Every hyperinflation in the historical record was run by people whose entire objective was extraction. A government printing money in a hyperinflation is not confused about what it wants. It wants real resources, right now, and the printing press is how it takes them. These are the most motivated maximizers you will ever find, playing a game they understand, with the throttle in their hands.
And they overshot. There is a rate of inflation that pulls the most real value out of a currency, a peak past which faster printing extracts less, not more, because the thing you are taxing, people's willingness to hold the money, collapses faster than you can raise the rate. Milton Friedman put that peak somewhere under twenty percent a year. Phillip Cagan, studying the great interwar hyperinflations in 1956, put it near twenty percent a month. Wherever exactly it sits, the striking finding, tested repeatedly across the historical hyperinflations, is that governments blew past it. Average inflation rates in these episodes significantly exceeded the revenue-maximizing ones. For long stretches, the economies were on the wrong, descending side of the curve, printing harder and collecting less.
This is Cagan's paradox, and it is the cleanest fact I know for a claim that sounds like a paradox itself: for a whole class of things you might want to extract, total extraction is worth less than partial extraction, and the people best positioned to know this get it wrong in the same direction every time. The purpose of this piece is to convince you that the pattern is general, that it has three mechanically different causes hiding under one lazy phrase, and that knowing which cause you are facing is the difference between the right move and its exact opposite.
Economists call this an interior maximum, and the most famous instance is the Laffer curve. Tax revenue is zero at a zero percent rate, because you have taxed nothing, and zero again at a hundred percent, because nobody works or reports income they get to keep none of. Somewhere between those two zeros is a peak. The shape is not controversial. What the shape does not tell you, and this turns out to be the whole story, is where the peak sits.
Look at the serious attempts to locate it and you find not a number but a spread. Peter Diamond and Emmanuel Saez, in a well-known 2011 estimate, put the revenue-maximizing top rate for the United States around seventy-three percent once you include state and payroll taxes. Other credible estimates using dynamic models land lower, in the low sixties. Push the elasticity assumption around within its plausible range and the peak slides across a band roughly from the mid-fifties to eighty percent. The disagreement is not sloppiness. It is one parameter wide, the responsiveness of taxable income to the rate, and that single parameter is genuinely hard to measure. Everyone agrees the curve bends. Nobody can tell you where from the outside.
Hold onto that, because it is the load-bearing idea. The shape is cheap knowledge. The location is expensive, contested, and often unknowable in advance, and every real failure in this family is a failure of location rather than a failure to believe in the shape.
Now watch the same structure play out in a field with fish instead of taxes, because it adds a failure mode that the tax story hides.
In 1954, Milton Schaefer wrote down the model that still governs how much you can pull out of a fish population year after year without depleting it. The logic is that a population's surplus, the amount it regrows each year and can therefore be harvested sustainably, is largest not when the stock is full and not when it is nearly gone, but in the middle. In Schaefer's model, the maximum sustainable yield sits at exactly half of the ocean's carrying capacity. The largest durable catch requires deliberately holding the population at fifty percent of what the sea could support. Restraint is not a compromise with the yield. Restraint is the yield.
Here is the part I did not expect. Since the early 1970s, fisheries practitioners in several major countries quietly modified Schaefer's model so that the maximum sustainable yield is assumed to occur at forty, or even thirty, percent of unfished biomass. By the original model, thirty percent is not the optimum. It is overfishing. The peak did not get overshot the way the hyperinflation governments overshot theirs. The peak got redefined, downward, toward the level the harvesters wanted to fish at, while keeping the name maximum sustainable yield and with it the authority of the derivation it now contradicted.
The consequence has a name too. New Zealand's orange roughy fishery was managed with a target in that redefined thirty-to-forty-percent range, and the seamount spawning aggregations collapsed. By the early 1990s many of those stocks had fallen below twenty percent of their unfished biomass. The curve was real, the managers were not ignorant of it, and the resource died anyway, because the estimate of where the peak sat drifted in the direction of interest, one modest revision at a time.
This is the failure mode the tax and inflation stories cannot show you on their own. You do not need to ignore the curve to destroy the thing it describes. You only need to be slightly optimistic about the location of its maximum, repeatedly, in a direction that happens to serve you. It is the quiet cousin of Cagan's overshoot: one blows past the peak, the other moves the peak, and both end with less of the thing they were trying to protect or extract.
The three cases above are measured. Economists have spilled decades of ink locating these peaks. What I want to argue, and here I am reasoning rather than citing a dataset, is that the same curve governs a set of capabilities in security and social systems that are almost always modeled as if more is simply better.
Start with the one that sounds most like a paradox. Suppose you found a way to break the cryptographic signatures that secure Bitcoin. The naive picture is that you have won: the coins are yours. But a complete, public break is close to worthless to you, and the reason has nothing to do with cryptography. It is exchange rate and liquidity. The moment the break is visible, confidence in the asset collapses, and it collapses faster than you can convert meaningful size into anything you can spend. You have picked a lock on a vault whose contents evaporate on the way out the door. The valuable version of that break is the quiet, partial one: take a little, tell no one, let the asset keep its price because the world still believes it is safe. Maximal, public success is the failure case. The optimum is partial, and it is partial for the same structural reason the inflation tax is: the value you are extracting rides on a substrate, here belief in the currency, that your own extraction destroys when you push it to the limit.
Counterfeiting is the same curve with an extra descending force. A modest volume of good counterfeit currency is spendable and profitable. Push the volume up and two independent things happen at once. The counterfeits inflate the money supply and erode the value of the very notes you are forging, which is the substrate-collapse mechanism again. And the volume itself is what gets you caught, because a flood is detectable in a way a trickle is not. Two separate forces bend the curve down, and the peak, the most you can profitably print, sits well below all of it.
Credential inflation is the third, and it bends for a different reason entirely. A degree, a certification, a membership is valuable because it is scarce. Its worth was made of the exclusion. Issue more of it and you do not add value proportionally, you convert the signal into a floor. When everyone has the credential, having it means nothing, and the institutions that issued freely discover they have zeroed the thing they were selling. The sociologist Randall Collins made a version of this argument in 1979 in The Credential Society, and anyone who has watched a once-distinguishing qualification become a baseline requirement has watched the peak get passed in real time.
The honest version of this essay does not overclaim. Adjacent literatures surely exist under names like optimal extraction or the attacker's dilemma, and the claim is not that nobody has seen these before. It is narrower and more useful: these scattered cases are the same curve, and lumping them together tells you something the individual folk explanations do not.
Because here is what the folk phrase costs you. When a capability turns out to be Laffer-shaped, people say "it doesn't scale," and then they stop thinking. But "doesn't scale" is a description of the shape, and the shape is the part everybody already agrees on. What you actually need to know is why it bends, because there are three mechanically distinct reasons, and they imply opposite actions.
The first reason is that the substrate collapses. Extraction destroys the thing you are extracting from. This is the inflation tax, the fishery, the crypto break: the value rides on a stock, a base, a belief, that shrinks as you pull on it. The right move when the substrate collapses is to find the interior peak and sit on it quietly, harvesting at maximum sustainable yield and never approaching the limit, because the limit is where the substrate dies. Go slowly.
The second reason is that an adaptive defense is induced. The system notices you and responds. This is counterfeiting detection, and it is every security exploit that gets patched once it is used at scale, and every trading edge that decays as the market learns it. The substrate is not destroyed; a defense is grown against you. The right move here is the exact opposite of the first: go fast, extract hard, bank the value before the response matures, and accept that the window closes. Blitz.
The third reason is that the value was constituted by scarcity in the first place. Nothing is destroyed and no defense is induced; the asset simply is the exclusion, so distributing it universally zeroes it directly. This is credentials, licenses, memberships, and the "secret" that stops being valuable the instant it is widely known. The right move is neither harvest nor blitz. It is to guard the boundary, because the boundary is the entire asset, and every unit of the thing you issue past the line is a unit of your own value you are spending.
Substrate collapse says go slow. Induced defense says go fast. Constituted scarcity says protect the fence. Those are three different instructions, and "it doesn't scale" covers all three while telling you nothing about which one you are in. The phrase is not wrong. It is empty in exactly the place where you needed it to be full.
I could end on that taxonomy and it would be a tidy essay, but it would also be dishonest, because it would leave you with the impression that the work is to identify your mechanism and then find your peak. The three economic cases are precisely the ones that show why finding the peak is the hard part, and often an impossible one.
For the Laffer curve, the peak is contested by a full parameter's width, mid-fifties to eighty percent, and reasonable economists have argued the point for decades without settling it. For the inflation tax, the peak was not just missed but blown past reproducibly, in every hyperinflation, by the maximizers themselves, and even that finding is now contested: a more recent analysis at daily rather than monthly frequency argues the economies were actually on the correct side of the curve most of the way through, which is not a refutation of the pattern so much as a demonstration that even in hindsight, with the whole dataset in hand, we cannot agree on where the peak was. For the fishery, the peak did not move on its own. It was moved, from fifty percent to thirty, by the people with an interest in fishing harder, wearing the authority of the model they were quietly abandoning.
So the practical insight is not "find the peak," because in the cases where we have the most data and the most motivated experts, the peak's location is exactly what nobody can pin down, and worse, it is the quantity that drifts toward whoever is estimating it. The practical insight is this. When you meet a capability that bends, and you now have three mechanisms to recognize one by, the shape is the cheap knowledge and you should treat it as settled. Spend your real attention on the location of the peak: who is estimating it, how wide the honest uncertainty is, and in whose interest the estimate would drift if it drifted. If the answer is that you cannot locate it from where you stand, that is not a gap to paper over with confidence. That is the finding. It is the reason restraint beats optimization here, the reason an independently measured peak beats a self-reported one, and the reason the operators who last are the ones who assume they are closer to the edge than their own numbers say. The curve is generous to the humble and merciless to the certain, and the certainty it punishes is never about the shape. It is always about the location.
The hyperinflation governments believed the shape. They were destroyed by the location. So was the orange roughy. Do not be the third example.
Sources. Cagan's paradox and the seigniorage-maximizing inflation rate: Cagan (1956) and the subsequent hyperinflation literature, including work revisiting the paradox at daily frequency that reaches a different conclusion, which is cited here as the honest state of the disagreement rather than a settled result. Laffer-curve peak estimates: Diamond and Saez (2011) at roughly seventy-three percent, with the broader range across serious estimates spanning the mid-fifties to eighty percent, the spread driven by the taxable-income elasticity. Maximum sustainable yield at fifty percent of carrying capacity: Schaefer (1954); the downward redefinition to thirty to forty percent since the early 1970s and the New Zealand orange roughy collapse are documented in the fisheries-management critique literature (Sea Around Us; ICES Journal of Marine Science, "MSY needs no epitaph but it was abused"). Credential inflation: Collins, The Credential Society (1979). The security and social cases are argued from these mechanisms rather than independently measured, and no claim of novelty is made for the connection.
An independently measured peak beats a self-reported one, and the only way to have the first is to make the record something you cannot quietly revise.
The fishery peak did not move because anyone lied. It moved one modest revision at a time, in the direction of interest, wearing the authority of the model it was abandoning. That is what a self-reported number does under pressure. Chain of Consciousness is an append-only, tamper-evident record of what an agent reasoned and did, written on the hot path and living where the application cannot reach back and adjust it. It will not tell you where your peak is. It will tell you, later and checkably, what your estimate was before the outcome was known.
See Hosted Chain of Consciousness
pip install chain-of-consciousness · npm install chain-of-consciousness
Provenance is one layer of the full agent trust stack: pip install agent-trust-stack / npm install agent-trust-stack.