Irreversibility raises the bar

We tend to talk about the standard of evidence as if it were a fixed property of a decision — as though a given call simply requires a certain amount of proof, and the job is to assemble it. But the right amount of proof is not fixed. It moves, and it moves with a variable we rarely name out loud: how hard the decision will be to undo. A choice you can reverse cheaply and a choice you can never take back are not the same kind of choice, even when they look identical on the page, and they do not deserve the same evidentiary bar before you act on them.

The reason is simple, and it has nothing to do with caution as a temperament. A reversible decision has a second chance built into it. If the evidence turns out to be thin or wrong, reality corrects you, and you correct the decision — the error is real but bounded, paid off the moment you notice. An irreversible decision has no such mechanism. When the evidence turns out to be wrong, there is no contact with reality that buys it back, because the thing is already done. The cost of being wrong is not refunded over time; it is fixed at the moment of action and carried forever. That asymmetry is the whole argument. Where you can be corrected, you can afford to be uncertain. Where you cannot be corrected, you cannot.

Two decisions, two standards

Hold two decisions side by side. The first flags an account for a brief review — a soft hold, lifted the moment a person looks and sees nothing wrong. The second closes the account permanently, severs the relationship, and deletes the history. The inputs feeding both might be the same signals, the same model, the same threshold. But the appropriate bar is not the same, because the consequences of error are not symmetric. A wrongful soft hold costs minutes and an apology. A wrongful permanent closure costs something that may never be recovered, to someone who may have no way to make themselves heard. To demand the same evidence for both is to misunderstand what evidence is for. Evidence is not a tax levied evenly on all decisions; it is insurance priced against the harm of getting it wrong, and the premium should rise with the stakes.

This reframes a familiar instinct. The conventional wisdom says: gather more evidence for important decisions. True, but imprecise. Importance is not quite the axis. A consequential decision that can be unwound the next morning can tolerate genuine uncertainty, because the unwinding is the backstop. A minor-seeming decision that happens to be permanent cannot, because there is no backstop at all. The axis that actually governs the bar is not how much the decision matters but how completely you are committed to it the instant you act — whether the door behind you stays open or locks.

Where a decision can be corrected, thin evidence is a risk you can afford. Where it cannot, thin evidence is not a risk at all. It is a verdict you have already passed and called permanent.

The record has to exist beforehand

Here is the part that makes this more than a maxim about prudence. For a reversible decision, the evidentiary record can afford to be reconstructed after the fact, because if anyone objects, the remedy is to look again and undo it — the record is a convenience, and the reversal is the real protection. For an irreversible decision, there is no looking again that helps the person harmed. The only protection they will ever get is the protection that existed before the action was taken. Which means the record cannot be something assembled later to explain the decision. It has to be something that existed at the moment of decision and was strong enough, then, to have justified it.

This is a real constraint, and a demanding one. It says that for the decisions you cannot take back, you owe a contemporaneous account — the evidence that was actually in front of you, the rules that were actually in force, captured at the time and preserved against later revision — because there will be no second decision in which to get it right, and no reversal in which to make the harmed party whole. The justification has to have been there in advance, or it was never there at all. A Decision Receipt that captures the basis of an irreversible call at the instant it is made is not paperwork. It is the only thing standing in for the second chance the decision does not have.

None of this argues for paralysis, and it does not argue that irreversible decisions should never be made quickly. Some must be. What it argues is that the standard should track the stakes honestly, and that the honest way to lower the bar is to lower the irreversibility — to build a way back in. A decision you have made reversible is a decision you have earned the right to make on less. This is why reversibility is worth engineering for its own sake: every door you keep open is evidence you no longer have to gather in advance, an error you are now allowed to make and fix rather than forbidden to make at all. The bar and the back door move together. Raise one only when you have closed the other, and know, when you close it, exactly what you have taken on.

So before acting, it is worth asking the two questions in the right order. Not first how confident am I, but first can this be undone — and only then, in light of that answer, how confident do I need to be, and is the record I am standing on strong enough to have justified this in advance? When the door stays open, you can move on thin evidence and let reality teach you. When it locks behind you, the evidence is all you will ever have, and it had better have been enough before you turned the key.

— Dispatches · Summit Cognitive