The record is the policy

Every organization carries two policies, and it usually only knows about one of them. The first is the written one — the document with the headings and the version number, the thing that gets approved in a meeting and posted to an internal page where it is read mostly by people who are about to be disciplined under it. The second policy is harder to find, because nobody wrote it down on purpose. It is the policy the organization actually has: the one assembled, decision by decision, out of what it really did when the cases came in. The two are not the same document. They are not even the same kind of thing. And when they disagree, only one of them is binding on the people the organization touched.

The written policy is a statement of intent. It describes what the institution means to do, how it would like to be understood, the principles it would defend if asked. That is not nothing — intent matters, and a clear statement of it is a real form of accountability. But intent is not conduct. A policy document predicts behavior the way a budget predicts spending: it is the plan, not the ledger. And anyone who has compared a budget to a year's actual spending knows which of the two tells you what the organization is.

The real policy is the behavior. It is the pattern that emerges from the decisions themselves — who got approved and who got denied, which exceptions were granted and which were refused, where the edges turned out to be once the abstractions met particular people. You cannot read it off the document, because the document was written before the cases existed. You can only read it off the record of the cases. The decisions are the policy; the document is the press release.

Where the two disagree

Most of the time, no one notices the gap, because no one is looking at both halves at once. The document lives in one place and the decisions live in another — in tickets, in queues, in the outputs of systems that were never built to be read back as a body of conduct. The institution can recite its written policy on demand and has no comparable way to recite what it actually did. So the gap goes unmeasured, which is not the same as it going away.

And the gap is where everything that matters lives. A written rule that says applications are judged on the merits is contradicted, quietly and completely, by a thousand decisions that turned on something the merits never mentioned. A stated commitment to consistency is refuted by two near-identical cases that came out opposite ways for reasons no one recorded. None of this requires bad faith. It requires only that decisions get made faster than anyone reconciles them against the document — which is to say, it requires nothing at all. The drift is the default. Effort is what holds the two together; absent effort, they separate on their own.

When they have separated, the question is which one governs, and the answer is not close. The person who was denied was denied by the decision, not by the document. The pattern that disadvantaged them was the real pattern, whatever the written policy professed. You cannot defend a decision by citing a rule the decision did not follow, and you cannot reassure someone with a commitment your conduct contradicted. The record is what happened to them. The document is what was supposed to.

An institution that cannot see its own decisions does not have a policy it is failing to follow. It has a policy it cannot read, and a document it mistakes for one.

You cannot govern what you cannot read

This is why the gap is not merely an embarrassment to be managed but a failure of self-knowledge. An organization that cannot reconstruct its own decisions — cannot lay them side by side, cannot see the pattern they form, cannot tell where its conduct diverged from its document — does not actually know its own policy. It knows its intentions and mistakes them for its conduct. It is governing by a map it has never checked against the territory, and it will keep being surprised, from the outside, by people who have walked the territory and found it does not match.

The fix is not a better document. You cannot write your way to consistency you are not measuring. The fix is a record built so that the decisions can be read back as what they are: the institution's true policy, available for the institution to inspect before someone else does it for them. This is the quieter half of keeping the minutes and not only the verdict — not so an outsider can contest a single case, though that matters, but so the institution can see, in aggregate, the policy it has actually been running. A Decision Receipt is, among other things, one line of that ledger. Enough of them, read together, are the only honest description of what the organization does.

There is a version of legitimacy that rests entirely on the written policy — on having stated good intentions clearly and obtained the authority to act on them. It is a real thing, but it is the smaller part, and it is shrinking. Authorization is not legitimacy, and a clean document is a form of authorization: a claim to be acting properly. The larger and growing part is the demand that the conduct match the claim — that the record, when read, shows the institution did what it said. An organization that can produce that record is in a position to find its own gaps and close them before they become someone else's grievance. An organization that cannot is left defending a document while the record says something it cannot see and cannot answer for.

So the test of whether an institution knows its own policy is not whether it can hand you the document. Anyone can hand you the document. The test is whether it can hand you the record — the decisions themselves, legible as a body — and whether, when the two are laid against each other, it is the one not surprised by what they show. Where the document and the record agree, the institution is what it says. Where they disagree, it is what it did, and the only question left is whether it knew.

— Dispatches · Summit Cognitive