The appeal nobody can file

Almost every automated decision comes with the promise of an appeal. There is a line at the bottom of the notice: if you believe this decision was made in error, you may request a review. The line is important to the institution. It is the sentence that lets everyone involved feel the process was fair, that the affected party was not simply told no by a machine but offered recourse. And in a great many systems that sentence is a fiction — not because the institution lies about it, but because the appeal it promises cannot, in any practical sense, be filed. The door is painted on the wall. You can walk toward it. You cannot walk through it.

An appeal that cannot be filed is worse than no appeal at all, and it is worth being precise about why. No appeal is at least honest; it tells the affected party plainly that the decision is final and they are on their own. A phantom appeal does something more corrosive. It manufactures the appearance of recourse, which discharges the institution's felt obligation and disarms the affected party's sense that they are being wronged. They think they have a remedy. They spend their energy pursuing it. The remedy was never real, and its unreality is the thing they discover last, after the window has closed.

The distinction I want to draw is between a verdict and a process. A verdict is a conclusion imposed. A process is a conclusion reached in a way that could have gone otherwise had the affected party shown it should. The presence of the word appeal does not make something a process. Only the existence of a real path from objection to reversal does.

An appeal you cannot actually file is not a safeguard against error. It is a way of laundering a verdict into the appearance of a fair process, and it fails exactly the people it names.

How an appeal becomes unfileable

There are a few reliable ways to make an appeal impossible while keeping the promise of one on paper. The first is to withhold the reasons. You cannot contest a decision whose basis you cannot see; an appeal against an unstated reason is a swing in the dark. The affected party is told they may appeal but not told what they are appealing — what evidence was used, which rule was applied, why the answer came out as it did. The appeal collapses because there is nothing specific to grab. This is why contestability is downstream of the record: no legible account of the decision, no fileable appeal.

The second is to route the appeal to the same process that produced the decision, with the same inputs and the same rules, so that the review is guaranteed to reproduce the original result. This is a review in form and a rubber stamp in substance. For an appeal to be real, the affected party has to be able to introduce something the first decision did not have — a correction to the evidence, a contest of the rule, a demonstration that the inputs were wrong. If the appeal cannot take on new argument, it is not a second look. It is the first look, run again, to the same end.

The third, and most modern, is that no one on the institution's side can actually reconstruct what happened. The decision was automated, the reasoning discarded, and the person handling the appeal has no more access to why the machine decided as it did than the appellant does. The appeal fails not from bad faith but from amnesia: there is nothing to review because nothing was kept. The reviewer and the appealed-against are equally in the dark, staring at the same bare output, and the review becomes a negotiation between two parties who both lack the record.

What a fileable appeal requires

For an appeal to be real, three things have to be true, and each is a property of the record before it is a property of the procedure. The affected party must be able to see the reasons — the evidence consulted and the rule applied, in a form they can read. They must be able to introduce a contest — new evidence, a corrected input, a challenge to the rule — that the review is capable of taking on. And the review must be able to replay: to feed the corrected inputs back through the same process and show whether the conclusion still follows. A Decision Receipt is what makes each of these possible, because it carries the evidence, the frozen rules, and enough state to rerun. Without it, an appeal has nothing to appeal against and nothing to appeal with.

Notice that this makes contestability a design property, not a customer-service function. You cannot bolt a real appeal onto a decision that kept no record of itself; the appeal has no surface to work on. The institution that wants genuine recourse has to build the record at the moment of decision, precisely so that later — when someone objects — there is something to reopen. The appeal is only as fileable as the decision was recordable.

So when you are handed a decision and pointed toward an appeal, test the door before you trust it. Ask what reasons you will be shown, what new argument you are permitted to bring, and whether the review can actually rerun the decision with your correction in it. If the answers are none, none, and no, you do not have an appeal. You have a verdict, and a sentence at the bottom of the notice designed to make it feel like something else. The right to be wrong about you means nothing unless someone can be made to check — and checking requires a record built to be reopened.

— Dispatches · Summit Cognitive