BOB STEWART

The Architecture of Accountable Autonomy

Systems that can act must also be able to answer.

As intelligence becomes capable of pursuing goals, invoking tools, moving information, recommending decisions, and acting across institutional boundaries, trust can no longer remain an assumption. It must become architecture.

See the obligations Explore IRB.Institute (opens in new tab)
The Doctrine — Canonical Text

Reviewable Intelligence

Version 1.0 — August 1, 2026 · Author and sole amending authority: Bob Stewart

The law

Authority before action. Evidence after action.

Power is not made trustworthy by good intention. It becomes trustworthy through bounded authority, visible consequence, and durable evidence.

The registers

This doctrine is quoted at three lengths. All three are the same law.

The aphorism
Authority before action. Evidence after action.
The compression
Logged. Traceable. Reviewable.
The full statement
The five questions, connected in a durable record.

The five questions

A consequential system — human, institutional, or autonomous — is reviewable when it can answer five questions, and when the answers remain connected in a durable record.

  1. 1 · Evidence. What is known, observed, inferred, missing, and materially disputed?
  2. 2 · Authority. Who is empowered to decide, under which role, policy, and risk context?
  3. 3 · Review. What evaluation makes the recommendation fit to act upon?
  4. 4 · Execution. What action was authorized, performed, bounded, and observed?
  5. 5 · Receipt. Can the institution reconstruct why the decision existed and what followed?

Break any link and the record becomes folklore. Governance is not a layer added after intelligence. It is how intelligence earns the right to act.

The obligation

The work is not to slow the future down. The work is to make the future worthy of acceleration.

Reviewable Intelligence, v1.0 · BLAKE3-256 b6133986fa14da3b · bobstewart.com/architecture/

Downstream surfaces cite this document by version, hash, and home URL. They do not restate it. Amendments create a new version upstream; no version is ever rewritten. Value and expression accrue downstream; reference flows upstream, immutably.

The Estate Map

Two node classes, one doctrine.

Expressions

Surfaces owned and accountable to Bob Stewart. Each carries the estate lineage sentence and cites the doctrine.

The estate publishes engagement post-mortems through one instrument: the field report (opens in new tab). One instrument upstream; reports downstream.

Lineage

Independently founded and owned ventures, doctrine-informed at their founders’ election. The map shows doors, never pipes.

Lineage ventures are independently founded and owned. They inform themselves by the doctrine at their founders’ election. The estate holds no authority over them, and their value accrues entirely to their founders.

Naming: legal names in footers; brand names in prose; no bare acronyms anywhere in the estate. Apex Velocity Catalysts, PBC-style legal designations, and similar formal names appear in footers and legal pages only.

01 — The Threshold

The question has changed.

The first generation of software waited for human hands. The next calculated, searched, ranked, and recommended. Emerging systems can plan, delegate, call tools, exchange information, and initiate consequential work.

The governing question is therefore no longer merely, “Can it perform?”

We must also ask: Who authorized it? What may it pursue? Which evidence informed it? What limits bind it? Who can review, interrupt, revoke, or appeal its action? What durable record will remain?

These are not only technical questions. They are questions of institutional legitimacy.

Power was never the only problem. Stewardship was.

02 — Five Obligations

Accountable autonomy must remain connected across the whole decision.

01

Identity

Who or what is acting? Is the identity sufficiently established for this role, context, jurisdiction, risk, and moment?

02

Authority

Who delegated permission? What scope, policy, credential, purpose, and prohibition define the boundary of action?

03

Review

What human or institutional judgment is required before, during, or after action? How are dissent, appeal, conflict, and uncertainty handled?

04

Evidence

What was known, observed, inferred, disputed, or missing? Can the institution reconstruct why the decision existed and what followed?

05

Revocation

How can authority be paused, withdrawn, expired, or superseded? What happens to work already initiated, and what evidence remains after permission changes?

No single model, policy document, identity token, audit log, or approval screen satisfies these obligations alone. They must remain connected before, during, and after action.

03 — The Operating Architecture

Five functions. One reviewable chain of consequence.

Lifecycle governance

Expression: AI-SDLC Institute

How is a system proposed, classified, designed, tested, approved, deployed, monitored, changed, suspended, and retired? Lifecycle gates turn responsible intent into repeatable operating discipline.

Institutional review

Expression: AI-IRB, IRB.Institute, OpenIRB, Decision Forum

Who is competent and authorized to review consequence? What evidence must be present? How are conflicts, dissent, holds, approvals, waivers, incidents, and appeals recorded?

Bounded authority

Expression: Autonomous Verifiable Credentials and explicit delegation

What is a person or agent permitted to do, for which purpose, in which context, under what policy, until what time, and subject to which revocation path? Identity is necessary. Bounded authority makes identity actionable without making it unlimited.

Durable evidence

Expression: EXOCHAIN (opens in new tab)

What proof remains after action? EXOCHAIN connects identity, consent, authority, policy, provenance, execution, time, revocation, and receipts so institutional memory does not collapse into disconnected logs and retrospective narrative.

Domain consequence

Expression: CyberMedica, LiveSafe, and other governed applications

How does the architecture become operational where failure has specific human meaning? Clinical systems make the question immediate: patients, professionals, protocols, credentials, consent, evidence, and oversight must remain connected. Personal-safety and continuity systems make it intimate: the people being protected are specific, known, and loved.

These are related functions and active fields of work. This page does not represent every component as a completed integration, generally available product, or substitute for an institution’s legal, clinical, regulatory, security, or ethical obligations.

04 — What Must Remain

A consequential decision should survive its moment.

A conventional log may show that something happened. A reviewable decision record should preserve enough context to understand why it was allowed to happen.

At minimum, the institution should be able to reconnect:

  • the person or agent;
  • the role being exercised;
  • the proposed or completed action;
  • the governing context and risk;
  • the evidence available at decision time;
  • the authority and policy relied upon;
  • the review or approval that occurred;
  • the execution and observed result;
  • any dissent, exception, incident, or revocation; and
  • the durable receipt binding the record together.

In the autonomous age, memory is not nostalgia. Memory is defense, accountability, and the possibility of learning.

05 — Boundaries

Governance is not omniscience, surveillance, or centralized control.

The architecture does not assume that every decision can be made correct in advance. It does not eliminate human judgment, uncertainty, conflict, or moral responsibility. It does not grant one platform unlimited command over people, agents, or institutions.

Its purpose is narrower and more demanding: make authority explicit, evidence inspectable, decisions reviewable, execution bounded, and accountability difficult to erase.

That is enough to change the quality of institutional action.

06 — The Work Ahead

The future must be governed, witnessed, and remembered.

The age of autonomous intelligence is not waiting for institutions to become ready. Capability is already moving into decisions, workflows, products, care, infrastructure, finance, security, and government.

The choice is not between acceleration and governance. The choice is between acceleration with memory, authority, and recourse—or acceleration that leaves institutions unable to explain their own actions.

The work is to make the future governable enough to deserve its power.

Convene institutional review (opens in new tab) Explore AI-SDLC Institute (opens in new tab) Bring a consequential mandate to MaineSoftware (opens in new tab)