The most-shared chart in bitcoin is a straight line on a log-log plot, and most of the accounts that post it could not say why it should be straight. Giovanni Santostasi can. An astrophysicist by training, he did the thing the chart's other promoters skip: he wrote down a mechanism, derived the exponents from it, and invited the result to be tested. His essay, "The Bitcoin Power Law Theory," is the document the popular versions are quietly borrowing from.
An earlier piece on this site argued that bitcoin's price-versus-time chart is a power trendline, a regression, not a statistical power law, and that the genuine power laws sit in the return distribution instead. That argument was aimed at the casual claim: the one that treats a fitted curve as a law of nature because it looks straight on the right axes. Santostasi's claim is not casual, and it should be met on its own terms.
Taken seriously, his theory is the best argument the power law has, and it is better than its critics usually allow. It is also load-bearing in one spot that cannot hold the weight: a single adoption exponent that is fitted rather than derived, propped up by a methodological stance that excuses itself from the tests which could prove it wrong. What survives the scrutiny is a strong empirical regularity worth using, not the law of nature it is dressed as.
What the Theory Claims
The model is a chain, and the chain is short. Adoption, measured by active addresses, is taken to grow as the cube of time. Network value scales with the square of users, which is Metcalfe's law, so price grows as time cubed, squared, which is time to the sixth. Miners are pinned to the edge of profitability by the difficulty adjustment, so the hash rate has to rise as the square of price, which is time to the twelfth. Three exponents, six, three, twelve, and they all fall out of one number run through two pieces of economics.
Around that arithmetic sits a feedback loop. Adoption lifts price through Metcalfe. Price pulls in hash rate. Hash rate buys security. Security earns trust. Trust feeds adoption, and the loop turns again. Santostasi's deeper point is that this shape is not a coincidence: processes where the output becomes the next input, iterated, are exactly the kind that tend to produce power laws. He is right about that. Preferential attachment, compounding, recursive growth, these are the standard generators of scale-invariant behaviour, and bitcoin's security-and-adoption flywheel is a plausible member of the family.
On its own terms the model is coherent. The exponents do not contradict each other. They multiply out of a single story.
Credit Where It Is Due
Three things the theory gets right, each of which its detractors tend to skip.
It buries stock-to-flow. The most famous bitcoin price model of the last cycle was built on the ratio of existing supply to new issuance, and it has real mathematical and conceptual problems. Santostasi names them and does not lean on the same machinery. A model that discards a popular but broken predecessor, rather than quietly reusing its appeal, is doing science and not marketing.
It is honest about its horizon. He does not promise a number for 2060; asked for one, he answers with a joke about the technological singularity and a warning not to extrapolate past 2040. The relationship, he says, is good for one or two more orders of magnitude, roughly a decade. That restraint is the opposite of the screenshots that put a dollar value on a named day. The popularisers added the false precision. He did not.
And the mechanism is more than ornament. Metcalfe's square, the difficulty adjustment holding miners at the margin, adoption slowed by the risk of an unproven asset: these are real forces. Tying the exponents to them is a genuine attempt at first principles, not a backstory bolted onto a line. This is the steelman, and it is why the theory earns a real answer rather than a smirk.
The Exponent Holding Up the Tower
Everything stands on one number. Price as time-to-the-sixth and hash rate as time-to-the-twelfth are not independent discoveries; they are adoption-as-time-cubed run through Metcalfe and through mining math. Pull the cube and the tower falls. So the question that decides the theory is narrow: where does the three come from?
The honest answer is that it is read off the history. Santostasi concedes the price-to-addresses exponent is closer to 1.95 than to a clean 2, rounded for tidiness, and he justifies the time-cubed adoption by analogy: the spread of phenomena slowed by risk-aware human decisions, the diffusion of a disease like AIDS being his example, can grow as the cube of time. An analogy is not a derivation. No step in the argument shows why bitcoin adoption must scale as the third power of time rather than the second or the fourth. The exponent is measured, then narrated. Strip the narration and the theory is one fitted number plus algebra, which is the earlier piece's power trendline wearing a lab coat.
Even the economics it leans on is softer than it looks. Metcalfe's square is itself an idealisation; a large literature finds real networks scaling closer to n log n than to n squared. The model borrows a contested approximation and treats it as a fixed gear. A law derived from a mechanism is strong. A law derived from one fitted exponent and a disease analogy is a regression with good manners.
A Theory That Declines Its Own Exam
The sharpest tell is methodological. The standard objection to any price-versus-time power law is statistical: price is autocorrelated, autocorrelated series throw off spuriously high goodness-of-fit, so a straight log-log line and a fat R-squared prove far less than they appear to. It is the objection a careful proponent has to answer.
Santostasi's answer is candid, and it is the moment the theory steps outside science. Of course price is autocorrelated, he says, because the claim is that the process is deterministic; and once causation is asserted through a mechanism, the formal tests can be set aside. As philosophy that is internally consistent. As epistemics it is an escape hatch. A model that names the very tests that could embarrass it and then rules them inadmissible has bought its confidence by giving up its exposure. It can no longer be caught being wrong, which is not the same as being right.
This is exactly where this site's posture diverges. The earlier piece treats the same trendline as an empirical fit with a strong R-squared that could still break, and says so in those words. That is the smaller claim and the more honest one, because it leaves the result standing in front of the data with nothing to hide behind. A claim that cannot fail a test has not passed one.
Two Different Power Laws
Underneath the disagreement is a confusion worth naming, because it is the same one the earlier piece turns on. Two distinct things wear the name "power law." One is a heavy-tailed distribution: Pareto, Zipf, the distribution of city sizes and earthquake magnitudes, where extreme events are rare but not vanishingly so. The other is power-law growth in time: a quantity rising as a power of time. They are not the same object, and they do not have the same causes.
The iterative, output-becomes-input processes Santostasi invokes are famous for generating the first kind. Preferential attachment produces power-law degree distributions; that is textbook network science. They are not the standard explanation for the second kind, a smooth deterministic growth curve in calendar time. The theory borrows the hard-won credibility of distributional power laws, which do emerge from network dynamics, and spends it on a growth trajectory, which does not emerge from them in the same way.
The distinction is not pedantry, because bitcoin does contain real, well-behaved power laws, but not the one on the poster. They live in the return distribution. The daily return tail follows a Pareto law with a measured index a little below three, heavier than equities; the autocorrelation of absolute returns decays hyperbolically, the long-memory signature of volatility clustering. The earlier piece fits both, with goodness-of-fit numbers and the data to back them. Those are scale-invariant distributions of the genuine kind: measured, falsifiable, and unglamorous. The price-versus-time curve is the one feature of the system that is not a power law in that sense, and it is the one the theory crowns.
How to Use a Curve That Is Not a Law
None of this makes the trendline useless. A regression with a fifteen-year, multi-order-of-magnitude fit is a serious object even when it is not a law, and using it as a long-horizon anchor is reasonable. That is precisely how this dashboard treats it. Power Law Position, the price's distance from its fitted trend, is one of thirteen indicators, and across the walk-forward machine-learning runs it ranked first more often than any other feature, in 6 of 12 rounds (with the model-size caveats the SHAP analysis documents). It is the single most useful reading on the board.
But it is weighted, not worshipped. It carries 17% of the composite, not 100%, because a fit that could break should size a position rather than anchor a belief. Twelve other indicators are allowed to argue with it, and the backtests are run on the assumption that the trend might one day fail.
That is the whole difference between Santostasi's framing and this one, and it is a difference of kind, not degree. A law claims to know the price in 2032. A tool only locates bitcoin against a trend that has held so far, leaves explicit room for the trend to fail, and refuses to bet the portfolio on a single line. Credit the trendline for being the strongest regularity in the data. Do not mistake it for the law underneath.