The map is not the decision

A map is a useful thing precisely because it leaves almost everything out. It flattens a landscape into a sheet you can hold, throws away the texture and the smell and the slope underfoot, and gives you back a few lines you can act on. The compression is the point; a map the size of the territory would be useless. But the usefulness comes with a quiet condition attached, one we usually forget. The map is trustworthy only as long as the territory it was drawn from still exists, and only as long as you can go back and check it. A map of a coastline that has since washed away is not a map anymore. It is a picture of something that is gone.

Modern decisions are increasingly made from maps. A risk score, a dashboard tile, a single summary figure, a green light or a red one — these are all maps of a decision's underlying basis, dramatic compressions of a great deal of evidence into something a person or a process can act on in a second. And like all maps they are valuable for exactly the reason they are dangerous: they leave almost everything out. The risk score is not the applicant's circumstances; it is a flattening of them. The dashboard tile is not the quarter; it is a projection of the quarter onto a single axis. We know this, abstractly. The trouble is what we do next.

What we do next is act on the map and throw away the territory. The score is computed, the inputs that produced it are discarded or never retained, and from that moment on the score is not a pointer to evidence — it is the evidence, or stands in for it. The compression becomes the original. And a compression you cannot reverse is not a summary of a decision. It is a decision you can no longer examine, wearing the costume of one you can.

A number you can unfold back into its sources is a summary. A number you cannot is a verdict you have agreed in advance never to question.

Reversible compression and the kind that is not

There are two kinds of compression, and the difference between them is the whole subject. The first kind is reversible: the map is small, but the territory is still there, and you can always go back to it. The score is a convenience that summarizes evidence which remains available, in order, with its provenance, ready to be unfolded if anyone asks why. This kind of map is a genuine aid to judgment, because it speeds up the ordinary case without foreclosing the hard one. When a decision drawn from it is questioned, the answer is not the score; it is the territory the score was drawn from, produced on demand.

The second kind is lossy and final. The map is generated, the territory is released, and the compression cannot be undone. Ask why the score is what it is, and the honest answer is that no one can fully say, because the material that would answer the question no longer exists in a form anyone can interrogate. This is the kind of map that masquerades as the first. It looks identical on the surface — a number, a tile, a verdict — and the difference only becomes visible at the exact moment it matters, when someone with standing asks the decision to account for itself and discovers there is nothing underneath to consult.

The confusion of the two is what the old line about the map and the territory was always warning against, and the warning lands harder now than it used to. When a human analyst worked from a summary, the underlying material usually still sat in a file somewhere, recoverable with effort. Automated pipelines made the discarding efficient and invisible. The intermediate evidence is transient by default; the score persists; and so the system arrives, without anyone deciding it should, at a state where the only durable artifact of a decision is the one artifact that cannot be questioned. The map outlives the territory, and we mistake its survival for reliability.

The discipline is keeping the territory

The remedy is not to stop drawing maps. Maps are how decisions get made at scale, and the demand to abolish summaries in favor of raw evidence would simply make every decision unmanageable — which is its own way of making them unaccountable. The discipline is narrower and more achievable: never let the map become the only thing you kept. A score is admissible as the basis of a consequential decision only if it remains an index into something, only if the evidence it compresses was preserved with enough provenance and sequence that the number can be unfolded back into the sources that produced it. The map is allowed to be a map. It is not allowed to be all that survives.

This reframes what a good indicator even is. We tend to evaluate a score by its accuracy — does it predict well, does it correlate with outcomes. Those are real questions, but they are first-reader questions, questions about whether the map is useful to the person holding it now. The second question, the one that decides whether a decision is defensible later, is whether the map is reversible: can it be unfolded, by someone who was not there, back into the territory that justified it. A score that predicts well but cannot be unfolded is a decision you got lucky with, not one you can stand behind, because standing behind it requires showing your work, and the work is gone.

So the test for any indicator that drives a consequential call is a single, awkward question, and it is worth asking out loud before the call is made rather than after it is challenged: if someone asks why this number is what it is, can I unfold it back into the evidence that produced it — or have I kept only the answer and discarded the grounds? If you can unfold it, you have a map of a decision, and the decision is yours to defend. If you cannot, you do not have a map of anything. You have a flattened indicator pointing at a territory you have already let wash away — and you have mistaken it, as we so often do, for the decision itself.

— Dispatches · Summit Cognitive