Applied Evidence Layer

Human-Governed AI Workflows

Three public cases on authority boundaries, artifact-scoped provenance, and delegated execution, with declared controls, implemented artifacts, tested properties, and measured outcomes kept distinct.

This page presents three public repository-level cases concerning AI authority boundaries, provenance and artifact-scoped reconstruction, and delegated execution with retained human answerability. Each case separates declared controls, implemented artifacts, tested technical properties, and measured outcomes.

The cases also identify limited, case-specific correspondence with the NIST Artificial Intelligence Risk Management Framework (AI RMF 1.0) and the public ISO/IEC 42001 overview. Correspondence here means a bounded structural comparison. It does not mean that Meta-Writing Ecology has adopted or implemented either framework.

Evidence boundary

Declared control ≠ implemented artifact ≠ tested property ≠ measured outcome.

A public rule does not prove that a mechanism exists. A mechanism does not prove that a property was tested. A passing technical test does not establish an operational or organizational effect. Where no outcome measure is public, this page says Not measured.

The framework notes below are limited to exact MWE public evidence and official public framework concepts. Each correspondence identifies the shared structural function, the difference in scope, and the stronger inference that is not permitted. Shared vocabulary alone is not treated as correspondence.

AI Authority Boundaries and Human Oversight

Human-authority rules: Declared control
Metadata and machine-reading artifacts: Tested technical property

Condition. AI-assisted interpretation and implementation can encounter decisions about naming, classification, relation status, publication, navigation, and public/private boundaries that agents are not authorized to settle.

Control objective. Preserve final human authority and reduce the inference authority of automated interpretation when evidence is missing, conflicting, or unconfirmed.

Declared controls. Public repository instructions reserve specified conceptual and publication decisions to the repository owner. Candidate and navigation relations may not be promoted into confirmed relations. Unknown, conflicting, and unconfirmed interpretation states fail closed.

Implemented artifacts. Public route-metadata policy rejects unregistered BaseLayout routes and excludes authority-bearing metadata keys. A machine-reading state model represents source-access states, uncertainty flags, and claim scopes.

Tested properties. Public tests and validators check the route-metadata contract and the machine-reading state model. These tests establish properties of the artifacts; they do not test whether the human-authority rules are followed in practice.

Publicly established result. The repository publicly declares authority limits and exposes tested technical mechanisms that preserve selected metadata and machine-reading boundaries.

Measurement state. Not measured.

Inference ceiling. The evidence does not establish effective human oversight, reduced unauthorized promotion, measured governance improvement, or organizational risk reduction.

NIST AI RMF 1.0 correspondence. MWE’s documented authority boundaries and fail-closed interpretation rules correspond at a limited structural level to selected NIST AI RMF GOVERN and MAP concerns about documented roles, responsibilities, knowledge limits, and human oversight. This does not establish adoption, implementation, or completion of GOVERN or MAP.

ISO/IEC 42001 public-overview correspondence. At public-overview level, MWE’s documented authority limits and tested fail-closed metadata behavior have limited conceptual correspondence with ISO/IEC 42001’s high-level treatment of responsibilities, policies, processes, and controls. This does not establish an Artificial Intelligence Management System, operational control as a clause-level conclusion, implementation, adoption, or conformity.

Direct evidence. Link only to approved source-repository documents. Refer to website-repository implementation files by visible path without a GitHub URL.

AI Provenance, Version Control, and Artifact Reconstruction

Highest established level: Tested technical property

Condition. Public AI-readable artifacts can be versioned, validated, and traceable while still failing to establish the validity of a claim or the reconstructability of an entire publication system.

Control objective. Preserve source identity and scope technical reconstruction claims to the artifact actually covered by public evidence.

Declared controls. Public source-use and machine-reading boundaries distinguish provenance from validity, traceability from truth, metadata from conceptual authority, and version identity from complete reconstruction.

Implemented artifacts. Public manifests, schemas, version identities, validators, and a scoped correction-register mechanism retain selected evidence about public artifacts.

Tested properties. One tracked public dataset and its manifest are covered by public tests asserting byte-identical output across independent rebuilds.

Publicly established result. One tracked public dataset and its manifest are covered by public tests asserting byte-identical output across independent rebuilds. This artifact-scoped result does not establish deterministic reconstruction for the website, repository, publications, or other public releases.

Measurement state. Not measured.

Inference ceiling. The evidence does not establish complete reproducibility, preservation-grade reconstruction, organizational monitoring effectiveness, AI-risk measurement, or conceptual validity.

NIST AI RMF 1.0 correspondence. The version identities, validators, and artifact-scoped rebuild tests provide evidence infrastructure that may support selected NIST AI RMF MEASURE activities where those artifacts are used within an independently defined AI risk-measurement process. Validators, version identities, and reconstruction tests provide evidence infrastructure that may support selected NIST AI RMF Measure activities. They do not by themselves measure AI risk, system trustworthiness, or organizational effectiveness.

ISO/IEC 42001 public-overview correspondence. At public-overview level, MWE’s versioned evidence and artifact-scoped checks have limited conceptual correspondence with ISO/IEC 42001’s high-level emphasis on traceability, documented policies and information, monitoring, and improvement processes. This does not establish “documented information” or “operational control” as clause-level mappings, an organizational corrective-action process, an Artificial Intelligence Management System, or conformity.

Direct evidence. Link only to approved source-repository artifacts and tests. Describe website-repository files by visible path without a GitHub URL.

Multi-Agent Execution and Retained Answerability

Highest established level: Declared control with partial public execution history

Declared Workflow Architecture

Condition. Delegated AI-assisted work can distribute execution and review without transferring final answerability.

Control objective. Bound delegated work, separate execution from review where specified, and retain final user authority over publication and boundary-sensitive decisions.

Declared controls. Public rules describe bounded task scope, role separation, review requirements, retained answerability, and final user authority over publication, naming, classification, relation confirmation, top navigation, and merge decisions.

Observed Public Execution

Aggregate statement. Public worklogs record selected bounded tasks, separate review events, specified corrections, and test results. The reviewed public material does not provide an explicit human final-acceptance record for a complete normalized execution chain.

Measurement state. Not measured.

Inference ceiling. The evidence does not establish formal segregation of duties, independent audit, proven oversight, measured review effectiveness, multi-agent superiority, transferred responsibility, or a complete public execution history.

NIST AI RMF 1.0 correspondence. MWE’s declared role boundaries, retained answerability, and final user authority have limited structural correspondence with selected NIST AI RMF GOVERN concerns about documented responsibilities, human review, and accountability. This does not establish implementation or completion of GOVERN or effective oversight.

ISO/IEC 42001 public-overview correspondence. At public-overview level, MWE’s declared role boundaries, review requirements, and retained human authority have limited conceptual correspondence with ISO/IEC 42001’s high-level emphasis on defined responsibilities and oversight. This does not establish formal segregation of duties, internal audit, an Artificial Intelligence Management System, implementation, adoption, or conformity.

Direct evidence. Use source documents that declare the boundaries. Do not include a named end-to-end execution chain, consolidated worklog references, complete PR/commit/reviewer chains, reusable task packages, prompts, internal routing logic, or full defect-and-correction sequences.

Bounded correspondence with the NIST AI RMF 1.0

NIST’s official publication is the Artificial Intelligence Risk Management Framework (AI RMF 1.0). Its Core contains GOVERN, MAP, MEASURE, and MANAGE. The functions are not a checklist or a required linear sequence; they may be applied in an order suited to the user and should be iterative. GOVERN is cross-cutting across AI risk management.

Case 01 has limited structural correspondence with selected GOVERN and MAP concerns about documented responsibilities, knowledge limits, and human oversight. Case 02 identifies technical evidence infrastructure that may support selected MEASURE activities, but does not itself measure AI risk or trustworthiness. Case 03 has limited structural correspondence with selected GOVERN concerns about documented responsibilities, human review, and accountability.

Validators, version identities, and reconstruction tests provide evidence infrastructure that may support selected NIST AI RMF Measure activities. They do not by themselves measure AI risk, system trustworthiness, or organizational effectiveness.

These correspondences do not establish that Meta-Writing Ecology has adopted or implemented the AI RMF, completed any Core function, created an AI RMF profile, reduced AI risk, or established effective governance.

Public-overview correspondence with ISO/IEC 42001

ISO’s public overview describes ISO/IEC 42001:2023 as specifying requirements for establishing, implementing, maintaining, and continually improving an Artificial Intelligence Management System. Public ISO material also discusses responsibilities, policies, processes and controls, traceability, monitoring, corrective actions, and continual improvement.

Case 01 has limited conceptual correspondence with high-level responsibility and control concepts. Case 02 has limited conceptual correspondence with high-level traceability, documentation, monitoring, and improvement concepts. Case 03 has limited conceptual correspondence with defined responsibilities and oversight.

This is a public-overview conceptual correspondence, not a clause-level ISO/IEC 42001 mapping.

These correspondences do not establish an Artificial Intelligence Management System, adoption or implementation of ISO/IEC 42001, conformity, certification readiness, internal audit, a compliance gap analysis, or the effectiveness of any monitoring, corrective-action, or improvement process.

This page presents bounded correspondence only. It does not establish adoption or implementation of the NIST AI RMF or ISO/IEC 42001, conformity with ISO/IEC 42001, certification readiness, an audit result, a compliance assessment, or a measured organizational outcome.

This page does not establish

This page is a bounded public evidence surface. It is not a Model, Cross, Log, Protocol, ontology layer, Registry entry, confirmed formal relation, governance product, compliance method, audit system, certification service, or complete implementation methodology. Direct source documents remain necessary for conceptual claims.