Predicate-Scoped Authority for Conflict Resolution Between Mutually Untrusted Agents

Abstract

When autonomous agents from different trust domains exchange facts, those facts conflict, and a receiving agent must decide which value governs. We make two claims. First, authority over a contested fact is properly scoped to the (principal, predicate) pair, not to the principal alone: we prove that a per-predicate authority assignment is not representable by any global ranking of principals (Theorem 1), and we measure the ceiling this imposes on predicate-independent resolvers. Second, when the authority class is derived at an authenticated boundary rather than read from the submission, a writer that over-asserts its own authority cannot change which value governs. We evaluate four deterministic resolvers on generated conflicts whose governing authority varies by predicate, show that predicate-independent strategies (a global per-principal ranking; a static precedence table) top out well below full accuracy while a per-predicate rule resolves them by construction, and show in a controlled adversarial setting that a boundary-derived per-predicate resolver is not captured by an over-asserting writer whereas a recency rule always is. A small study with two instruction-tuned models indicates that when the same conflicts are put to a model to adjudicate, it resolves them more reliably when the authority signal is made explicit than when it is left implicit. The transport and the idea of trust-based resolution are not new; the contribution is authority scoped to the predicate and derived at the boundary, evaluated as a conflict-resolution primitive.