“It is the mark of an educated mind to rest satisfied with the degree of precision which the nature of the subject admits.”

— Aristotle

A leader cuts the price on the flagship product, and over the next quarter revenue climbs. At the review, the story writes itself: the price cut worked. But pause on that word, worked, because it hides an impossible claim. To know the cut worked, you would have to compare this world to the one where you left the price alone and everything else played out the same. That second world is the only real test of the claim. And it does not exist. The moment we pull the lever, we destroy the world where we did not. So the question buried under every confident arrow on every slide is the one no one asks: compared to what?

That question is the whole game, because a cause is not a thing that happened. It is a subtraction: what happened, minus what would have happened without you. The first half you can see. The second half is a ghost. We can never watch the same customer both get the discount and not get it, so every causal claim we make is a claim about a parallel world we cannot visit. And that changes what kind of thing a causal arrow can be. It can never be proven the way a theorem is proven, true by definition, true everywhere. It can only be earned the way a verdict is earned in a court, by ruling out the rival explanations until yours is the one left standing. The arrow is a verdict, not a theorem.

The trouble is that a causal claim can wear the costume of proof while owning none of its substance. A leader puts up a clean equation, revenue equals price times volume, or a model stamped proven, and the room relaxes, because it looks like geometry. But geometry lives in a closed world, where the axioms hold and nothing outside can intrude. Our decisions live in an open one, where competitors react, tastes shift, supply chains break, and the other things that were meant to stay equal never do. The equation is closed. The world it describes is not.

There is a quieter version of the same mistake: confusing prediction with causation. Prediction asks, if I see X, can I expect Y? Causation asks, if I change X, will Y move? A signal can answer the first perfectly and the second not at all. Ice-cream sales predict drownings beautifully. Ban the ice cream and no one is saved.

Dressing a bet as a theorem is dangerous exactly where it matters most. A verdict can be revised when new evidence lands. A theorem cannot bend, it can only break, and it breaks at scale, after the policy is set and the capital is spent, when the unstated conditions quietly shift. Worse, calling it proof shuts down the questions that would have caught its limits, because nobody interrogates a theorem. The most expensive errors on record were not honest mistakes. They were arrows drawn in a vacuum and defended as proof, so the organisation learned where the arrow failed the only way left to it: by crossing the line, at full cost.

So change the question. Stop asking whether the arrow is true, because in an open world that answer never arrives. Ask instead what would entitle you to draw it at all, and what it would take to convict it beyond reasonable doubt.

The fear is obvious: if no arrow is ever proven, a rival can always demand one more study, and nothing ever closes. The opposite is true. Warrant is exactly what lets a decision close. Certainty cannot be the bar, because in an open world it never comes, so a room that waits for proof waits forever. A criminal court shows the way out. It does not demand mathematical proof that the defendant pulled the trigger. It asks whether guilt explains the evidence better than any rival story, whether those rivals were tested, and then it reaches a verdict and acts. That is not paralysis, and it is not recklessness. It is the only standard that lets us commit hard while staying honest about what we know.

Convicting an arrow takes three plain lines, stated before it is allowed onto the slide. The first is the counterfactual: if we had not done this, what would have happened instead? Name the world you did not get, even roughly, a before-and-after against a comparable group that did not change, a matched case. You are not trying to show you acted. You are trying to tell acting apart from causing. The second is the boundary: this holds under these conditions and may fail outside them. That one sentence turns a pretend universal law into an honest scoped claim, and it tells you in advance where to watch for the collapse. The third is the breaker: name the strongest rival explanation, and say what you would see if that rival, not your arrow, were driving the result, the number that should move first, the pattern that should follow, the signal that should be there and is not. A claim that cannot say how it differs from its best competitor is not a finding. It is a preference with a chart.

We cannot rewind the universe to check our work. The world where we did nothing is gone the instant we act, and no chart brings it back. So when you cannot describe the road not taken, do not draw the arrow. Write correlated with, and hold it there. Act with the humility of a gambler who knows the odds can turn, not the false certainty of an engineer whose bridge was never load-tested. A causal claim is never proven. It is convicted against its rivals, or it is only a story with good production values.

Decision Rule — The Conviction Test

Before a causal arrow is allowed to steer a decision, make it stand trial, because it can be convicted but never proven:

• The counterfactual. If you had not acted, what would have happened instead? If you cannot describe that missing world, you have a correlation, not a cause.
• The boundary. Under what conditions does this hold, and where would it fail? A claim that cannot name its own limits is a story posing as a law.
• The breaker. What is the strongest rival explanation, and what would you see if it, not your arrow, were driving the result?
• An arrow that survives all three is warranted, not certain, and warranted is enough to act on. One that survives none is a preference with a chart.

Convict the arrow against its rivals. Never mistake the verdict for a theorem.