline, trunk-based development, the test pyramid, flaky tests and why they quietly kill a whole team's trust, and the deployment patterns — blue-green, canary, feature flags, expand-contract — that let you be wrong on purpose without taking users down with you. Sun, 12 Jul 2026 00:00:00 +0000 technology/concepts/041-ci-cd-pipelines 4055 false Any signal, however jagged, is a sum of pure sine waves. That one stubborn claim got Joseph Fourier laughed out of the room by the greatest mathematicians alive in eighteen oh seven — and then quietly became the load-bearing idea inside every phone call, MRI scan, JPEG, and Wi-Fi packet on Earth. This deep dive walks the whole mechanism carefully: orthogonal decomposition, the Fourier series, the continuous transform, the DFT, sampling and aliasing, windowing and Gibbs, and the algorithm that made it all run in real time — the Fast Fourier Transform. Tue, 07 Jul 2026 00:00:00 +0000 technology/concepts/040-fourier-transforms 4853 false Late summer, eighteen twenty-three. A trapper named Hugh Glass walks between a grizzly sow and her cubs in a South Dakota riverbed, and the bear nearly kills him. His friends bury him alive and run. Then he gets up and crawls two hundred miles to find them. Tue, 30 Jun 2026 00:00:00 +0000 montana-history/hugh-glass/crawls-home 1033 false In the spring of eighteen oh five, Meriwether Lewis wrote that the white bears the Hidatsa had warned him about were no match for a Kentucky long rifle. Within six weeks, his men were jumping off twenty-foot cliffs to escape them, his best hunters were emptying eight balls into a single bear before it would die, and Lewis himself was wading backward into the Missouri with an empty rifle and a spear. Tue, 30 Jun 2026 00:00:00 +0000 montana-history/colter/white-bears-missouri 1146 false In the spring of eighteen ten, the first American mountain man — the one who saw Yellowstone, the one who ran naked across a hundred miles of plain to escape the Blackfeet — walked into a stockade on the Jefferson River and quit. Out loud. In front of witnesses. He was home in six weeks. Tue, 30 Jun 2026 00:00:00 +0000 montana-history/colter/quits-mountains 992 false In the spring of eighteen oh seven, a Spanish-born merchant from St. Louis paddles up the Missouri River and intercepts John Colter heading the other way — a week from civilization, after almost four years with Lewis and Clark. The merchant turns him around. Six months later, at the mouth of the Bighorn, they build the first American outpost on the upper Missouri. Tue, 30 Jun 2026 00:00:00 +0000 montana-history/colter/manuel-lisa-fort 1257 false In the winter of eighteen oh seven, a former Lewis and Clark scout named John Colter walked five hundred miles alone through country no one from his world had ever seen. He came back with stories about boiling springs and steaming canyons — and spent the rest of his short life being laughed at for them. Tue, 30 Jun 2026 00:00:00 +0000 montana-history/colter/colters-hell 1169 false In the fall of eighteen oh eight, a Lewis and Clark veteran named John Colter is stripped, asked if he can run, and turned loose on a six-mile plain of cactus with a Blackfeet war party at his heels. He has no clothes, no weapon, no help, and roughly two hundred fifty miles between him and the nearest American fort. This is the story of what happened next. Thu, 25 Jun 2026 00:00:00 +0000 montana-history/colter/naked-run 959 false On a Monday in June twenty twenty-six, AbbVie agreed to pay almost eleven billion dollars for Apogee Therapeutics — a company three and a half years old, with no products on the market and a lead drug that hasn't finished a Phase 3 trial. This episode works through why that price is not crazy, walking from the molecule up to the deal. Tue, 23 Jun 2026 00:00:00 +0000 health/landscape/002-apogee-therapeutics-abbvie-acquisition 901 false On April eleventh, nineteen seventy, Apollo thirteen had its first emergency — two days before the famous one. Climbing on its second stage, the Saturn V's center engine started flexing three inches at sixteen hertz, accelerations roaring past thirty g, one cycle of growth away from tearing the thrust structure apart. The crew never knew. This is the story of POGO, the closed-loop instability where a rocket shakes itself to death with no wind and no turbulence — and the story of Coupled Loads Analysis, the discipline built to see it coming before anyone lights the candle. Tue, 23 Jun 2026 00:00:00 +0000 aviation/concepts/007-coupled-loads-analysis-for-launch 6698 false This is the capstone finale. Episode thirty-eight ended on a cliffhanger — "thirty-nine is where the metal wins" — and this is where it wins. We take the coupled thermo-plastic loop from thirty-seven and thirty-eight, crank it to high rate so the heat can't diffuse away, and watch the feedback localize into a razor-thin adiabatic shear band, microns wide, where the metal sheared clean through itself, mirror-smooth, because for a few microseconds it went molten-soft. A ballistic plug punched out of armor. A serrated titanium chip. A depleted-uranium penetrator that self-sharpens by failing while tungsten mushrooms by refusing to. Same physics, bug versus feature. Sat, 20 Jun 2026 00:00:00 +0000 technology/concepts/039-thermo-viscoplasticity-adiabatic-shear-bands 3863 false Episodes thirty-six and thirty-seven built the coupled, self-heating thermo-elasto-plastic physics — the two-way loop where plastic work makes heat, heat softens the metal, and softer metal flows more freely. They left us with a complete map and no vehicle. This is the episode that builds the vehicle: how you actually solve that coupled system on a computer without it diverging. The monolithic option — one big Newton solve with a two-by-two block Jacobian that is, for the first time in the whole series, non-symmetric, because the two coupling blocks are not transposes of each other. The staggered option — split the problem into sub-steps, except the naive isothermal split goes conditionally unstable and blows up, while the adiabatic Armero–Simo split is unconditionally stable. Fri, 19 Jun 2026 00:00:00 +0000 technology/concepts/038-coupled-thermo-elasto-plastic-solve 3752 false Episodes thirty-four and thirty-five built the von Mises yield surface and treated it as fixed furniture — a wall in stress space that never moves. They quietly held the temperature constant: a thermostat bolted on without comment. This episode rips the thermostat off. Take a real metal and deform it fast — a bullet, a forging hammer, a cutting tool — and roughly ninety percent of the plastic work turns straight into heat, the metal warms, the yield surface shrinks, and a softer metal flows more easily, which dumps in more heat. The metal is heating itself soft, in a loop. Thu, 18 Jun 2026 00:00:00 +0000 technology/concepts/037-thermo-plasticity-temperature-dependent-yield 3320 false The series finale. For seven episodes we balanced an inverted pendulum on a cart by pretending two things were true: that we could trust the model, and that we could trust the sensors. Both are lies. This episode drops them. With a motor that randomly shoves the cart and an encoder that jitters on every read, we ask the only question left — given a model you half-trust and a measurement you half-trust, what is the optimal estimate of where you really are? The answer is the Kalman filter, and it turns out to be last episode's observer with one thing swapped: a gain that is provably the best. Wed, 17 Jun 2026 00:00:00 +0000 technology/concepts/051-kalman-filter-optimal-estimation 2618 false For six episodes we balanced an inverted pendulum by feeding the whole state back through a gain. We cheated. A real cart measures its position and its angle — and nothing else. The two velocities the controller depends on are not on any sensor. So where do they come from? They come from a piece of software that runs a copy of the cart, watches where its prediction disagrees with the real sensor, and corrects itself toward the truth. That is the Luenberger observer, and it is the first half of the estimator the whole series has been pointing at. Wed, 17 Jun 2026 00:00:00 +0000 technology/concepts/050-observers-hidden-state 2656 false Two yes-or-no questions decide whether the whole modern control toolkit even applies to a machine. Can the input steer every state? Can the output reveal every state? Controllability and observability are exact mirror images of each other — one theorem read two ways — and Rudolf Kálmán wrote both down in a single 1960 paper. We build the rank tests from physical intuition on the cart-and-pendulum we've carried all series, then unlock the superpower they grant: full-state feedback that can place a system's poles anywhere you want. Wed, 17 Jun 2026 00:00:00 +0000 technology/concepts/049-controllability-observability-state-feedback 2623 false For four episodes we watched the cart-pendulum from the outside — one input, one output, a single ratio in the s-domain. This episode walks inside. State space is the modern representation: instead of a transfer function, you write down the state — the smallest set of numbers that captures everything the machine currently remembers — and let a matrix carry it forward in time. We build the language from scratch: x-dot equals A-x plus B-u, the four matrices in plain words, the matrix exponential as the operator that slides one freeze-frame to the next, and the bridge back to classical control. Then the recognition scene: the eigenvalues of A turn out to be the exact same poles we met in episode two. Eigenvalues are destiny, in a second language. Wed, 17 Jun 2026 00:00:00 +0000 technology/concepts/048-state-space-representation 3301 false A loop that balances a pendulum perfectly in simulation can still topple on the bench — because the real machine is never quite the model, and the sensors always lag. This episode is about the cushion: the gain and phase margins that say exactly how much extra delay, gain, or model error a feedback loop can absorb before it tips over into instability. We build the frequency domain from a child on a swing, walk the Bode plot and the Nyquist criterion in plain words, and arrive at the number that killed a fighter prototype and nearly wrecked a Space Shuttle on its own runway. Wed, 17 Jun 2026 00:00:00 +0000 technology/concepts/047-frequency-domain-bode-nyquist 3532 false We finally close the loop. The PID controller — proportional, integral, derivative — is the three-term workhorse behind most of the control loops on Earth, and it was reverse-engineered from watching a sailor steer a ship. We walk what P, I, and D each buy and what each costs: P leaves a permanent gap, I erases it but can wind up into a runaway, D anticipates but amplifies noise. Then we meet the root locus, Walter Evans's trick for sketching where the closed-loop poles travel as you crank a gain — a movie of the s-plane instead of a polynomial solved in the dark. Wed, 17 Jun 2026 00:00:00 +0000 technology/concepts/046-pid-and-root-locus 2399 false Every object has frequencies it secretly wants to ring at. Hit one hard enough and the thing tears itself apart. Those frequencies have a name — poles — and a precise address on a map called the s-plane. In this episode we turn the inverted pendulum's differential equation into algebra using the Laplace transform, find the four numbers that decide its fate, and pin its instability to a single dot in the wrong place. Wed, 17 Jun 2026 00:00:00 +0000 technology/concepts/045-transfer-functions-poles-stability 2868 false A broom balances on your open palm for minutes. A pencil falls before you can blink. Same physics, wildly different outcome — and the gap between them is a billion-dollar idea called feedback. This is episode one of an eight-part series that takes a single unstable machine, an inverted pendulum on a motorized cart, and builds it all the way up to a Kalman filter. No transforms, no matrices yet. Just the one idea everything rests on: measure, compare, correct, repeat. Wed, 17 Jun 2026 00:00:00 +0000 technology/concepts/044-feedback-and-the-control-problem 3188 false Clamp a steel bar between two rigid walls and warm it a hundred degrees. Nothing moves — every instrument that measures displacement reads zero — and yet the bar is now pushing on the walls at roughly two hundred fifty megapascals, right at the edge of yield. In ordinary linear elasticity, zero strain means zero stress. So where does a quarter-gigapascal come from in a bar that, by every measure of motion, did nothing? This episode is the story of the missing term. Wed, 17 Jun 2026 00:00:00 +0000 technology/concepts/036-thermo-elasticity-duhamel-neumann 2321 false In the 1980s, the finite-element codes that simulated metal yielding had a quiet bug that wasn't a bug. They computed the right stresses. They landed on the right yield surface. And then they crawled — Newton's method, which is supposed to double your accuracy every step, was limping along adding a fixed few digits at a time. The culprit turned out to be the most physically honest object in the whole theory: the textbook elastoplastic stiffness, the tangent you get by differentiating the rate equations the way the constitutive law tells you to. It was right about the material and wrong about the algorithm. This episode is the story of that gap and the 1985 paper that closed it. Mon, 15 Jun 2026 00:00:00 +0000 technology/concepts/035-plasticity-return-mapping-tangent 3761 false Below yield, stress is a clean formula — the elasticity tensor double-dot the strain, the same map we built in Linear Elastostatics. Feed in the strain, read off the stress, done. But the instant a metal yields, that map breaks. The same strain can now produce different stresses depending on everything the material did before. The constitutive law stops being a function and becomes a decision — one that has to be made at every point, every step, by consulting a history the material is carrying with it. This episode walks that change out loud, with no page in front of you, all the way from the additive strain split to the continuum elastoplastic tangent. Mon, 15 Jun 2026 00:00:00 +0000 technology/concepts/034-plasticity-yield-flow-drucker 2846 false In episode thirty-two, a material grew a memory — Boltzmann's hereditary integral, the stress at any instant written as the whole history of strain increments, each one faded by how long ago it happened. It was physically perfect and computationally lethal: evaluate it naively at every timestep and the work grows like the square of the number of steps, while the strain history piles up into an array that never stops growing. This episode is about the fix — and the fix is a small miracle of matched physics and algorithm. Mon, 15 Jun 2026 00:00:00 +0000 technology/concepts/033-viscoelasticity-recursive-update 2762 false Hand a finite-element solver the law of linear elasticity and stress is a function of strain — where the body is right now. But pour pitch into a funnel and wait eighty years, or drop Silly Putty and then leave it on the table overnight, and that picture breaks. The same strain today can carry two different stresses, depending on the path that got you there. Stress stops being a function of state and becomes a functional of history. This episode builds the machine that makes that precise — Boltzmann's hereditary integral, the convolution that says today's stress is the whole history of strain increments, each one faded by how long ago it happened. Mon, 15 Jun 2026 00:00:00 +0000 technology/concepts/032-viscoelasticity-hereditary-integral 2816 false Last episode we bought a stored-energy function and a beautiful postulate: the stress is the gradient of an energy. This episode we have to actually solve for the displacement field — and the comfortable single stiffness solve of linear elastostatics is gone. The boundary-value problem is nonlinear, Lax–Milgram does not apply, and the answer is a Newton iteration whose Jacobian carries a term linear elasticity never had. We walk that tangent out loud, with no page in front of you: the material tangent — four times the second derivative of the energy in C, recomputed at every step because the rubber's stiffness changes as you stretch it — plus the geometric, initial-stress tangent that carries the current stress straight into the stiffness and is exactly why a column buckles and a guitar string sings sharp under tension. Mon, 15 Jun 2026 00:00:00 +0000 technology/concepts/031-hyperelasticity-newton-tangent 2643 false Take a solid rubber block, clamp one face, and twist the other through twenty degrees. Nothing stretches. Nothing compresses. The material stores no energy at all. And yet the strain measure that serves engineers perfectly for a steel bridge — the linearized strain of episode 022 — reports that the block has deformed. That single lie, a spurious strain conjured out of a pure rotation, is where this episode begins, and it forces us to rebuild elasticity from the ground up: a new kinematics that tells the truth under rotation, three different stresses that only agree when nothing much happens, and a constitutive law where stress is no longer proportional to strain but the gradient of a stored-energy function. Mon, 15 Jun 2026 00:00:00 +0000 technology/concepts/030-hyperelasticity-energy-function 3072 false A jetliner can be perfectly designed and still kill people if the paperwork doesn't prove it. In the closer of our Structural Analysis series, we follow one load all the way home — from air pressure on a wing to a boxed margin on page eleven of a signed, archived stress report. We walk the regulation, the anatomy of the report, a real worked example down to the bolt, and the testing pyramid that turns analysis into proof. Mon, 15 Jun 2026 00:00:00 +0000 aviation/concepts/006-certification-reporting 3245 false The entire output of a stress engineer — the thing the FAA accepts as proof that a wing spar will survive — comes down to one arithmetic act: take the worst load, find what the metal can carry, divide, subtract one. If the answer is positive, people fly. This episode walks through everything that has to be true for that subtraction to mean anything: the 1.5 factor nobody ever derived, the allowables you design to that almost no real part actually has, the dozen failure modes a single part can have, and the second engineer who re-does your whole calculation before either of you signs. Mon, 15 Jun 2026 00:00:00 +0000 aviation/concepts/005-margins-of-safety 3491 false An aircraft wing might be sized by two or three flight conditions out of ten thousand. So how does a single load travel from a million-row table of forces and moments, through a finite element model, into the handful of certified numbers a stress engineer actually signs against? This is the machine in the middle — the load database that recovers internal loads, traces every value back to its source, and envelopes the worst-of-the-worst onto each part. Mon, 15 Jun 2026 00:00:00 +0000 aviation/concepts/004-load-databases 2776 false A wing that ripped itself off in twenty seconds, with no gust, no overload, and no pilot error. The structure did it to itself. In part three of our walk through aerospace structural substantiation, we leave the static world behind and follow the load into the place where the airframe is no longer frozen — where it resonates, oscillates, and can pump energy into its own destruction. Mon, 15 Jun 2026 00:00:00 +0000 aviation/concepts/003-structural-dynamics 2765 false A real wing is a continuum — billions of atoms, riveted skins, fastener holes, sealant — and no computer on Earth can solve it. So loads and stress engineers do something that sounds reckless and is actually the whole craft: they replace the airplane with a few hundred thousand springs that are wrong on purpose. This is episode two of Structural Analysis, and it's about how you turn an airframe into a finite element model. Mon, 15 Jun 2026 00:00:00 +0000 aviation/concepts/002-model-creation 3157 false A modern airliner's wing gets bent twenty-five feet upward in a hangar before the type ever carries a passenger — to a force that traces back, through hundreds of thousands of load conditions, to one number nobody actually invented. This is episode one of a six-part deep dive that follows a single load from the air to the archive, and it starts upstream of everyone: with the loads engineer who decides what the structure has to survive before a single stress report is written. Mon, 15 Jun 2026 00:00:00 +0000 aviation/concepts/001-loads-development 3044 false In the first half of 2025, strip data centers out of the American economy and growth was basically zero. Four companies are now planning to spend three-quarters of a trillion dollars in a single year — more than the entire output of Switzerland — on buildings full of chips. Is this the most important infrastructure bet in history, or the most expensive mistake? And how would you even know? This episode hands you a tool for telling the difference. Sat, 13 Jun 2026 00:00:00 +0000 technology/economics/001-the-ai-capex-ledger 3491 false Rome, February nineteen sixty-three. A dying professor's last recorded words are not about his family or his life's work — they're about a proof. A week earlier, his student had handed him the answer to a question he'd posed at the end of his career: does a heavy elastic body, resting under its own weight on a frictionless floor, have an equilibrium you can actually find? It sounds trivial. It was not. The hard part is that you don't know in advance where the body touches the floor — the contact region is itself part of what you're solving for. The boundary is part of the answer. Sat, 13 Jun 2026 00:00:00 +0000 technology/concepts/026-unilateral-contact 2422 false Every time an engineer trusts a finite-element stress plot — a wing spar, a bridge deck, a hip implant — they're trusting something nobody ever prints on the report: that the problem they handed the solver actually has a solution, and only one. This episode is about that fine print. We take the linear-elastic boundary-value problem from its strong, pointwise PDE form all the way to a unique weak solution, and we walk the proof out loud, with no page in front of you. The spotlight is coercivity: why the elasticity tensor being positive-definite, plus Korn's first inequality on a clamped body, forces the strain energy to control the whole displacement field — and why pulling out the clamp quietly destroys the whole guarantee. Sat, 13 Jun 2026 00:00:00 +0000 technology/concepts/022-linear-elastostatics 3065 false This is where the whole series starts. Before any of the immersed, enriched, contact-and-topology machinery in the later episodes makes sense, you need three words and one theorem: a form that is bounded, a form that is coercive, and a load that is continuous. That is the Lax–Milgram checklist, and once you've seen it, you'll see it under every well-posedness argument we ever run. We build it from the ground up — Sobolev's reframing of what a "derivative" even is, the affine trial space and the lifting trick, then the proof itself, walked one spoken inequality at a time, with the Korn-inequality digression exactly where it earns its keep: the single glued edge that keeps the whole argument from collapsing. Sat, 13 Jun 2026 00:00:00 +0000 technology/concepts/021-mathematical-foundations 3334 false A simulation of one millimeter of laser-melted metal can burn a hundred thousand CPU-hours on a supercomputer — and the model that has to handle the whole part runs by deleting the physics entirely. This is the sequel to "The Loop That Can't Catch the Pool," and it goes underneath the melt pool to the math: the energy equation term by term, recoil pressure that runs away exponentially, the Marangoni flow that reverses on a trace of sulfur, and the four families of numerical methods that each live at one rung of a ladder spanning ten orders of magnitude. Sat, 13 Jun 2026 00:00:00 +0000 technology/concepts/020-melt-pool-physics-and-numerical-methods 5465 false For five episodes this cluster has been quietly running away from one problem. Immersed finite elements skip the mesh and dunk the geometry in a dumb background grid. XFEM grows a crack on a fixed mesh because remeshing as it advances was called computationally fatal. The mesh-quality episode told you what a good element looks like. Every one of those is a clever way to not do the thing this episode is finally about — turning a dirty CAD solid into a hundred million well-shaped elements that fill it, conform to its boundary, and don't crash the solver. This is the prequel the whole cluster kept deferring to. Fri, 12 Jun 2026 00:00:00 +0000 technology/concepts/019-mesh-generation 6009 false Everyone who has ever run a finite element solve knows the rule: skinny triangles are bad, keep your aspect ratios near one. That rule is folklore, and the folklore is wrong about which angle to fear. This episode tears the phrase "bad element" in half. A bad element hurts you in two completely different ways — it can wreck how accurately you interpolate the answer, and it can wreck how solvable the linear system is — and the two failure modes are triggered by opposite geometric features. We follow the reversal across forty years, from a nineteen fifty-seven book about hypercircles to a UC Berkeley meshing paper that finally said out loud that there is no single number called element quality. Fri, 12 Jun 2026 00:00:00 +0000 technology/concepts/018-mesh-quality 2513 false In the fall of 2012, a graduate student in Toronto trained a neural network on two gaming graphics cards in his bedroom — and won a contest by a margin nobody had ever seen. That was the starting gun. This episode walks the fourteen years that followed, from a Go board in Seoul to a database of every protein known to science to a video call where everyone but the victim was a deepfake. Fri, 12 Jun 2026 00:00:00 +0000 technology/concepts/003-ai-impacts-on-our-world 2197 false On the eighteenth of March, twenty twenty-six, the FDA approved icotrokinra — brand name ICOTYDE — the first oral peptide that blocks the IL-twenty-three receptor. By every rule of pharmacology, a nineteen-hundred-dalton macrocycle with a fraction of a percent oral bioavailability should be useless as a pill. It clears skin like an injectable. We use that paradox as the lens onto the entire psoriasis and psoriatic-arthritis market — the IL-twenty-three injectables, the IL-seventeen leaders, the oral TYK-two incumbent it was built to beat, and the biosimilar gravity every new launch now falls into. Fri, 12 Jun 2026 00:00:00 +0000 health/landscape/001-icotyde-pso-psa-landscape 2377 false Topology optimization spends almost all its wall-clock inside one operation it repeats hundreds of times — assemble the stiffness matrix, solve for displacement. And a sixty-year-old aerospace technology exists for exactly that bottleneck: superelements, which condense a whole region down to its boundary and never look at the interior again. The two ideas have barely been married. This episode works through why, and where condensation could plug into the optimization loop next. Thu, 11 Jun 2026 00:00:00 +0000 technology/concepts/017-superelements-topology-optimization 3001 false In nineteen oh four, an Australian engineer named Anthony Michell published an eight-page paper describing the lightest structure that can possibly exist — the mathematical floor no material at any budget can beat. Then the world ignored it for fifty years. The same man, the very next year, patented a bearing the size of a dinner plate that changed shipbuilding inside a decade. Two claims to immortality, same person, opposite fates. Thu, 11 Jun 2026 00:00:00 +0000 technology/concepts/016-michell-structures 950 false Every finite element computation you have ever run reduces, in its innermost loop, to the same two-hundred-year-old trick: replace an integral you can't do with a weighted sum of the integrand at a handful of cleverly chosen points. This episode opens the box that everyone treats as sealed — where the weight w and the Gauss point location in your assembly loop actually come from, why with n points you integrate polynomials of degree 2n minus 1 exactly, and what happens when a curved boundary or a sliver refuses to cooperate with the textbook rule. Thu, 11 Jun 2026 00:00:00 +0000 technology/concepts/015-quadrature 3135 false Last episode ended on a confession. Topology optimization hands you a gray density field — every cell scored somewhere between solid and void — and then asks you to draw the boundary by hand. Pick a threshold, call everything above it solid, and live with the fact that the part you certify is provably not the part the optimizer found. Whose part is it, then? This episode resolves that gray-boundary problem by changing the one thing nobody questioned: instead of asking how dense each cell is, make the location of the edge itself the design variable. Thu, 11 Jun 2026 00:00:00 +0000 technology/concepts/013-level-set-xfem 7200 false For most of human history, finding your way meant looking outside — at stars, a coastline, a compass needle pulled north. This is the story of the eighty-year quest to do it sealed in a box, in the dark, with no window at all: a machine that knows where it is purely by remembering every push it has ever felt. From a gyroscopic gunsight on a rocking battleship to a German vacuum-tube computer riding the first rocket into space, from a washing-machine-sized box that flew a B-29 across America untouched to the alarm nobody had seen that nearly aborted the first Moon landing. Wed, 10 Jun 2026 00:00:00 +0000 aviation/concepts/guidance-navigation-control 3960 false A computer was handed nothing but the outer skin of a Boeing triple-seven-class wing and told to fill the inside however it liked. It came back with curved spars, diagonal ribs, and intricate internal trusses that no human draws by hand — and when someone held up a cross-section of a hornbill's beak, it looked the same. This is topology optimization, the method that discovers structural geometry from physics alone, and this is the full two-hour, method-level walk through how it actually works. Tue, 09 Jun 2026 00:00:00 +0000 technology/concepts/012-topology-optimization 6958 false CUDA and MPI are both SPMD — you write one program and launch many copies. So why do they feel like they come from different planets? This episode walks two working developers through the mental models, line by line, kernel by message, until the split clicks. We name the index incantation, the warp, the rank, the deadlock, the race condition, and the one inversion that explains all of it: in CUDA communication is free and synchronization is the danger, while in MPI communication is the thing you write by hand and where every bug lives. Tue, 09 Jun 2026 00:00:00 +0000 technology/concepts/011-cuda-vs-mpi-programming-models 2918 false Almost everything sold as "closed-loop melt-pool control" in metal 3D printing isn't really control at all. It's monitoring with good marketing. In this full two-hour technical deep dive, Chris and Alex work through why — starting from the counter-intuitive fact that the laser isn't melting powder, it's drilling a hole into the metal, and ending at the wall the whole industry keeps running into: the physics is faster than any feedback loop you can build. Tue, 09 Jun 2026 00:00:00 +0000 technology/concepts/010-closed-loop-melt-pool-control 5869 false In nineteen forty-six an astronomer named Fred Whipple noticed something backward: a thin, spaced layer of material protects a spacecraft better than a thick solid wall. Eighty years later, that idea has a room made of woven fabric bolted to the International Space Station — a balloon you'd trust your life to, tougher against space debris than three inches of solid aluminum. This is the story of how it got there. Mon, 08 Jun 2026 00:00:00 +0000 technology/concepts/009-inflatable-space-habitat 2681 false In nineteen ninety, the most expensive science instrument ever built opened its eye for the first time — and the stars came back as smears. Hubble's main mirror had been polished to one of the smoothest surfaces in human history, and ground to exactly the wrong shape. This is the story of how that happened: a single test device trusted absolutely, two backup tests that flagged the error and were overruled, a correctly-figured mirror sitting unused in a warehouse, and a flake of black paint smaller than a grain of rice. Mon, 08 Jun 2026 00:00:00 +0000 technology/concepts/008-telescope-that-couldnt-see 3137 false For sixty years, the hardest problem in spaceflight has been getting a vehicle back down through its own atmosphere without burning up. There are only three ways to do it — let the shield burn away, insulate against the heat, or pump fluid through the skin to carry it off — and every generation of engineers has had to bet on one. This is the history of that bet, told through the people who made it: John Glenn flying a fireball by hand, the Avco technicians who filled three hundred seventy thousand honeycomb cells one at a time, the NASA scientist who crossed into SpaceX to cut his own material's cost tenfold, and the astronaut who once dangled beneath a Space Shuttle to fix it with his bare hands. Mon, 08 Jun 2026 00:00:00 +0000 technology/concepts/007-heat-shield-tiles-sweating-steel 3926 false In June of twenty twenty-four, two astronauts launched on an eight-day test flight of Boeing's Starliner and didn't come home for two hundred and eighty-six days — and when they did, they rode a competitor's capsule. This is the technical story of how it happened: not one catastrophic flaw, but the same chemistry biting the same spacecraft three different ways across five years, and an organization that kept reading each bite as a one-off. Mon, 08 Jun 2026 00:00:00 +0000 engineering-failures/disasters/002-starliner-what-went-wrong 3403 false A working SR-71 Blackbird leaked fuel onto the runway every time it taxied — by design. So did every other Blackbird. So would yours, if you were running an airframe at Mach three. This is an hour-long tour of the strangest engineering decisions in aerospace history, told as a chain of surprises: the materials people invented by accident, the failure modes nobody had words for yet, and the crashes that taught the rest of us how airplanes really break. Mon, 08 Jun 2026 00:00:00 +0000 aviation/concepts/aerospace-structural-engineering#fullhour-2026-06-08 3427 false The most powerful rocket ever flown is built out of the same material as a beer keg and a kitchen sink. SpaceX scrapped an entire carbon-fiber factory to make that choice — and it wasn't a quirk. It was the first of six interconnected engineering bets, each one counterintuitive on its own, each one only making sense in the context of the others. This episode walks through all six: the steel reversal, the methane fuel that's secretly a Mars plan, the engine the propulsion community had written off as impossible, the booster that gets caught by a tower instead of landing on legs, the "blow up and learn" iteration culture, and the one bet that still hasn't paid off. Sun, 07 Jun 2026 00:00:00 +0000 technology/concepts/006-starship 2754 false For more than twenty-five years, there has not been a single moment when every human being was on Earth. The International Space Station — the most expensive object ever built — began as a Cold War boast, became a peace treaty disguised as plumbing, and turned into a place where the most ordinary morning is one clogged filter away from death. This is the story of how two space programs born as rivals ended up sharing a home above our heads, and what it means that we plan to deliberately destroy it. Sat, 06 Jun 2026 00:00:00 +0000 technology/concepts/005-international-space-station 3074 false In the early nineteen eighties, an Australian spy plane dropped to mast height over the Indian Ocean and photographed a small winged spacecraft the West wasn't supposed to know existed. This is the sixty-year story of that shape, and how a Soviet sketch became an American spaceplane named Tenacity. Sat, 06 Jun 2026 00:00:00 +0000 aviation/frontiers/001-dream-chaser-space-plane 3600 false Twice, NASA destroyed a space shuttle and killed its crew. Twice, the engineers saw it coming. In nineteen eighty-six it was a rubber O-ring that went stiff in the cold; in two thousand three it was a chunk of foam that punched a hole in a wing. Different hardware, same story — a known danger that the institution had quietly decided to call normal. This episode makes the machine legible — what a solid rocket field joint is, what reentry does to a wing — and then walks the human decisions that turned warnings into wreckage. Fri, 05 Jun 2026 00:00:00 +0000 technology/concepts/004-challenger-columbia-engineering 3194 false On January twenty-seventh, nineteen sixty-seven, three astronauts died in seventeen seconds on a launch pad in Florida. The rocket wasn't even fueled — the test was officially non-hazardous. The men were sealed inside a pure-oxygen cabin lined with seventy pounds of flammable material, behind a hatch that couldn't open against internal pressure. We trace the chain of engineering decisions that made the fire inevitable, and the chain of human moments — Gus Grissom hanging a lemon on the simulator, a prayer-portrait given to a program manager, Wally Schirra warning his friend the night before — that almost stopped it. Mon, 25 May 2026 00:00:00 +0000 aviation/disasters/apollo-1 1377 false Between January twenty twenty-four and the spring of twenty twenty-six, the United States started building data centers at a scale with no real precedent in industrial history. This is what's actually happening — the announcements, the grid that's already breaking under the weight of them, the tax code that just chose gas and nuclear, the small Black neighborhood in Memphis where unpermitted turbines were photographed from the air, and the six-hundred-billion-dollar question that even Sequoia Capital can't answer. Sun, 24 May 2026 00:00:00 +0000 technology/concepts/002-gigawatt-buildout 3600 false On a July morning in nineteen eighty-two, a trash truck driver in Rocky Mountain National Park became the only working warning system for a dam that had been failing in slow motion for seventy-nine years. This is the story of how Lawn Lake came apart, how three people died who didn't have to, and how the disaster came down to a patch of soft lead the size of a pencil eraser and a broken axle the day before. Sun, 24 May 2026 00:00:00 +0000 engineering-failures/disasters/001-lawn-lake-dam 900 false Before there was a discipline called aerospace structures, there were two brothers, a bicycle shop in Ohio, and a hand-built wooden airplane covered in fabric meant for women's underwear. This is the story of the 1903 Wright Flyer — how it was made, what it was made of, and what made it so different from every airplane that came after. Sat, 23 May 2026 00:00:00 +0000 aviation/pioneers/wright-flyer 947 false In nineteen twenty-seven, a young California designer named Jack Northrop laminated plywood inside a concrete mold, glued the halves together, and produced an airplane that looked like a teardrop with a propeller. The Lockheed Vega became the airplane Amelia Earhart flew solo across the Atlantic — and the proof of concept that turned every airliner that came after it into a smooth metal egg. Sat, 23 May 2026 00:00:00 +0000 aviation/pioneers/vega-dc3 781 false In Southampton in nineteen thirty-four, an aircraft designer named Reginald Mitchell had a problem he had to solve: he had to fit eight machine guns inside a wing without making the wing thick enough to slow the airplane down. His Canadian aerodynamicist proposed an ellipse. The result was the Supermarine Spitfire — the wing that kept Britain out of German hands four years later, designed by a man who never knew it had. Sat, 23 May 2026 00:00:00 +0000 aviation/pioneers/spitfire 830 false In nineteen fifteen, while Europe was killing itself in the trenches, a fifty-six-year-old German professor decided he was tired of waiting for the rest of the world to catch up to metal. He built an airplane out of sheet steel — because he couldn't get the aluminum alloy he wanted — and flew it. The German Army called it the Tin Donkey. It was slow. It was ugly. And it didn't care if you shot it. Sat, 23 May 2026 00:00:00 +0000 aviation/pioneers/junkers 802 false In nineteen forty-four, Boeing built an airplane that had to fly higher than any combat aircraft had ever flown. The crews needed pressurized cabins because at thirty thousand feet, the air is unbreathable and the temperature is forty below. The structural problem was that the fuselage had to be a pressure vessel — a balloon with a tunnel running through it — and that pressure vessel had to survive thousands of cycles. Nobody had built one before. The B-29 was the first. Sat, 23 May 2026 00:00:00 +0000 aviation/pioneers/b29 718 false At Mach two cruise, the Concorde's aluminum airframe stretched by six to twelve inches. Pilots used the gap. They wedged their hats into a particular slot on the flight deck during cruise, and on descent, as the airplane cooled and contracted, the gap closed. The hat got stuck. Forever. Sat, 23 May 2026 00:00:00 +0000 aviation/frontiers/xb70-concorde 899 false In late two thousand eighteen, Elon Musk announced that SpaceX's next-generation rocket — the one supposed to fly to Mars — would be built out of stainless steel. The internet collectively assumed he had lost his mind. The math turned out to be defensible. The bet was that iteration speed mattered more than mass efficiency. The first decade of Starship results suggests the bet might be right. Sat, 23 May 2026 00:00:00 +0000 aviation/frontiers/starship 963 false The fastest airplane the United States ever built. Made out of titanium it had to buy from the Soviet Union, through CIA shell companies, while the cover story said the metal was for pizza ovens. Designed by a man who decided that if you can't make a fuel tank that doesn't leak, you shouldn't try — you should make a fuel tank that only seals itself once the airframe gets hot enough. Sat, 23 May 2026 00:00:00 +0000 aviation/frontiers/sr71-blackbird 1050 false In June of two thousand three, NASA's solar-powered Helios prototype — two hundred and forty-seven feet of wingspan, lighter than a small car — flexed into a permanent U-shape after hitting mild turbulence off Kauai. Within two minutes, it had torn itself to pieces in the air. The investigation didn't blame the structure exactly. It blamed something subtler: the math we use to design rigid airplanes does not work for very flexible ones, and we haven't finished writing the new math yet. Sat, 23 May 2026 00:00:00 +0000 aviation/frontiers/helios 963 false In nineteen sixty-nine, NASA landed a spacecraft on the moon whose pressurized hull was twelve thousandths of an inch thick. Three-tenths of a millimeter. Three sheets of household aluminum foil. Engineers genuinely worried that a dropped tool inside the cabin could puncture it. It worked twelve times. Sat, 23 May 2026 00:00:00 +0000 aviation/frontiers/apollo-lm 885 false The Boeing seven-eighty-seven was the first airliner with a fuselage built in single continuous carbon-fiber barrels, laid up layer by layer on a giant rotating mandrel and cured in an autoclave the size of a building. Half the airplane, by weight, is composite. It burns twenty percent less fuel than its predecessor. And it represents a structural bet — that composites can replace aluminum for primary aircraft structure — whose long-term consequences we will not fully understand for another fifty years. Sat, 23 May 2026 00:00:00 +0000 aviation/frontiers/787 1018 false In the fall of nineteen fifty-nine and the spring of nineteen sixty, two brand-new Lockheed Electra turboprops broke up at altitude over Texas and Indiana. Ninety-seven people were killed. The chief test pilot flew the same conditions and could not reproduce the failure. The investigation took most of a year and turned on a phenomenon that didn't have a name yet — and that, today, every aerospace engineering student learns as a fundamental lesson in structural dynamics. Sat, 23 May 2026 00:00:00 +0000 aviation/disasters/electra-flutter 887 false In nineteen seventy-two, an American Airlines DC-10 over Windsor, Ontario, lost a cargo door at altitude. The floor collapsed. The captain saved everyone aboard. The fix was known, cheap, and not made mandatory. Twenty-one months later, the same exact failure killed three hundred forty-six people outside Paris. The fix was the same fix. The price of waiting to make it mandatory was the largest single-airplane disaster in aviation history up to that point. Sat, 23 May 2026 00:00:00 +0000 aviation/disasters/dc10-door 913 false The world's first commercial jet airliner. Two fatal in-flight breakups in three months. The Royal Navy lifting wreckage off the floor of the Mediterranean. A British prime minister handing the engineers a blank check. An airliner drowned in a giant water tank until it finally split open. And, sixty years later, a textbook story that still gets the cause wrong. Sat, 23 May 2026 00:00:00 +0000 aviation/disasters/comet 1538 false At twenty-four thousand feet, halfway between Hilo and Honolulu, the entire upper section of a Boeing seven-thirty-seven peeled off in one explosive failure. Eighteen feet of skin. Gone. A flight attendant was sucked out. The captain brought the airplane down to Maui with no roof and one fatality out of ninety-five souls. The cause was a kind of structural decay nobody had named yet — and the regulations that followed changed how every airliner in the world is maintained. Sat, 23 May 2026 00:00:00 +0000 aviation/disasters/aloha-243 1067 false By nineteen thirty, the global aviation industry was building stressed-skin metal airliners. There was just one problem. The math everyone was using said the airplanes shouldn't have been flying. Real thin metal sheets buckled at a fraction of the load the textbook equations predicted, and nobody could explain why. A Hungarian named Theodore von Kármán arrived at Caltech and spent twenty years figuring it out. Sat, 23 May 2026 00:00:00 +0000 aviation/concepts/von-karman 1011 false In nineteen sixty-six, NASA contracted a small California company to build a single piece of structural analysis software that the whole agency could use. The project was called NASTRAN. It became the most influential aerospace software ever written, was released to the public, and saved an estimated seven hundred million dollars in non-aerospace industries — bridges, cars, buildings — in its first thirteen years. Most engineers today have never heard the name. Sat, 23 May 2026 00:00:00 +0000 aviation/concepts/nastran 773 false In late September of nineteen oh six, a German metallurgist named Alfred Wilm finished a sample of aluminum alloy on a Saturday afternoon, didn't have time to test it before going home, and left it sitting on a shelf for the weekend. On Monday morning, he ran the hardness test, and the metal had gotten harder by itself. Sat, 23 May 2026 00:00:00 +0000 aviation/concepts/duralumin 764 false