The number and the judgment

A score is a remarkable thing to receive. It is small, definite, and self-contained. It does not argue or qualify; it simply states. Seventy-three. Above threshold. High risk. It has the texture of a measurement, the same clean finality as a temperature or a weight, and that texture is doing an enormous amount of quiet work, because a score is not a measurement. A measurement reads a quantity that exists in the world. A score manufactures a quantity by compressing a long sequence of human choices into a single figure — and then presents the figure with all the authority of the temperature and none of the visible seams of the choices. The number looks like something it found. It is something someone built.

Walk back up the chain and the choices are everywhere. Someone decided what to measure — which features count, which are ignored, what stands in as a proxy for the thing actually cared about, since the thing actually cared about is usually not directly measurable at all. Someone decided how to weight what was measured — what matters twice as much as what, what gets folded together, what dominates and what merely nudges. And someone decided where to draw the line — the threshold above which the score means one thing and below which it means another, a cut point that is almost never given by nature and almost always chosen to balance one kind of error against another. Each of these is a judgment. Each could have gone differently. And every one of them is invisible in the figure that results.

Compression hides exactly what could be contested

This is the property of the number that deserves the most attention: the compression does not hide its contents at random. It hides precisely the parts that someone might want to argue with. The raw fact that an applicant has a certain attribute is not, by itself, controversial. What is controversial is the decision to treat that attribute as relevant, to weight it as heavily as it was weighted, and to place the threshold where it sits relative to it. Those are the contestable choices — and those are exactly the choices that the single figure dissolves. What survives the compression is the verdict. What is lost is everything you would need to dispute it.

So the number arrives stripped of its own argument. It does not say "we chose to treat this as a proxy, weight it thus, and cut here, and a reasonable person might have chosen otherwise." It says seventy-three. The judgment calls have been laundered into a quantity, and a quantity does not invite disagreement the way a judgment does. You can argue with a claim that someone made a choice. You cannot easily argue with what looks like the readout of an instrument — and the score is engineered, whether deliberately or not, to look like exactly that. Its persuasive power comes from concealing that it is the output of contestable decisions rather than the reading of an external fact.

A judgment invites you to disagree with it. A number, made of the same judgments, dares you to argue with arithmetic.

And once produced, the number travels with a freedom that no judgment ever enjoys. It moves between systems, gets copied into other people's decisions, accumulates an authority that grows with each hop precisely because each new reader is further from the choices that made it. By the third or fourth use, the score is treated as a property of the person it describes rather than as the residue of a modeling decision someone made on a particular afternoon under particular assumptions. It has become, functionally, a fact — not because anything about it changed, but because the judgments inside it were never visible to begin with and the distance from them keeps increasing.

To contest the number, decompress it

Which means that contesting a score is not, fundamentally, an argument about the score. It is an act of decompression. To dispute the seventy-three, you cannot stay at the level of the seventy-three; there is nothing there to grab. You have to run the compression backward — to recover what was measured, how it was weighted, and where the line was drawn — and conduct the argument at the level of those buried choices, because that is the only level at which there is anything to argue about. The number is not the claim. The number is a sealed container for a stack of claims, and the dispute lives inside the container.

This is why a score offered without its decomposition is not really evidence, however authoritative it looks. Evidence is something you can examine and answer. A bare figure forecloses examination by hiding the very judgments that examination would target. A score becomes contestable only when it carries its own decomposition — when the record around it preserves what was measured, the weights applied, the threshold chosen, and the inputs that produced this particular figure for this particular person. That decomposition is what gives a person standing to dispute the result rather than merely receive it. Without it, the affected party is handed a verdict and denied the only thing that would let them contest it: the judgments the verdict was made of.

None of this is an argument against scores. Compression is useful; that is why we do it. A single figure that summarizes many factors is often exactly the instrument a decision needs, and demanding that every number be carried around with its full derivation at all times would be its own kind of paralysis. The argument is narrower. It is that the moment a score is used to do something to a person, the compression has to be reversible — the judgments have to be recoverable, on demand, by the person they were used against. A number you can decompress is a number you can contest. A number you cannot is not a fact, whatever it looks like. It is a judgment that has stopped admitting it is one.

— Dispatches · Summit Cognitive