Cataloged reference
A documented multi-agent architecture with a defined domain purpose. Inclusion does not imply production readiness.
EVIDENCE BEFORE AUTHORITY
Know exactly what each claim proves.
The Atlas separates architecture, implementation, testing, simulation, independent review, and real-world deployment. These are different kinds of evidence and they should never be presented as interchangeable.
THE EVIDENCE LADDER
Six labels with six different meanings.
Structured implementation
Repository structure exposes roles, orchestration, tools, state, safety boundaries, tests, and documentation where applicable.
L3 Gold Standard
A project-defined engineering maturity status backed by recorded commit, CI, and benchmark evidence. It is not a regulatory certification.
Interactive simulation
A browser demonstration using synthetic scenarios to make coordination, evidence, failure modes, and human approval visible.
Independent review
A review claim requires a named or authorized reviewer, defined scope, date, findings, and disposition. No review is implied without that record.
Deployment evidence
A deployment claim requires an authorized organization, bounded use case, operating context, measured outcome, and permission to publish.
CURRENT PUBLIC EVIDENCE
A precise snapshot, not an inflated claim.
| Collection | Public evidence | Boundary |
|---|---|---|
| 170-system AtlasReference architectures across 15 domains | Public taxonomy, repository links, architecture documentation, and source history | Catalog inclusion alone does not establish L3 maturity or production use |
| Verified maturity cohortF36 through F60 | Recorded commits, CI runs, and held-out benchmark evidence in the maturity registry | L3 is the Atlas project standard, not legal or regulatory certification |
| Ten flagship simulationsInteractive browser scenarios | Visible agent handoffs, evidence, failure modes, and human approval points | Educational simulations are not operational control systems |
| F117 case studyPackaging-line motor digital twin | Synthetic data, deterministic calculations, traceable agent steps, and protected-action blocking | No claim of measured downtime, safety, maintenance, or production improvement |
Inspect the evidence directly.
The maturity registry records the exact systems that have completed the current acceptance gate.
CORRECTION AND REVIEW
The reference should become stronger when challenged.
Technical corrections
Document the affected system, claim, evidence, expected behavior, and reproduction path. Accepted corrections remain visible in repository history.
Independent review
Reviewers may assess architecture, evaluation, security, domain boundaries, documentation, or reproducibility without endorsing the whole collection.
Adoption records
Teaching, research, and organizational use are recorded only with a verifiable source or explicit permission from the adopting organization.
Release evidence
Material releases identify the source commit, affected systems, completed tests, known limitations, and any maturity impact.
Security findings
Unpatched vulnerabilities follow private responsible disclosure rather than public issues.
Commercial confidentiality
Public evidence supports scrutiny without publishing client information, private delivery methods, or organization-specific implementation details.