CONCEPTS

Decision rights

Most of a claim is about evidence — what the decision rested on. The optional decision_rights field is about something else: who was entitled to make it. Authority and evidence are different axes, and a defensible record carries both.

A claim submitted to POST /v1/evaluate can include a decision_rights array — described in the reference as principal/role/permission assertions. It is the part of the record that names the actor and the authority under which they acted: a principal (a person, an agent, or a service), the role they held, and the permissions that role carried at the moment of the decision.

Two questions, not one

It is tempting to collapse accountability into a single question — was this decision good? — but a decision can fail in two independent ways. It can be poorly grounded: thin evidence, low confidence, no provenance. Or it can be made by someone with no business making it: the right answer reached by the wrong hand is still a governance failure, because next time the same actor will reach the wrong one and nobody will have standing to object.

The evidence model — sources, confidence, provenance — answers the first question. Decision rights answer the second. A receipt that records both lets a later reviewer ask, separately and clearly: was the decision well-founded, and was the actor entitled to make it?

What an assertion looks like

The reference specifies decision_rights only as principal/role/permission assertions, so treat the shape below as illustrative — the authoritative schema is the live OpenAPI spec. Conceptually, each assertion binds an actor to the authority they claimed:

Illustrative — confirm field shapes against the OpenAPI spec

"decision_rights": [
  {
    "principal": "svc:deploy-bot",
    "role": "release-manager",
    "permission": "deploy:production"
  }
]

Carried inside the claim, these assertions travel into the receipt as part of the recorded subject and context — alongside the agent, agent_model, and human_operator on the action object. The result is a record that does not just say what happened and on what evidence, but by whom, in what capacity.

Authority is necessary, not sufficient

Recording decision rights is how a system shows an actor was authorized. That matters — but it is worth being precise about what it buys. Authorization establishes that the actor was permitted to act. It does not, by itself, establish that the affected party is owed nothing further. You can have every right to make a decision and still owe an account of it; the two are different standards, and conflating them is how institutions mistake permission for legitimacy.

This is exactly why decision rights live in the receipt rather than gating it from outside. The point is not to wave authority as a conversation-ender — we were allowed to — but to put the authority on the record next to the evidence, so it can be examined like everything else. The companion essay Authorization is not legitimacy develops the argument; here it is enough to say that an admissible decision is one where both the grounds and the standing are written down and contestable.

Reviewing standing after the fact

When a decision is later challenged, decision rights give the reviewer something concrete to check: did the named principal actually hold that role, and did that role actually carry that permission, at that time? A claimed right that doesn't hold up is itself a finding — the same way a gap in provenance is. Recording the assertion is what makes the check possible; an unrecorded authority can only be reconstructed from memory and goodwill, which is to say it cannot be reconstructed at all.