An Argument About Knowledge

A trained fighter who stops to reason mid-exchange will lose. An investor who trusts the same reflex will, eventually, lose far more. The two activities feel similar from the inside — both reward a kind of practiced instinct — but they sit at opposite ends of a property most people never think to measure. Understanding which end you are standing on is, we think, the first discipline of capital allocation. It is also the least discussed.

We are indebted here to Emanuel Derman, the physicist-turned-quant whose central warning — that a model is not a theory, and that the confusion between them is where the largest errors live — has shaped how we hold every position. What follows builds on his account of how we come to understand the world, and extends it from the question of how we know to the question we actually have to answer each morning: how much should we bet on knowing.

Four modes, and how each one lies

Derman distinguishes several modes through which we apprehend a domain. Four matter for our work.

Intuition is compressed experience — pattern recognition operating below the level of speech. It is fast, holistic, and trained by repetition against feedback. Its failure is silent: it pattern-matches to the wrong situation and feels exactly as certain when it is wrong as when it is right.

Theory claims something true about the structure of the world — a causal account that holds whether or not today’s data cooperates. Real theory is rare, and Derman reserves genuine reverence for it. Its failure is over-extension: a true theory applied just past the edge of the world it describes.

Models are analogies. They say this unfamiliar thing resembles that familiar one, and they are reductive on purpose. Their failure is reification — the moment a user forgets the model is a metaphor and begins to treat its outputs as facts about the world.

Data and statistics describe what has happened without explaining why. Their failure is induction: the regularity holds until the regime that produced it dissolves, and the data carries no warning of its own expiry.

The essential move, the one we return to constantly, is Derman’s: a model is not a theory. Most domains that believe they possess theories in fact possess models wearing a theory’s clothes. Mistaking the second for the first is the costliest error in our field.

The three frictions

If all four modes are always present, the practical question is which one should lead in a given arena — and which one’s confidence should be distrusted precisely because it feels most natural. We find the answer is not a matter of temperament. It is a readable property of the domain itself, legible through three frictions that resist easy knowledge.

The first friction is feedback. How quickly, and how cleanly, does the world tell you that you were wrong? Fast, unambiguous, repeated feedback trains intuition into something trustworthy. Slow, noisy, or rare feedback corrupts it — you accumulate confident judgments that were never once corrected, and you cannot tell them apart from the validated ones.

The second friction is stationarity. Does the structure of the domain hold still, or does it adapt — and worse, does it adapt to you? A stable structure rewards theory and data, because the rules do not move while you study them. A reflexive structure, one composed of other minds reacting to your moves, punishes both: the data you gather is drawn from a regime your own participation is already dissolving.

The third friction is the shape of the stakes. Are errors recoverable, or ruinous? Symmetric, survivable stakes let you lean on instinct and iterate toward competence. Asymmetric stakes — where a single rare outcome can erase the gains of many ordinary ones — demand the opposite reflex: distrust the mode that feels most confident, because the catastrophic case is exactly the region where intuition and fitted models hold the least evidence and the most conviction.

Read these three frictions and the appropriate hierarchy of modes usually reveals itself. The fighter’s domain has fast clean feedback, a stationary structure, and recoverable stakes — which is why instinct rightly governs, and deliberation mid-exchange is a liability. One does not reason one’s way through a domain that has already trained a better reflex.

Why investing is the adversarial case

Markets are the inversion of the fighter’s domain on all three frictions at once, and this is the fact from which everything in our practice follows.

Feedback in markets is slow and treacherous. You may be correct for entirely wrong reasons and be rewarded; you may be correct and wait years to be told so, by which point the lesson is unrecoverable. The structure is reflexive in the deepest sense — any edge that becomes legible is competed away by the very participants who perceive it, so that the act of knowing alters the thing known. And the stakes are not merely asymmetric but fat-tailed: the rare event is not a footnote to the distribution, it is the distribution’s most consequential feature.

A domain with all three frictions at their most severe is one in which no single mode of understanding can be trusted to lead. Intuition is uncalibrated, because the feedback never trained it cleanly. Data extrapolates from a regime that is already expiring. Models, however elegant, are analogies that will fail exactly when it matters most. And genuine theory — durable truth about how value compounds — exists, but covers only a narrow part of the terrain.

This is not cause for paralysis. It is a design specification.

A portfolio is an argument about knowledge

If no mode can be trusted to lead, then a portfolio must not depend on any one of them being right. This single constraint, taken seriously, produces a structure rather than a forecast. We call our governing posture compounders at the core, convexity at the edge, and each half of it is the expression of a different epistemic stance — not a different return target.

The core is the part of the terrain where something close to theory is available. A small number of businesses possess economics so structurally durable — tollbooth positions, compounding advantages that widen rather than erode — that their value does not depend on which regime prevails. Here we are willing to hold with conviction and concentration, because the claim is theory-grade: it should hold whether or not this quarter’s data cooperates. This is the only part of the book where we permit ourselves something like certainty, and even here we hold it as a strong model rather than a proof.

The edge is the explicit admission of everything the first three frictions imply. It is the part of the structure that concedes, in advance, that our models will break, that our intuition is uncalibrated against catastrophe, and that the tail will arrive without announcing itself. Convexity is not a market view. It is a confession — capital arranged so that the failure of our own understanding is survivable, and occasionally profitable. One does not need to predict the rare event to be positioned for it; one needs only the humility to assume it is unpredictable.

Between the two sits the single thing intuition is genuinely suited for in a low-feedback domain: not prediction, but the recognition that a situation has left the region where the current model applies. We call this discipline Kairos — the Greek word for the opportune moment, the qualitative right time as distinct from mere chronological time. Its task is not to forecast the next regime but to notice, as early as the evidence allows, that the prevailing one is changing — to convert a vague unease into a legible signal, and so to discipline the one instinct worth keeping into something that can be examined rather than merely felt.

The result is a structure that does not require us to be right about the future. The core compounds through the ordinary regimes that occupy most of market time. The edge protects, and converts, the rare ones. And Kairos governs the weight between them, not by prophecy, but by reading the same three frictions — adjusting posture as the domain reveals which mode it is, for now, rewarding. A portfolio built this way is suited to all regimes precisely because it does not bet on any single one. It is sized for the limits of its own knowledge.

A closing humility

We are aware of the irony, and we hold it deliberately. Everything above is itself a model — a useful, reductive analogy, wrong at its own edges, valid only within a domain we have tried to mark honestly. It is emphatically not a theory. Were we to forget that, to mistake this framework for a law rather than a lens, we would commit the precise error it was built to guard against.

So we offer it as we hold our positions: with conviction proportioned to what the evidence has earned, and not one degree more. The discipline is not certainty. It is calibration — confidence matched, in every domain, to what that domain has actually given us the right to believe. In markets, that right is hard-won and easily overdrawn. A portfolio is how we keep the account honest.

— Shash Hegde