The provenance of a policy

We have learned to ask where evidence comes from. A figure that turns up in a decision invites the question of who produced it, from what source, by what method, and whether anything happened to it on the way. We call this provenance, and we treat it as a precondition of trust: a number with no history is a number we are right to doubt. But there is a second object in every decision whose origin we almost never interrogate, and it is the more powerful of the two. It is the rule. We demand the provenance of the evidence and accept the policy that judged it as though it had no history at all — as though the rule were simply true, rather than something someone wrote, for a reason, at a time, and has perhaps quietly changed since.

A policy is not a fact of nature. It was authored. Someone decided that this threshold, this criterion, this cutoff would govern a class of decisions, and they decided it in a particular context, toward a particular end, under particular assumptions. All of that is the provenance of the policy: where it came from, who set it, what it was meant to achieve, and when it last changed. And almost none of it is typically recorded in a way that survives to the moment the policy is applied. The rule arrives at the decision stripped of its own history, presenting itself as given — which is precisely the condition under which no one can be held to it, because there is nothing behind it to hold anyone to.

What a rule with no history costs you

The absence of policy provenance produces three failures, and each of them is a way of not being able to answer a question that matters. The first is that you cannot tell whether a decision actually followed the rule, because you cannot be sure which version of the rule was in force when the decision was made. A policy that has been amended since leaves the past decision suspended between two texts, and without a record of which one governed, the question did this decision comply has no determinate answer. The second is that you cannot detect drift. A policy that has shifted gradually — a threshold nudged, a criterion reinterpreted, an exception widened — will look, at any given moment, exactly like a policy that has held firm. Only its history reveals the movement, and a policy with no recorded history has no movement to reveal; it has simply always been whatever it now is.

A rule that cannot account for its own origin governs by inertia. It is not obeyed because it is right. It is obeyed because no one can remember it being anything else.

The third failure is the deepest. Without provenance, you cannot ask whether the policy still serves the purpose it was written for. Every rule was written to achieve something — to prevent a harm, to allocate a scarce thing fairly, to encode a judgment about what matters. Circumstances change, and a rule that once served its purpose can come to defeat it while remaining word-for-word the same. But you can only ask whether a rule still fits its purpose if the purpose was recorded alongside the rule. A policy that carries only its text, and not the intent behind the text, cannot be tested against that intent. It can only be applied, and applied, until the reason it exists has been forgotten and the rule persists as a habit that no one can either justify or repeal.

Provenance is part of legitimacy

The claim I want to make is that the provenance of a policy is not administrative housekeeping but a component of the policy's legitimacy. A rule that is merely in force — enforced, applied, obeyed — is not the same as a rule that is legitimate. Legitimacy is the difference between a rule you must follow and a rule you can be justly held to, and that difference rests on whether the rule can account for itself. A legitimate rule can say where it came from, who set it, toward what end, and how it has changed. A rule that can say none of those things exerts power without warrant. It may be obeyed, but obedience extracted from a rule that cannot explain its own existence is not consent to the rule; it is merely submission to whatever the rule happens currently to say.

This reframes what a policy record is for. We tend to think of documenting a rule as a matter of publishing it — making its current text available so people know what it says. But the current text is the least of it. The record that matters is the one that carries the rule's genealogy: the successive versions, the dates of change, the author of each change, and the rationale that accompanied it. Such a record turns a policy from a standing assertion into an accountable object. It lets a past decision be checked against the rule that actually governed it. It makes drift visible as drift. And it keeps the purpose attached to the text, so the rule can be measured against what it was for, not merely against what it now is.

Reference the standard from Admissible Reality: a decision is admissible only if it can be traced to a rule that was itself knowable and stable at the time it was applied. That standard is empty if the rule has no provenance, because a rule with no recorded history is neither knowable in its past form nor demonstrably stable — it is whatever the present says it always was. The provenance of the evidence tells you whether the facts can be trusted. The provenance of the policy tells you whether the rule that judged them can be. We have spent our scrutiny almost entirely on the first and almost none on the second, and the second is where the power actually sits. A policy has a history. Keeping that history is not bureaucracy. It is the part of a rule that makes it legitimate rather than merely current.

— Dispatches · Summit Cognitive