The cost of being right too late

There is a comfortable fiction that being right is a matter of substance and being on time is a matter of logistics — two separate concerns, one important and one merely operational. The fiction is comforting because it lets an institution defer the second in favor of the first. We will get you the correct answer, it says; we just need a little more time. But for a great many decisions, the two are not separable at all. The rightness and the timing are the same property, and an answer that is correct but late is not a slightly worse version of the same answer. It is a different object entirely, and often a worthless one.

Consider the account that is suspended pending review. The review, six weeks on, concludes correctly that nothing was wrong — that the flag was a false alarm, that the person is exactly who they said they were, that the transaction was legitimate. This is the right answer. It is also, for the person who could not make rent in week two, an answer that arrived after the thing it was supposed to protect had already been lost. The institution congratulates itself on getting it right. The person experiences having been gotten wrong for six weeks and then, belatedly, released. The correctness is real. The timeliness was the part that mattered, and it was never delivered.

This is why I want to insist that timeliness is not a service level. It is a property of accountability itself — of whether a decision can be answered for at the moment the answer is still capable of doing any good.

A verdict that arrives after the sentence has been served is not justice arriving late. It is justice that never arrived, wearing the costume of the thing.

Why the window is part of the decision

Every consequential decision has a window — a span of time during which an intervention can still change the outcome, after which the outcome hardens into fact. Inside the window, a correction repairs. Outside it, a correction can only apologize. The window is not a soft preference; it is a hard feature of the situation. And crucially, the window belongs to the affected party, not to the institution processing the case. The institution feels time as a queue depth. The person feels it as a countdown.

The reason records matter here is that the window is exactly the thing institutions forget to preserve. A decision made under time pressure, with real stakes, produces its evidence and its reasoning in a particular moment — and that moment is where a Decision Receipt earns its keep. The receipt does not merely record what was decided; it records when, against which rules as they stood at that instant, and on which evidence as it was then available. A record built after the fact, reconstructed weeks later from memory and logs, can be accurate about substance and still be silent about the one dimension that determined whether the correction could arrive in time. Provenance includes provenance in time. A record that cannot tell you when the window opened cannot tell you whether it was met.

There is a related failure that hides inside slow correctness: the delay is frequently the institution buying itself cover. Taking six weeks to conclude nothing was wrong is not always diligence. Sometimes it is an absence of a record clean enough to reach the conclusion faster — the answer was always available in the evidence, but no one could assemble it quickly because the decision left no traceable account of itself. The delay, in other words, is the symptom of the missing receipt, not a separate problem. Speed of correction and quality of record are the same variable seen from two angles.

Building for the window, not the verdict

If timeliness is a property of accountability, then a system that wants to be accountable has to be built to answer inside the window, which is a stricter requirement than being built to answer correctly. It means the record has to be assembled at the moment of decision and available on demand, not reconstructed on request. It means the party who can be harmed by delay must have standing to trigger the account, not merely to wait for it. And it means replay — the ability to rerun the inputs and show what the decision actually did — has to be a fast operation, because a replay that takes six weeks to stand up is a replay that arrives after the window it was meant to serve.

None of this lowers the bar on correctness. The point is not to trade accuracy for speed; it is to notice that for decisions with real windows, a late answer is already an inaccurate one — inaccurate about the thing that counted. The institution that measures itself only on whether it eventually reached the right conclusion is grading itself on the wrong exam. The exam is: did the record make it possible to reach the right conclusion while it still helped?

So the question to ask of any system that touches consequential decisions is not only will you get this right but can you show me you got it right before it stops mattering. If the honest answer is that correctness is available but only after the window closes, then the system is not accountable. It is merely eventually correct, which is a private virtue the affected party is not invited to share. Being right too late is a way of being wrong that flatters itself as rigor. It should be named as what it is.

— Dispatches · Summit Cognitive