Fast is not final

There used to be a gap, however small, between deciding something quickly and deciding it for good. A manager could make a snap call and still mean it provisionally — could say yes now and reserve the right to look again, could deny a request in the moment and leave the door open to reconsider once the rush had passed. Speed and finality were two separate properties of a decision, and the separation was useful precisely because it let urgency and care coexist. You could move fast and still hold the outcome loosely, in the knowledge that fast was not the same as fixed.

Automation closes that gap, and it closes it so smoothly that the closing is invisible. An automated decision is produced fast — that is the point of it, the reason it was built — and the speed is wonderful and not the problem. The problem is what travels alongside the speed without being chosen. The same motion that delivers the decision quickly also delivers it as settled. There is no provisional register in the system, no place for a fast-but-tentative outcome, no built-in pause in which the decision sits as a draft before becoming a determination. It is computed and it is done, and the two are the same event.

This is worth slowing down on, because the collapse is easy to wave away as a mere matter of tempo. It is not about tempo. A human deciding fast retains, by default, the capacity to revisit; revisiting is part of how human judgment normally works, and we have to actively suppress it to make a decision truly final. An automated system has the opposite default. Revisiting is not part of how it works unless it has been deliberately built in, and so the decision arrives final unless someone went to the trouble of making it otherwise. The natural state of an automated decision is settled. The natural state of a human one, even a quick one, is open.

Two properties that were never the same

Speed is a genuine virtue. A decision that comes quickly spares the person waiting on it real cost — the loan that funds in time, the claim that pays before the bills are due, the access granted before the moment to use it has passed. Nobody should want to slow good decisions down for its own sake, and a great deal of suffering in older systems came precisely from sluggishness, from decisions that were correct but arrived too late to matter. The case for speed is strong and it should not be conceded grudgingly. It should be affirmed.

Finality is a different thing, and its virtues are real but narrower. A decision has to settle at some point; disputes cannot stay open forever, and there is a kind of cruelty in a process that never lets anyone know where they stand. But finality earns its place only at the end of something — after the look, after the chance to be heard, after the moment in which the decision could still have been wrong out loud. Finality that arrives before any of that is not the closing of a process. It is the skipping of one. And when speed and finality are fused, finality stops being something a decision earns by surviving review and becomes something it is simply born with.

Speed spares the person the wait. Finality without review spares the system the second look. Only one of those is a kindness.

The fusion is dangerous in exact proportion to how reasonable each half seems on its own. We want decisions to be fast. We accept that decisions must eventually be final. Each premise is sound, and the conclusion that fast decisions should be final smuggles itself in between them as if it followed. It does not follow. The speed of reaching a decision tells you nothing about whether it deserves to stand. A decision can be both instantaneous and badly mistaken, and the instantaneousness does nothing to recommend it for permanence. We have let the impressiveness of the speed vouch for the soundness of the result, and those are unrelated facts.

The record reopens the gap

What is missing in the fused version is the pause — the interval in which a decision could be reconsidered before it hardens. You cannot easily reintroduce that pause as a delay, because the delay would forfeit the speed that was the whole point, and rightly nobody wants to. The pause has to be reintroduced in a different dimension: not as time before the decision takes effect, but as the standing possibility of looking back at it afterward with everything needed to judge it. A decision can be delivered fast and still be revisable, if and only if it leaves behind enough to revisit.

This is the work a Decision Receipt does against the collapse of fast into final. By preserving the evidence the decision actually rested on, the rules active at the time, and enough state to replay it, the record keeps the decision available for a second look long after the speed has done its job. The outcome can take effect immediately — the loan funds, the claim pays, the access is granted — and the decision can still be reopened, because the means of reopening it were captured at the moment it was made. The receipt does not slow the decision down. It refuses to let the decision's speed double as its acquittal.

The distinction this restores is between a decision that is settled and a decision that is merely fast. With a record, a fast decision is provisional in the only sense that now matters: it is reviewable on its own terms, against its own real inputs, by anyone with standing to question it. Without a record, a fast decision is final in the worst sense — settled not because it was examined and held, but because nothing was kept that would let anyone examine it at all. Its finality is an accident of having no memory, dressed up as a conclusion.

So the thing to insist on, as more and more decisions move at the speed of computation, is the separation that automation quietly erased. Let decisions be fast; the speed is a gift to the people who wait on them. But do not let the speed be mistaken for a verdict. Keep, at the moment of every fast decision, the record that lets it be looked at again — because a fast decision with a record is a quick judgment that can still be questioned, and a fast decision without one is just an unexamined decision that happened to arrive on time.

— Dispatches · Summit Cognitive