Public architecture brief

Put authority
at the action boundary.

TrustCore’s design direction separates a proposed action from permission to create an external effect. This is a conceptual overview, not a product specification or proof of deployment.

The conceptual control flow

INPUT

Proposed action

Intent, target and supporting evidence.

CONTROL BOUNDARY

Authority + policy

Identity, scope, current conditions and human authorization where required.

ONLY IF PERMITTED

Controlled execution

A bounded external effect and a reviewable outcome.

If required authority or evidence is missing, the proposed action must not silently become execution. A refusal or escalation is a valid control outcome.

  1. 01Observe

    Collect relevant state and evidence.

  2. 02Decide

    Evaluate authority, policy and consequence.

  3. 03Enforce

    Apply the decision before an external effect.

  4. 04Execute

    Perform only the permitted action.

Decision vocabulary

CONTINUEVERIFYESCALATEINTERRUPTDENY

Human authority stays explicit.

Consequential actions need a named decision owner and agreed approval conditions. Reviewing an answer is not the same as authorizing its execution.

Evidence spans the transition.

Review what was known at the decision point and what effect followed. A successful execution does not retroactively prove that permission was valid.

Implementation details, internal schemas and proprietary control mechanisms are intentionally outside this brief. A pilot determines the exact scope and evidence requirements.

Start with one workflow

Where does AI output become action?

Define the action, name the authority and agree what evidence would demonstrate control.

Discuss a pilot