Architecture and orchestration
Agent contracts, shared state, evidence ledgers, handoffs, memory boundaries, protected actions, and explicit human authority.

THE OFFICIAL ATLAS FIELD GUIDE
Engineering Observable, Governed, Human-Accountable Agent Systems
This standalone book is the technical companion to the public Atlas and its GitHub repositories. It gives readers enough evidence to understand and evaluate the method, while organization-specific assessment, implementation, training, and delivery remain private professional work.
Mahsa Keikha, PhD
INSIDE THE FIELD GUIDE
Agent contracts, shared state, evidence ledgers, handoffs, memory boundaries, protected actions, and explicit human authority.
Deterministic reference implementations, test structure, CI, CodeQL, failure injection, red teaming, and release evidence.
Ten governed decision rooms, including a complete F117 Digital Twin Engineer case study and interactive browser simulations.
How leaders can recognize evidence gaps, authority risks, and the point where organization-specific professional review becomes necessary.
WHY IT EXISTS
A trustworthy agent system should let us inspect the evidence, the handoffs, the permissions, the failures, and the exact point where a person still has authority.
FOR ORGANIZATIONS
Identify the system boundaries, evidence gaps, protected actions, and authority questions that require attention before deployment.
Build shared engineering judgment around orchestration, evaluation, observability, risk, and meaningful human control.
Translate one consequential workflow into a bounded architecture and evidence plan shaped around your organization.
Give leadership a defensible view of priorities, blockers, ownership, and the next responsible investment decision.
OPEN FIELD GUIDE
The public field guide helps teams ask better questions. Private engagements begin where generic guidance ends, inside the organization's actual workflow, evidence, controls, and accountable decision structure.