# Authority

> Bounded source authority, actor authority, review responsibility, and delegation.

Authority answers two different questions:

1. What may this source establish?
2. What may this actor cause the system to do?

Conflating them lets a credible source become an operator, or a powerful
operator make unsupported claims true.

## Source authority

**TM-AUT-001 — Scoped source authority.** Every source used to support current
truth **MUST** have an authority scope no broader than the claims it is
entitled to establish.

**TM-AUT-002 — No authority by location.** A source **MUST NOT** gain universal
authority because it is called a system of record, is important, is recent, or
stores a value in a canonical repository.

**TM-AUT-003 — Conflicting sources.** When differently authoritative sources
conflict, reconciliation **MUST** preserve the conflict or record the reviewed
reason one claim governs the current value. It **MUST NOT** silently choose by
recency, confidence, or ingestion order unless that rule is itself declared
and authorized.

## Actor authority

**TM-AUT-004 — Declared actor capabilities.** The implementation **MUST**
declare which actors may inspect, draft, review, accept, apply, publish,
communicate, or operate within the coherence boundary.

**TM-AUT-005 — Review responsibility.** Every accepted consequential change
**MUST** identify the authority responsible for the reviewed judgment. A
service or AI Peer may be the actor performing mechanical application without
becoming the reviewing authority.

**TM-AUT-006 — Exact delegation.** Delegation **MUST** state the delegating
authority, delegate, capability, scope, and applicable conditions. Delegation
**MUST NOT** be inferred from tool access or prior convenience.

**TM-AUT-007 — No self-expansion.** An actor **MUST NOT** expand its own
authority through the same capability being expanded. The expansion requires
the review path already authorized to make that change.

## AI Peers

An AI Peer may inspect broad context, compare evidence, draft changes, explain
consequences, and invoke deterministic tools within its declared authority.
Model confidence, reasoning quality, continuous presence, or operator trust do
not create authority.

**TM-AUT-008 — Model results are claims.** A model-produced extraction,
classification, recommendation, or conclusion **MUST** enter the system as a
derivative claim or proposed work unless an independently authorized,
deterministic rule establishes otherwise.