Then someone found the receipts.

For forty years, the official consensus was clear: dietary fat causes heart disease. Cut the fat, save the heart. The entire food industry restructured around it — low-fat yogurt, fat-free cookies, margarine instead of butter. Governments published food pyramids with fat at the narrow top, the thing to avoid. Hospitals put patients on low-fat diets. Schools pulled whole milk. A generation learned to read nutrition labels for one number.

When the anomalies piled up, the framework absorbed them. People on low-fat diets weren’t getting healthier — they were getting heavier, sicker, more diabetic. But the patches came fast: they weren’t disciplined enough, they were eating the wrong carbohydrates, they needed more exercise, the effect was long-term. Every contradictory result got folded into the existing model. The abstraction stretched.

Then, in 2016, researchers at UC San Francisco found the paper trail. Internal documents from the 1960s showed that the sugar industry had paid Harvard scientists — the equivalent of $50,000 in today’s dollars — to publish a review that shifted the blame for heart disease from sugar to fat. The review appeared in the New England Journal of Medicine. It cited the studies the sugar industry selected. It didn’t disclose the funding. And it set the trajectory of nutritional science for half a century.

But the receipts were the narrative capstone, not the cause. The consensus had been cracking for over a decade before anyone found them. The instruments broke first: the Women’s Health Initiative — the largest dietary study ever conducted, using their own methods — reported in 2006 that low-fat diets didn’t reduce heart disease risk.

When your best tool produces the wrong answer, you can’t blame the tool without undermining everything else it’s validated. The vocabulary cracked next: writers and researchers began splitting “fat” into saturated, trans, and omega-3 — giving the public language to articulate distinctions the old framework had been collapsing into a single word.

The institutional layer eroded more slowly: meta-analyses through the early 2010s kept failing to find the expected link, and the 2015 Dietary Guidelines quietly dropped the cholesterol limit and softened the language on fat — not a reversal, but a retreat. By the time the sugar industry documents surfaced, they didn’t break the consensus. They gave the break a story.

And notice: nutritional science still hasn’t converged on a replacement. Is the real culprit sugar? Processed food? Seed oils? Metabolic syndrome as a systemic condition? There are competing candidates, and none has won. The old abstraction cracked, but the new one hasn’t landed. That gap — between the break and the replacement — turns out to matter a great deal.

This isn’t a story about people being stupid. It’s a story about how understanding works. Existing concepts absorb what they can’t explain. They stretch. They accommodate. They hold contradictions for as long as they possibly can, because the cost of abandoning a working framework is always higher than the cost of one more patch.

Until it isn’t.

The Abstraction Cycle

Here is what I think happens, across every domain where humans try to make sense of complicated things.

First, Absorption. Whatever framework we already have stretches to accommodate new observations. For most of human history, “bad air” — miasma — absorbed everything from cholera to the plague. “Melancholy” absorbed what we’d now separate into depression, anxiety, grief, bipolar disorder, and a dozen other distinct conditions. Newton’s gravity absorbed the anomalous orbit of Mercury for over a century.

Absorption isn’t failure. It’s how understanding works most of the time.  Existing abstractions are load-bearing. We don’t abandon them at the first sign of strain. And what holds it in place is not just habit — it’s a stack of constraints all reinforcing the same framework:

• Instruments that can only detect what they were built to detect
• Vocabulary that can only name what it already has words for
• Institutions organized around existing categories.

The absorption holds because all the layers hold together — which is why, when a break comes, it often cascades.

You’ve already seen this stack in action. The dietary fat consensus held because instruments couldn’t isolate single macronutrients over a lifetime, vocabulary collapsed all fats into one word, institutions had built careers around the lipid hypothesis, and industry funding had shaped the initial framing. The stack cracked layer by layer over fifteen years, each break exposing the next, until the cascade became irreversible.

Then, Anomaly. Something stops fitting. The mismatch between what the current abstraction predicts and what actually happens grows too large to patch. People on low-fat diets develop metabolic syndrome. Cholera follows water lines, not wind patterns. Mercury’s orbit drifts by 43 arc-seconds per century in a direction Newton’s equations can’t account for.

The anomaly doesn’t arrive because the phenomenon changed. It arrives because either the signal got louder — what I’d call amplitude — or our ability to detect it got sharper — resolution.

Finally, Abstraction. Something gets a name. Oxygen replaces phlogiston. Germ theory replaces miasma. General relativity replaces Newtonian gravity at extreme scales.

But there’s something underneath the naming that’s worth pausing on. Humans are pattern-finding machines. We see faces in clouds, constellations in scattered stars, trends in random noise. It’s not a flaw — it’s the engine. Every abstraction that ever landed started as someone noticing a regularity and deciding it was real. The question is: what determines which patterns stick?

I am not sure, but I think the constraint stack is the filter — and its thickness determines what kind of force a new abstraction needs to survive.

What makes an abstraction stick

When the stack is thick — instruments, vocabulary, institutions, funding, and careers all reinforcing the existing framework — a new abstraction needs enough force to crack multiple layers at once. That’s a phase transition. Wegener cracked the observational layer in 1915 (the coastlines fit, the fossils match) but couldn’t crack the mechanistic layer (no one could explain how continents move through solid rock) or the institutional layer (geologists had organized their discipline around fixed continents). One layer broke, the others held, and the abstraction died for forty years. It stuck only after seafloor spreading provided the mechanism and seismology provided new instruments and generational turnover weakened institutional resistance. Three layers cracked in close succession. Rare, slow, and looks like a revolution — because it is one.

But not every abstraction fights through a thick stack. When there’s no entrenched framework defending the space — just an unnamed gap in shared experience — the barrier can be as thin as a single missing word. Someone coins “doom scrolling” and within weeks millions of people are using it. “Quiet quitting.” “Situationship.” “Main character energy.”

These abstractions spread through demonstrated utility within a community. Someone names the pattern, others start using it because it makes their work more legible, and the name persists because it earns its place.

So the constraint stack has a dual nature. When conservative — it protects established frameworks from noise, premature ideas, and false patterns. And it’s the reason genuine breakthroughs take so long. The same mechanism that filters out wrong abstractions also filtered out continental drift for forty years. The filter doesn’t distinguish between wrong and premature.

The naming step(Abstraction), when it lands, produces two distinct outcomes:

Sometimes the replacement is clean — convergence. The anomaly points to a single missing entity or mechanism, and once it’s named, the landscape reorganizes around it. Oxygen converged because the weight-gain problem in combustion pointed to a single substance that phlogiston had inverted. One entity, one correction, one new framework.

Sometimes the new abstraction doesn’t refine the existing space — it unlocks space we couldn’t previously reach. Maxwell’s equations didn’t improve mechanics. They opened electromagnetism — a territory in which radio, electronics, and information theory could exist as questions for the first time. Nobody was failing to understand transistors in 1820. The space in which transistors are even a question didn’t exist until the abstraction created it. DNA didn’t refine biology — it opened genomics, gene therapy, CRISPR. Quantum mechanics didn’t settle an old debate — it unlocked semiconductors, lasers, quantum computing. The interpretive disagreements that persist (Copenhagen, Many-Worlds, pilot wave) aren’t a failure of the abstraction. They’re just the next cycle running in the new territory, not yet converged.

Unlock can happen from two directions. Sometimes an anomaly persists despite every increase in resolution — consciousness, turbulence, quantum gravity — and that persistence signals the phenomenon lives at a different level of organization. The cycle has to ascend. Other times, a level shift arrives first. Darwin reads Malthus and imports population-level thinking into biology. Suddenly variation and death — visible to everyone, anomalous to no one — become anomalies demanding explanation. Either way: new territory opens, new cycles begin.

Creation of Anomaly

Two things push absorption past its breaking point to Anomaly.

Amplitude — the signal gets louder. The Black Death killed a third of Europe; “bad air” strained under the weight. The 2008 crisis overwhelmed models rating mortgage-backed securities as safe — not because the models got worse, but because the scale of failure exceeded them.

Resolution — we get better at seeing. Van Leeuwenhoek’s microscope didn’t make germs exist, but it gave us the resolution to detect them. The telescope didn’t create Jupiter’s moons. fMRI didn’t invent distinct neural patterns in depression versus anxiety. The instrument changed, and the anomaly became visible.

The fuels are different too. Curiosity is a resolution amplifier — it increases detection capacity before any anomaly demands it. Someone wonders what pond water looks like through ground glass. Someone points a radio antenna at the sky for no practical reason and picks up the cosmic microwave background. Necessity drills toward a specific problem. The military needed to detect submarines, and seismometers built for that purpose revealed seafloor spreading, which finally gave Wegener’s continental drift its mechanism forty years late. And Serendipity, it requires the observer to recognize something as anomalous— penicillin, X-rays, the microwave background itself — reminds us the cycle is descriptive, not prescriptive.

The Store That Grew a Brain

The cycle doesn’t just run in laboratories. It runs anywhere people try to make sense of complicated things — including a store.

When a store opens, the founders have a model: we sell goods, customers buy them, money comes in. Then the anomalies start. Customers leave trash. Traffic creates bottlenecks at certain hours, seasonal demand the original plan can’t hold. None contradict the concept of “store” — but they break the founders’ working model of what running one requires. The absorption fails. The gap between what they expected and what they’re seeing forces new vocabulary into existence: “inventory management,” “loss prevention,” “customer experience.” Each name is a genuine abstraction — a pattern someone detected, documented, and made visible to others. That’s the ratchet which keeps on finding new optimizations, increasing the resolution.

Eventually, no single person holds all of it. The store’s understanding of itself has become emergent — distributed across specialists, embedded in processes, existing in connections between departments rather than inside any individual mind.

The fuel here is mostly necessity, not curiosity. Nobody investigates the trash problem out of intellectual wonder. They investigate because the trash is piling up and something has to change. And the documentation matters. The knowledge transfers. The sensing surface grows.

But notice what the store develops vocabulary for: things that threaten it. Trash, theft, bottlenecks, churn. It does not grow toward what the store causes — the traffic it brings, the wages it sets, the community that depends on it. Those phenomena are real, but the store has no survival pressure to name them.

Think about your own work. What abstractions didn’t exist when you started that you now use without thinking? A finance team that once talked about “profit” learns to split it into EBITDA, CAGR, gross margin, runway. None arrived because someone was curious. They arrived because the old vocabulary couldn’t hold what people were seeing.

Where It Breaks

A model is only as useful as its failure modes are honest. So let me describe where this one cracks.

The most obvious problem is false abstractions. The cycle completes but lands in the wrong place. Phlogiston: chemists named an invisible substance released during burning, and the framework held for a century before Lavoisier showed combustion involved gaining oxygen, not losing phlogiston. He replaced it with caloric theory — also wrong. The cycle ran perfectly. Twice. But phlogiston wasn’t entirely wrong — it described a real transfer during combustion, just misidentified the direction. The false abstraction was a waypoint, not a dead end.

A subtler problem is what happens when resolution increases past the signal-to-noise ratio. If higher resolution always reveals real phenomena, the model has no built-in brake. But sometimes higher resolution detects noise and calls it signal.

FICO scores were designed to measure creditworthiness. But the score itself became a determinant of the economic conditions that produce creditworthiness. Someone with a low score pays higher interest rates, gets denied rental housing, sometimes loses job opportunities — all of which reduce income stability and increase the likelihood of default. The score predicted risk, then manufactured it.

The sensing surface didn’t just observe the system. It became a force inside it.

When the Cycle Gets Weaponized

Everything I’ve described so far treats the cycle as something that happens to us — a natural process, like erosion. But a mechanism this reliable can be exploited. Keep absorption artificially stretchy and you stall understanding indefinitely. Reverse the naming mechanism and you create phenomena rather than detect them. Narrow resolution structurally and you make the anomaly invisible before it accumulates.

Understanding doesn’t just grow through absorption, anomaly, and abstraction. It also stalls, reverses, and gets hijacked. And the forces that do the hijacking are not random. They are as patterned as the growth they interrupt.

Keep absorption artificially stretchy —In 1954, the tobacco industry published the “Frank Statement to Cigarette Smokers,” launching a half-century strategy summarized in one internal memo: “Doubt is our product.” Through the Tobacco Industry Research Committee, companies funded alternative explanations for every new study linking smoking to cancer — not to prove safety, but to maintain the appearance of scientific controversy. The framework never had to break because every anomaly was met with a counter-study designed to stretch it back into shape.

Reverse the naming mechanism —When British colonial administrators imposed rigid caste categories through the Indian census, they didn’t discover a social structure — they manufactured one. Categories that had been fluid and regional hardened into legal identities, and the differential outcomes that followed (in education, land rights, political representation) were then cited as evidence that the categories had been real all along. The abstraction step ran first; the phenomenon organized itself around it.

Narrow resolution structurally — Chernobyl (1986) On the night of the explosion, the plant’s dosimeters maxed out at 0.001 roentgens per second — the true levels were 5,600 times higher, but the instruments just read “off scale.” The Soviet state then narrowed every remaining channel: no public acknowledgment for two days, no proper dosimetric monitoring of cleanup workers for nearly two months, radiation data classified at the Politburo level. The anomaly was finally detected not by Soviet instruments but by Swedish ones, over a thousand kilometers away — the signal had to escape the country before anyone could name it.

Shoulders

Thomas Kuhn described the cycle first. His Structure of Scientific Revolutions (1962) gave us normal science, crisis, and paradigm shift — the architecture that absorption, anomaly, and abstraction are built on. What I’ve tried to add is the output taxonomy, the constraint stack, and the cross-domain mechanism.

What Comes Next

The cycle I’ve described here runs inside a single domain — one framework stretching, breaking, getting replaced by something that sees more. But it doesn’t run alone. Right now, thousands of these cycles are turning simultaneously, in different fields, at different phases. And they feed each other. One cycle’s completed abstraction becomes another’s instrument.

 Anomalies in separate domains converge on the same hidden structure. A name born in biology migrates to economics and keeps working. The real story isn’t the cycle. It’s the mesh between them.

The interesting thing about the cycle is that you can’t tell from inside the absorption phase whether you’re in a framework that’s about to break or one that will hold for centuries. Both feel exactly the same. Both feel like understanding.

— Sail