The appeal that goes in a circle

The letter denying you says, near the bottom, in a reassuring tone, that you have the right to appeal. You take it up. You fill in the form, you attach what you were told to attach, you press submit, and you wait. In due course a second letter arrives. It is worded a little differently. It thanks you for your submission, notes that your case has been reviewed, and informs you that the original decision has been upheld. What it does not tell you — what you have no way, from the outside, to know — is that the review was performed by the same model that made the first decision, reading the same inputs, following the same logic, and that it could not, structurally, have arrived anywhere else. You did not get a second look. You got the first look, run again.

This is a distinct and particular failure, and it is worth separating from its more famous cousin, the appeal you cannot file. That one at least announces itself: there is no form, no channel, no door, and the impossibility is out in the open where you can name it. The circular appeal is worse in a way that is hard to say out loud, because its futility is disguised as its opposite. There is a form. There is a channel. You can exercise the right. And exercising it accomplishes nothing, because the mechanism you have been invited to use was never capable of disagreeing with the outcome it was asked to review. It has the full outward form of an appeal and none of the function.

The machine that agrees with itself

Consider what actually happens when a decision system reviews its own output. The original decision was a function: certain inputs went in, a certain logic ran over them, a conclusion came out. Now the same inputs are fed back through the same function. There is no version of this that produces a different result. A deterministic process, given identical inputs, returns an identical output — that is what it means for it to be a process at all. So the appeal, insofar as it is the same system reconsidering the same case on the same evidence, is not a reconsideration. It is a repetition wearing the costume of one. It cannot find that the first decision was wrong, because "wrong" would require a standpoint outside the logic that produced it, and there is no such standpoint in the loop.

What makes this maddening from the receiving end is the tone. The second letter does not read as a repetition. It reads as a judgment — considered, independent, arrived at freshly. The serenity is the tell. A human reviewer who genuinely re-examines a hard case carries some trace of the difficulty into the reply: a caveat, a note of the close call, an acknowledgment of what cut the other way. A machine agreeing with itself has no such friction, because for it there was no second question — only the first one, asked twice, answered the same. The confidence you are reading is not the confidence of a check that was passed. It is the absence of a check having occurred.

An appeal requires a different mind

The whole point of an appeal — the thing that makes it an appeal and not an echo — is independence. To appeal is to ask that a decision be examined by something that could reach a different conclusion. The possibility of disagreement is not an incidental feature of the process; it is the entire substance of it. Remove it and what remains is not a weaker appeal. It is not an appeal. A review that cannot overturn is not a lenient check or an imperfect one; it is a confirmation with better manners.

Independence can be supplied in more than one way, but it always requires that something in the second pass be genuinely other than the first. A different decider — a person, or a system, that did not produce the original outcome and is not bound to reproduce it. Or new evidence — facts the first pass never weighed, which give even the same decider something honestly new to consider. Or a fresh human judgment, applied with the authority to set the machine's output aside. What none of these can be is the original function, re-executed on the original inputs, because that combination has already fixed the answer before the appeal begins. Self-review on unchanged inputs is not a diluted form of independence. It is its absence.

An appeal that runs your case back through the machine that decided it is not a second opinion; it is the first one, read aloud again.

And here is the specific harm, the thing that makes the circular appeal more corrosive than the honest dead end. It does not merely fail to help you. It launders. An unreviewed decision goes in, and a decision stamped "reviewed" comes out — carrying, now, the authority of a process it never actually underwent. The next person to touch your case, the auditor, the regulator, the court, sees that you appealed and were denied, and reasonably infers that a second, independent look confirmed the first. The record says the box was checked. The box was checked. What the record cannot show is that the check was performed by the very thing it was meant to check. Futility disguised as due process does not just waste your effort; it converts your unreviewed decision into a falsely-reviewed one, and hands that false review to everyone downstream as if it were the real thing.

What a real appeal owes the person

A real appeal owes you, at minimum, genuine independence — a decider that is not the original, and is not compelled to agree with it. In most cases that means a human with actual authority: not a person who reads the machine's output back to you, but one empowered to set it aside, looking with different eyes at a case they did not decide. Where the second pass is still automated, independence has to enter some other way — new evidence the first decision never saw, weighed under a standard the first pass did not apply. The one arrangement that cannot count, however it is dressed, is the same logic re-running the same inputs and calling the loop a review.

This is also why the record matters so much here, and why a Decision Receipt — the inputs a decision actually rested on, the rules that were active, enough state to replay it — is the difference between an appeal you can trust and one you have to take on faith. A reviewer who can see what the first decision was really built on can re-decide: weigh the evidence again, notice what was missed, reach a different conclusion on the merits. A system that can only re-run cannot do any of that; it can only reproduce. The distinction is exactly the one between a decision that was reconsidered and one that was merely re-executed. From the outside, in the calm language of the upholding letter, they look the same. They are not remotely the same, and the person on the receiving end is the one who pays for the difference — twice, and told each time that the matter has been reviewed.

So when a system offers you the right to appeal, the question worth asking is not whether the channel exists. It plainly does; you are holding the form. The question is whether anything on the far side of it can disagree with the machine that decided against you — a different mind, new evidence, a human who can say no to the model and mean it. If the answer is yes, you have an appeal. If the answer is that your case will be run through the same logic again, you have a circle, and the second letter is already written.

— Dispatches · Summit Cognitive