No self-authorization
An agent, workflow, or governed component cannot register, enlarge, bypass, or reinterpret its own authority.
Trust & boundaries
The governed system cannot grant itself authority.
CharterGate is designed around a strict separation: authority comes from verified external records, governance evaluation checks the exact proposal, and any real-world execution happens elsewhere.
Permanent boundaries
Authority is not a model inference.
Memory may remember, retrieval may find, and a model may reason, but none of those mechanisms determines what an actor is allowed to do.
Explicit trusted sources
A trusted source is a specific registered authority with a stable source identity, revision, provenance state, validity window, scope, exclusions, and explicit status. It is not a generic website or an AI guess.
Fail closed on material uncertainty
If required authority, provenance, tenant, actor, target, scope, condition, or other material binding cannot be established, the consequential lane stops.
What CharterGate is not
No hidden leap from evaluation to owner approval or action.
CharterGate does not claim legal, financial, medical, safety, or regulatory approval. An AUTHORIZED result is governance evidence only: it does not infer the owner's separate decision and it does not perform the proposed business action.
Read-only governance means CharterGate evaluates a proposal and may record its own governance results and audit evidence, but does not perform the proposed external action. Sending, publishing, deploying, paying, or changing an external system requires a separately authorized executor.
Trusted sourceDefines the bounded authority being checked
CharterGateEvaluates the exact request
Owner / decision-makerMakes any separate real-world decision
Caller / executorActs separately, if appropriate
Isolation and audit
One identity, the intended source context, one reconstructable evaluation.
Private-preview access can be scoped so a participant's credential is limited to intended trusted sources, while evaluator and registry-administration credentials remain separate and audit records identify the request, controlling source revision, evaluator/policy version, and returned result.
Tenant isolation
One tenant's source, authority record, proposal, or audit record must never authorize another tenant's action.
Credential separation
Evaluation credentials and authority-registry administration credentials should remain distinct, with secret values excluded from ordinary audit data.
Reconstructable evidence
Authorized operators should be able to determine what matched, what failed, what conditions applied, and which authority revision controlled.