# Autonomous Trust Resolution Protocol — Draft v0.1

**Publisher:** AgentProof  
**Status:** Public draft  
**Purpose:** Transaction-admission decisions for autonomous systems.

## Question

A resolver evaluates:

> May this agent perform this action, for this principal, with this counterparty, under this mandate, at this amount, right now?

## Request model

A request contains:
- principal;
- agent;
- mandate;
- counterparty;
- action;
- transaction;
- evidence;
- attestations;
- recourse context.

## Decisions

A conforming resolver returns exactly one primary decision:

- `ALLOW`
- `ALLOW_WITH_LIMITS`
- `REQUIRE_MORE_PROOF`
- `REQUIRE_RECOURSE`
- `REQUIRE_HUMAN_APPROVAL`
- `DENY`

The response also includes a policy version, reasons, effective constraints, evaluation time, short validity window, request digest, and resolution digest.

## v0.1 deterministic policy

AgentProof's first reference policy checks:
1. required identities and action fields;
2. mandate validity window;
3. action scope;
4. allowed counterparty restrictions;
5. currency restrictions;
6. transaction amount against a hard maximum;
7. minimum evidence requirement;
8. minimum attestation requirement;
9. recourse requirement above a configured threshold;
10. human approval requirement above a configured threshold.

A hard mandate violation returns `DENY`.
Missing required evidence returns `REQUIRE_MORE_PROOF`.
Required but unavailable recourse returns `REQUIRE_RECOURSE`.

## Non-goals

v0.1 does not claim universal legality, sanctions compliance, creditworthiness, insurance coverage, or objective truth. It is a deterministic admission protocol whose policy inputs can later be extended by recognized policy authorities and trust providers.

## Economic model

The open protocol is free to implement. AgentProof's hosted resolver is a paid machine endpoint. This preserves an open ecosystem while making the operating network commercially sustainable.