What is Computable Authority?

Every institution runs on authority: the accepted right of some party to decide something on behalf of others. A ministry may license. A hospital may admit. A bank may refuse. Authority is not information — it is permission bound to conditions, and it is the thing that makes a decision legitimate rather than merely correct.

For as long as institutions have existed, that authority has been carried by documents and interpreted by people. The document is not the authority; it is a description of it. A human reads the description, forms an interpretation, and acts. The interpretation step is invisible, unlogged and unverifiable — and until recently it was also unavoidable.

The definition

Computable authority is institutional permission expressed in a structure that a machine can evaluate deterministically, bound to a version, and verifiable by a party that did not execute it.

Three properties do the work in that sentence. Deterministic: the same inputs against the same artifact always produce the same outcome. Versioned: the artifact that governed a decision can be retrieved unchanged, years later. Independently verifiable: a regulator, auditor or counterparty can confirm the outcome without trusting the operator's word or re-reading the source text.

What it is not

• It is not automation. Automation executes a process; computable authority defines what a process is permitted to do.
• It is not a policy PDF with better metadata. Tagging a document does not make its logic executable.
• It is not a language model reading regulation. A model produces a plausible interpretation, not a bound decision.
• It is not compliance reporting. Reporting describes what happened after the fact; authority constrains what may happen at all.

Why it matters now

AI systems act at a speed and volume where human interpretation cannot sit in the loop for every decision. If the authority layer stays in prose, one of two things happens: institutions slow their systems down to human pace, or they let models improvise institutional judgement. Neither is acceptable. Computable authority is the third option — move the interpretation work upstream, once, into an artifact that machines execute and humans govern.

A regulator grants operating licences under a statute that runs to sixty pages. Today an officer reads the file, forms a view, and records an outcome. The reasoning that produced the outcome is not stored anywhere a machine — or a court — can replay.

Expressed as computable authority, the same statute becomes an artifact with named definitions, explicit conditions, and an evaluable decision path. The officer still governs the artifact; the artifact governs the decision, and the version that produced it can be retrieved unchanged.