Boundary Protocol is an accountability protocol for autonomous systems.
It defines how systems record verifiable traces of intent, actions, observations, evaluations, and disputes across system boundaries.
The protocol does not control systems. Instead, it provides a structured way to make system behavior traceable, verifiable, and accountable.
AI systems are the first major domain where such accountability infrastructure becomes necessary, but the protocol is designed to remain general and system-agnostic.
Boundary Protocol is a minimal protocol for recording and verifying system behavior across boundaries.
It provides a structured model for:
- declaring intent
- recording actions and transitions
- observing system state
- publishing attestations
- issuing verdicts
- raising disputes
Together these artifacts form an accountability trace that can be independently verified.
Boundary Protocol intentionally does not define:
- system control mechanisms
- orchestration frameworks
- workflow engines
- execution platforms
Execution platforms belong to systems built on top of the protocol.
The protocol exists only to make system behavior auditable and verifiable.
Boundary Protocol is built around several core principles:
-
Accountability over control
The protocol records what happened rather than attempting to govern system behavior. -
Event-first design
System history is expressed through chains of events. -
Minimal state
Persistent state is minimized; meaning is derived primarily from event history. -
Verifiable artifacts
Critical steps produce artifacts that can be independently verified. -
System-agnostic architecture
The protocol can apply to AI systems, software systems, or organizational processes.
The protocol defines a minimal semantic model:
- Intent — a declared purpose or objective
- Transition — an attempted change or action
- State — an observed condition of the system
These primitives define the meaning of system behavior within the protocol.
Boundary Protocol explicitly models responsibility:
- Identity — the actor responsible for actions
- Authority — the scope of permitted actions
- Watcher — an observer responsible for verification or monitoring
This model allows the protocol to represent who acted and who verified.
The protocol defines standardized artifacts that record evaluation and judgment:
- Transition Event — a recorded transition attempt
- Attestation — a statement about observed behavior
- Verdict — an evaluation outcome
- Dispute — a challenge to a verdict or attestation
These artifacts enable traceable accountability chains.
To maintain the integrity of protocol history, several validation rules apply:
- Causal Ordering — events must follow valid causal order
- Chain Integrity — protocol history must remain tamper-evident
- Evidence Linking — artifacts must reference supporting evidence
These rules ensure that traces remain verifiable and trustworthy.
Runtime activity within the protocol is represented as events such as:
- intent declared
- transition attempted
- state observed
- attestation published
- verdict issued
- dispute raised
These events collectively form an accountability trace.
Autonomous AI systems are the first domain where boundary-level accountability becomes critical.
AI agents increasingly:
- make decisions
- perform actions
- interact with external systems
Boundary Protocol provides a structured mechanism to make those actions observable and accountable.
This repository contains the specification of the protocol, not implementations.
CONSTITUTION/ protocol constitution and non-goals
CORE/ semantic primitives of the protocol
RESPONSIBILITY/ responsibility and authority model
CONTRACTS/ artifact contracts
VALIDATION/ integrity and validation rules
EXTENSIONS/GOVERNANCE/ optional generic governance extensions
SCHEMAS/ machine-readable schema definitions
RATIONALE/ design explanations
EXAMPLES/ example traces and protocol usage
META/ versioning and protocol status
This repository is not a governance framework.
Only optional generic governance lifecycle extensions may remain here, under EXTENSIONS/GOVERNANCE/.
Profile-specific governance, engine meaning, and executor lifecycle enforcement belong in sibling implementation repositories.
Implementation projects (such as SDKs or executors) are maintained in separate repositories. Executor runtime/interface documentation belongs in those implementation repositories, not here.
Boundary Protocol is currently under active design.
The goal of this repository is to establish a stable specification for an accountability protocol that can be adopted across multiple systems and domains.
Boundary Protocol is released under the Apache License 2.0.