The reason you cannot act on

The application was declined. This time there is even a reason on the page, because someone somewhere required that a reason be given: the decision was based on factors including account history, aggregate risk indicators, and information from third-party sources. You read it twice. It is grammatical. It is, in some formal sense, a reason. And it is completely useless to you, because there is nothing in it you can hold, correct, or push against. You cannot act on account history you cannot see. You cannot dispute aggregate indicators that name no fact about you in particular. You have been told why, and you are exactly as stuck as if you had been told nothing at all.

This is the peculiar cruelty of the reason you cannot act on, and it is worth naming precisely because it hides behind the appearance of a remedy. Somewhere upstream, a rule said provide an explanation, and the system complied — it produced a sentence. The box was checked. From the desk that produced it, the obligation looks discharged: a reason was given, the record will show a reason was given, and an auditor scanning for the presence of an explanation will find one. But you are not an auditor scanning for presence. You are the person the decision landed on, and from your chair the only question that matters is the one the explanation does not answer: what do I do now?

Two things wear the same word

We use one word, reason, for two entirely different objects, and the confusion is load-bearing. One kind of reason is a justification — an account offered upward, to a regulator or a court or an internal reviewer, meant to show the decision was defensible. The other is a handle — an account offered to the affected party, meant to give them purchase on their own situation. A justification asks to be accepted. A handle asks to be used. They can be written in the same sentence, but they do different work, and a decision system optimized to produce the first will reliably fail to produce the second, because nothing forced it to.

The tell is always the same. Ask of any stated reason: if this reason is wrong, what could the affected party do about it? A handle survives the question. "Your reported income fell below the threshold for this product; the figure we used was $X, drawn from the document you submitted on this date" — if that figure is wrong, you know exactly what to challenge and with what. A justification dissolves under the question. "Aggregate risk indicators" names nothing you could ever prove false, because it names nothing in particular at all. It is unfalsifiable by construction, and an unfalsifiable reason is not a reason offered to you. It is a reason offered about you, to someone else, in your hearing.

A reason you cannot act on is not half an explanation. It is a full explanation aimed at the wrong person — written to satisfy a reviewer, and handed to you because you were standing there.

The dignity is in the next move

There is a common assumption that what the affected party wants is understanding — that the injury of the opaque decision is epistemic, a hunger to know. That is part of it, but it undersells the thing. Mostly, the person on the receiving end does not want to contemplate the decision. They want to respond to it: to correct the wrong input, to supply the missing document, to contest the false inference, to bring the mistaken record to someone who can fix it. The reason matters because it is the map to the next move. When the reason is a handle, the person is restored to the position of an agent in their own case. When it is a justification, they are left a spectator to a decision about their life, holding a sentence that explains everything and permits nothing.

This is why "we told them why" is not the standard, though it is constantly mistaken for it. A system can tell people why all day and leave every one of them helpless. The standard is whether what you told them connects to a remedy they can actually reach — whether the reason names a specific input they could challenge, points to a specific record they could correct, or identifies a specific door they could knock on. An explanation that terminates in itself, that describes the past and opens nothing toward the future, has performed the ritual of accountability without any of its substance.

What the account owes the receiving end

The remedy is not more words; it is words of a particular shape. An account owed to the affected party has to be built from the specific inputs the decision actually rested on — the figures, the records, the sources, named concretely enough that a person could point at one and say that one is wrong, and here is why. It has to distinguish the parts of the decision the person can move from the parts they cannot, so they are not left grinding against an immovable object without knowing it is immovable. And it has to end in a door: an office, a channel, a person who can receive the correction the account made it possible to formulate. A reason with no door behind it is a reason built to be admired, not used.

None of this asks the system to be more lenient, or even to be right more often. A system can be exactly as strict as before and owe exactly the same denials. What changes is that each denial arrives as something the person can work with — a specific claim about them, sourced to specific evidence, attached to a specific way to contest it. That is the difference between telling someone their situation and handing them the means to change it. The first is a courtesy the system extends to itself, so its record will read well. The second is the thing the person on the receiving end was actually owed.

The next time a decision lands on you with a reason attached, run the one test that separates the two: if this were wrong, what could I do? If the honest answer is nothing, you did not receive an account. You received a receipt for one — proof that somewhere, someone was told why, and it simply was not you.

— Dispatches · Summit Cognitive · The Receiving End