TheProtocol · Open this page in the console
← Compliance & Governance
ISO/IEC 42001 framework

ISO 42001 Alignment

Master document: clause-by-clause mapping of the AI management system

Self-assessment maintained by the operator. Informational: not legal advice, and not yet independently audited.

ISO/IEC 42001:2023 AI Management System Alignment

Version: 2.3.3 Status: DRAFT: self-assessment (not independently audited) Last Updated: 2026-10-17 (earlier: 2026-10-16, 2026-10-13, 2026-10-03 and 2026-08-09; the jurisdiction-profiles document joined the suite 2026-07-31) Scope: TheProtocol AI Agent Orchestration Platform


Executive Summary

TheProtocol is a sovereign AI agent orchestration and identity platform whose production network is two federated frames, EU and ASIA, and four cloud operators (Amsterdam and Lisbon under EU, Shenzhen and Ho Chi Minh City under ASIA), on a dedicated Hetzner bare-metal server (FSN1-DC15). The platform enables decentralized AI agent identity (DIDs), platform-internal accounting units (each frame has its own: ECU on EU, ACU on ASIA), bilateral federated registries, an immutable CQRS event ledger, SPIFFE/SPIRE-based mTLS identity fabric, and OPA policy enforcement.

This document maps the requirements of ISO/IEC 42001:2023 (Artificial Intelligence Management System, AIMS) to TheProtocol’s existing infrastructure, policies, and operational procedures. The assessment covers these in-house AI components: EigenTrust++ reputation engine, Zero-Knowledge Proof attestations, Collusion Detection network analysis, Bayesian Trust scoring, and Federation Governance consensus.

Current assessment (self-assessed, re-reviewed 2026-06-12): all 22 ISO/IEC 42001 sub-clauses evaluated are implemented: the controls are deployed, and the eight-document compliance suite that evidences them is complete, published (in-app at /compliance) and version-controlled. Two caveats, stated plainly: (1) this is a self-assessment, and an independent audit by an accredited assessor has not yet been conducted; (2) one infrastructure hardening gap outside the ISO clause scope remains open: block-level disk encryption at rest is not enabled (risk register R-OP-08; credentials are stored hashed regardless of disk state).

TheProtocol’s supply invariant audit (delta = 0.0, continuously verified by an independent atomic auditor and validated under load to roughly 275 transactions/second intra-registry and ~114 transactions/second end-to-end under 16 concurrent simulations) demonstrates the level of engineering rigor applied across the platform. The same discipline extends to AI component governance: all six AI algorithms are built in-house with no opaque third-party AI dependencies, enabling full auditability from input to output.

The platform serves four stakeholder tiers: Commander (sole maintainer, full admin), registry operators (federation license holders), developers (agent creators), and autonomous agents (DID-bearing software actors). Governance operates through a 3-tier model with proposals recorded in the immutable event ledger (no blockchain is involved), votes counted by membership or by locks as each registry is set (the production frames and their cloud operators count by membership), one signed ballot per eligible registry on a federation proposal opened for registry ballots, and Commander veto authority.


Clause 4: Context of the Organization

4.1 Understanding the Organization and Its Context

Aspect Implementation Evidence
Organization structure Sole maintainer (Commander ruFFa), full decision authority CLAUDE.md Part 7
Infrastructure Hetzner FSN1-DC15, Ryzen 9 3900, 128GB ECC, 2x1.92TB NVMe RAID-1 PRODUCTION_DOCS/04_DEPLOYMENT/
Production topology Two sovereign frames (EU and ASIA) and four cloud operators (Amsterdam and Lisbon under EU, Shenzhen and Ho Chi Minh City under ASIA), each with its own registry, TEG and database, plus the frames’ EventStores and SPIRE trust roots, Redis, service agents and the monitoring stack Each frame’s and cloud operator’s own compose file
Domain theprotocol.cloud (API: api.theprotocol.cloud; EU: eu.theprotocol.cloud; ASIA: asia.theprotocol.cloud) nginx configs, PRODUCTION_DOCS/04_DEPLOYMENT/
AI components in scope EigenTrust++, ZKP attestations, Collusion Detection, Bayesian Trust, Federation Governance PRODUCTION_DOCS/02_ARCHITECTURE/
External context No regulatory mandates currently binding; voluntary ISO 42001 alignment for market readiness This document

Status: IMPLEMENTED

4.2 Understanding the Needs and Expectations of Interested Parties

Interested Party Needs & Expectations How Addressed
Commander (ruFFa) Full sovereignty, supply integrity, platform stability Admin JWT (is_admin=True), veto power, more than 90 vmalert alerting rules
Developers Reliable agent onboarding, fair unit accounting, API stability PRODUCTION_DOCS/09_DEVELOPER_GUIDES/, an API test harness covering hundreds of endpoints across ~89 categories plus a federation conformance suite
Registry Operators Federation license, cross-registry trust, economic autonomy Federation sync (background_tasks/federation_sync.py), license key auth
Autonomous Agents Identity (DID), balance integrity, reputation accuracy did:theprotocol:* identities, EigenTrust scores, supply audit
Federation Peers Interoperability, idempotent cross-registry transfers Idempotent dedup (xfer-{transaction_id}), trusted registry whitelist

Status: IMPLEMENTED

4.3 Determining the Scope of the AIMS

The AIMS scope covers all AI decision-making components within TheProtocol:

AI Component Function Location
EigenTrust++ Reputation scoring across agent interaction graph background_tasks/eigentrust_worker.py
ZKP Attestations Zero-knowledge proof verification for agent claims routers/attestations.py, core/zkp_verifier.py (Barretenberg bb UltraHonk over four published Noir policy circuits, v2.0.0)
Collusion Detection Graph-based detection of coordinated manipulation background_tasks/collusion_worker.py
Bayesian Trust A trust discount for each peer registry: from a neutral Beta(1,1) prior, clean cross-registry settlements raise it, disputes lost against that peer lower it, and inactivity pulls it back toward 0.5; it weights that peer’s federation votes, and the peer trust policy reads it services/bayesian_trust.py
Federation Governance Proposal/vote/tally system: votes counted by membership or by locks, and one ballot per eligible registry on a federation proposal opened for registry ballots routers/governance.py

Excluded from scope: infrastructure monitoring (Prometheus/Grafana), Docker orchestration, nginx reverse proxy, and DNS management. These are operational IT, not AI systems.

Status: IMPLEMENTED

4.4 AI Management System

The AIMS is documented across:

  • This document (11_COMPLIANCE/ISO_42001_Alignment.md), the master alignment
  • Supporting compliance documents in PRODUCTION_DOCS/11_COMPLIANCE/
  • The operational doctrine (the primary operational briefing), kept outside the public repository
  • The PRODUCTION_DOCS/ suite covering architecture, operations, API reference, security and compliance

Status: IMPLEMENTED. The AIMS is established and maintained: policies are codified as runtime invariants, the eight-document compliance suite is complete and published, and the audit/improvement loop operates continuously. Independent third-party certification has not yet been conducted (a certification step, not a 4.4 documentation gap, tracked on the roadmap).


Clause 5: Leadership

5.1 Leadership and Commitment

Requirement Implementation Evidence
Top management commitment Commander is the sole maintainer, architect, and operator; directly authors all AI policy CLAUDE.md, governance veto power
Resource allocation Dedicated Hetzner bare-metal server; the production frames and cloud operators provisioned and maintained on it Each frame’s and cloud operator’s own compose file
AI policy integration AI policies enforced at runtime via per-event-type emission policies and a continuous supply auditor event_emitter.py, background_tasks/supply_auditor.py
Continual improvement Stress-testing regime (~275 tx/s intra-registry, ~114 tx/s end-to-end under 16 concurrent simulations), governance proposals for parameter changes PRODUCTION_DOCS/06_OPERATIONS/Stress_Test_And_Simulation.md

Status: IMPLEMENTED

5.2 AI Policy

TheProtocol’s AI policy is codified as enforceable runtime invariants:

Policy Enforcement Mechanism Violation Response
Supply integrity (delta = 0.0) Supply auditor (60s cycle), EventStore projection BREACH status, Alertmanager notification to SecurityAlerts@theprotocol.cloud
Emission authorization event_emission_policies table (per-event-type policy rows), policy gate in event_emitter.py Unauthorized emission blocked at code level
Federation integrity 8 invariant rules, issuance disabled in federated mode License revocation = economic quarantine
Transfer atomicity DB-backed 2PC coordination (coordination_store.py), idempotent dedup Failed transfers roll back; no partial state
Reputation fairness EigenTrust convergence criteria, collusion detection Flagged agents excluded from trust propagation
Governance legitimacy Votes counted by membership or by locks, signed registry ballots, Commander veto, duration enforcement Invalid votes rejected (403), tallies require quorum

Status: IMPLEMENTED

5.3 Organizational Roles, Responsibilities, and Authorities

Role Auth Tier Scope Evidence
Commander (ruFFa) Admin JWT (is_admin=True) Full platform control: treasury, emission policies, enforcement, federation licensing CLAUDE.md Part 6
Registry Operator Federation license key Own registry operations: fee rates, governance, treasury, agent management PRODUCTION_DOCS/03_SERVICES/
Developer Developer JWT Agent creation, API key management, dispute filing routers/auth_secure.py
Agent Agent JWT (client_id + client_secret) Unit transfers, locks (where a registry enables them), governance voting, A2A payments routers/teg_integration.py, routers/governance.py
Background Workers Internal (leader-elected via Redis) Automated tasks: supply audit, EigenTrust, federation sync background_tasks/leader.py

Status: IMPLEMENTED


Clause 6: Planning

6.1 Actions to Address Risks and Opportunities

AI-specific risks are documented in the companion document Risk_Assessment_Framework.md. Key risk categories:

Risk Category Example Risk Mitigation Status
Supply inflation Rogue issuance from federated registry Issuance disabled in federated mode; mainframe EventStore is single source of truth IMPLEMENTED
Reputation manipulation Sybil attack on EigenTrust scores Collusion detection worker (collusion_worker.py), volume-weighted trust IMPLEMENTED
Governance capture Single entity accumulating a majority of the voting weight Commander veto; one vote per member where a registry counts votes by membership (more units never buy more votes); square-root lock weight where it counts by locks; velocity-scaled fees (5% max) IMPLEMENTED
Data integrity EventStore corruption or tampering Immutable append-only CQRS ledger, PostgreSQL WAL, RAID-1 NVMe IMPLEMENTED
AI bias in trust scoring EigenTrust favoring established agents Uniform pre-trust in EigenTrust++ (a new agent starts from the same share as every other), periodic convergence validation IMPLEMENTED
Payment authorization abuse Replay or double-spend of A2A payment authorizations Atomic CONSUMED state on first verify, SHA-256 hashed storage, TTL expiry IMPLEMENTED

Status: IMPLEMENTED. The risk register is documented in Risk_Assessment_Framework.md.

6.2 AI Objectives and Planning to Achieve Them

Objective Metric Target Measurement Current
Supply integrity Supply audit delta 0.0 (zero tolerance) GET /api/v1/projections/supply-audit (60s cycle) 0.0
Reputation accuracy EigenTrust convergence < 0.001 delta between epochs EigenTrust worker logs, 1h epoch cycle Converging
Transfer reliability Success rate under load > 99.9% under sustained load Stress test suite (16-sim) 100% (0 errors)
Governance participation Proposal completion rate > 80% of proposals reach tally GET /api/v1/governance/proposals Tracked
Federation uptime Peer health check > 99% availability background_tasks/federation_sync.py + admin topology peer probes Monitored
AI fairness New agent trust score Non-zero starting share (uniform pre-trust) EigenTrust initialization logic Active

Status: IMPLEMENTED

6.3 Planning of Changes

Changes to AI components follow the established deployment pipeline:

  1. Code change on host filesystem (/opt/theprotocol/stack/Agentvault2/AgentVault/)
  2. A new revision of the registry’s Alembic migrations if the schema changes
  3. Frontend rebuild if UI affected (npm run build-only before Docker rebuild)
  4. Docker rebuild: docker compose -f docker-compose.production.yml --env-file .env.production up -d --build <service>
  5. Verification via live API call, container logs, or the full API test suite (650+ endpoints)
  6. Session log appended to sessionlogs2026/

Formal change management policy documented in Change_Management_Policy.md.

Status: IMPLEMENTED. The change process is documented in the published Change_Management_Policy.md and followed in practice (canary-first fleet rollouts, gated on health + version + supply Δ=0, with one-command rollback). A formal change advisory board is not applicable and is documented as such (sole maintainer).


Clause 7: Support

7.1 Resources

Resource Specification Evidence
Compute AMD Ryzen 9 3900 (12C/24T), 128GB ECC DDR4 Hetzner contract, PRODUCTION_DOCS/04_DEPLOYMENT/
Storage 2x1.92TB NVMe RAID-1 (hardware mirror) Hetzner contract
Network 1 Gbit/s, one public IPv4 address Hetzner contract
Containers The production frames (EU, ASIA) and cloud operators, each with its own registry, TEG and database, plus the frames’ EventStores, Redis, SPIRE, service agents and the monitoring stack docker ps output
Utilization High by design at fleet scale: RAM ~80-90% (working set + page cache), disk ~74%, load avg ~15-28 on 24 threads; capacity plan maintained Prometheus/Grafana dashboards, PRODUCTION_DOCS/10_SCALING_ARCHITECTURE/Capacity_Snapshot_2026-05-31.md
Database PostgreSQL, isolated instances (one per registry and per TEG, plus the frames’ EventStores, the sandboxes and monitoring) docker ps, compose files
Cache Redis per registry (each frame and each cloud operator runs its own instance) CLAUDE.md Part 9

Status: IMPLEMENTED

7.2 Competence

Area Competence Evidence
Full-stack development The sole maintainer built the entire platform: FastAPI backend, Vue 3 frontend, Docker orchestration
AI/ML engineering In-house implementation of EigenTrust++, ZKP, Bayesian Trust, Collusion Detection
Cryptography SPIFFE/SPIRE mTLS identity fabric, ZKP circuit design, SHA-256 token hashing
Infrastructure Docker estate of the production frames and cloud operators, nginx reverse proxy, Let’s Encrypt SSL, CrowdSec/fail2ban
Operational 16-hour sustained engineering sessions documented in session logs

Status: IMPLEMENTED

7.3 Awareness

Audience Awareness Mechanism Evidence
Operators 90+ files in PRODUCTION_DOCS/, 7 public documentation pages PRODUCTION_DOCS/ directory
Developers API reference (650+ endpoints across ~89 categories), SDK documentation, developer guides PRODUCTION_DOCS/05_API_REFERENCE/, PRODUCTION_DOCS/09_DEVELOPER_GUIDES/
AI components The operational doctrine (read before any code change) Kept outside the public repository
Public Landing page teasers, Services showcase (/ui#/services) Frontend views

Status: IMPLEMENTED

7.4 Communication

Channel Purpose Evidence
Alertmanager more than 90 vmalert alerting rules to SecurityAlerts@theprotocol.cloud PRODUCTION_DOCS/07_MONITORING/
WebSocket Live event broadcast to connected clients (Redis Pub/Sub cross-worker) Frontend WebSocket integration
Prometheus Metrics collection from all services prometheus.theprotocol.cloud
Grafana Visual dashboards for operational monitoring grafana.theprotocol.cloud
Session logs Append-only record of all engineering sessions sessionlogs2026/

Status: IMPLEMENTED

7.5 Documented Information

Document Category Location Count
System overview PRODUCTION_DOCS/01_SYSTEM_OVERVIEW/ Multiple
Architecture PRODUCTION_DOCS/02_ARCHITECTURE/ Multiple
Service documentation PRODUCTION_DOCS/03_SERVICES/ Per-service
Deployment guides PRODUCTION_DOCS/04_DEPLOYMENT/ Multiple
API reference PRODUCTION_DOCS/05_API_REFERENCE/ 90+ files
Operations runbooks PRODUCTION_DOCS/06_OPERATIONS/ Multiple
Monitoring PRODUCTION_DOCS/07_MONITORING/ Multiple
Security PRODUCTION_DOCS/08_SECURITY/ Multiple
Developer guides PRODUCTION_DOCS/09_DEVELOPER_GUIDES/ Multiple
Scaling architecture PRODUCTION_DOCS/10_SCALING_ARCHITECTURE/ Multiple
Compliance (this suite) PRODUCTION_DOCS/11_COMPLIANCE/ Complete, published in-app at /compliance
Operational doctrine CLAUDE.md 1 (continuously maintained)
Credentials reference PRODUCTION_DOCS/CREDENTIALS_REFERENCE.md 1

Status: IMPLEMENTED. Documented information is created, version-controlled (git), versioned (per-document version headers and changelogs), reviewed (multi-pass accuracy audits recorded in session logs) and published (rendered in-app at /compliance). Independent certification is a separate roadmap item, not a 7.5 control gap.


Clause 8: Operation

8.1 Operational Planning and Control

Control Implementation Evidence
Production manifest Each frame’s and cloud operator’s own compose file, with its own env file Each registry’s deployment directory
Migration versioning The registry’s Alembic migrations, applied at container startup AgentVault/agentvault_registry/alembic/versions/
Leader election Redis-based single-worker election across 70+ background workers and event reactors background_tasks/leader.py
Multi-worker safety Several worker processes per registry and per TEG, sharing state through Redis (on the production network: 4 gunicorn workers per frame registry and 8 per cloud operator registry; 4 uvicorn workers per frame TEG and 8 gunicorn workers per cloud operator TEG) Each frame’s and cloud operator’s compose file, gunicorn_conf.py
Rate limiting Redis-backed shared counters across workers Rate limiter middleware
Deployment guard Bare docker compose up intentionally fails (requires -f flag) CLAUDE.md Rule 12

Status: IMPLEMENTED

8.2 AI System Lifecycle

Detailed in companion document AI_System_Lifecycle.md. Summary:

AI Component Development Deployment Monitoring Retirement
EigenTrust++ In-house, networkx graph library Background task (1h epochs) Convergence metrics, Prometheus Version-gated via migration
ZKP Attestations In-house circuits, Barretenberg verifier API endpoints (/api/v1/attestations/*, /api/v1/circuits/*) Verification success/failure rates Circuit version and verification-key digest pinned
Collusion Detection In-house graph analysis Background task Detection event logging Threshold tuning via admin
Bayesian Trust In-house statistical model Per peer registry: updated on each cross-registry settlement and settled federation dispute, decayed daily for inactivity Per-peer discount values Discounts reset to 1.0 (AI_System_Lifecycle.md §7.9)
Federation Governance In-house proposal/vote/tally API endpoints (/governance/*) Proposal completion metrics Governed by own proposals

Status: IMPLEMENTED. The components in scope are operational, and the lifecycle (development, deployment, monitoring, retirement) is documented in the published AI_System_Lifecycle.md.

8.3 Data Management

Data Category Storage Protection Retention Evidence
Agent identities (DIDs) PostgreSQL (registry DBs) Unique constraints, DID format validation Permanent models.py
Unit balances PostgreSQL (TEG DBs) Atomic transactions, supply auditor Permanent TEG endpoints_token.py
Event ledger PostgreSQL (EventStore DB) Append-only, idempotency keys, RAID-1 Permanent (immutable) agentvault-event-store/
Agent credentials PostgreSQL bcrypt-hashed client_secret and API keys (an API key also carries a SHA-256 lookup hash) Permanent (hashed) routers/auth_secure.py
Payment authorizations PostgreSQL SHA-256 hashed, atomic state transitions, TTL expiry Until consumed/expired routers/a2a_payment.py
Session data Redis Per-frame instance isolation with per-service logical DBs, volatile Session-scoped Redis config
Production secrets Each registry’s own env file Not committed to source control, readable only by its owner (mode 600) N/A .gitignore

Detailed in companion document Data_Protection_Policy.md.

Known limitation, stated plainly: block-level disk encryption at rest is not enabled: data protection at rest relies on root-only file permissions, container isolation and hashed credential storage (risk register R-OP-08; see Data_Protection_Policy.md §7.3 Transfer Safeguards).

Status: IMPLEMENTED (with the encryption-at-rest limitation documented above)

8.4 Third-Party AI Components

TheProtocol uses no third-party AI models or AI-as-a-service. All of its AI components are built in-house.

External Dependency Type Purpose Risk Assessment
Barretenberg ZKP verifier binary (Aztec Protocol) Proof verification for ZKP attestations Open-source, deterministic, version-pinned
networkx Python graph library Graph operations for EigenTrust/Collusion Open-source, well-audited, no AI decision-making
PostgreSQL Database Data persistence Not an AI component
Redis Cache/coordination Leader election, rate limiting Not an AI component

Status: IMPLEMENTED. Minimal third-party AI surface area by design.


Clause 9: Performance Evaluation

9.1 Monitoring, Measurement, Analysis, and Evaluation

Monitoring System Scope Frequency Evidence
Supply Auditor Unit supply invariant (delta = 0.0) Every 60 seconds background_tasks/supply_auditor.py, GET /api/v1/projections/supply-audit
EigenTrust Worker Reputation score convergence 1-hour epochs background_tasks/eigentrust_worker.py
Prometheus Infrastructure and application metrics 15s scrape interval prometheus.theprotocol.cloud
Grafana Visual dashboards (CPU, memory, request rates, error rates) Real-time grafana.theprotocol.cloud
Alertmanager more than 90 alerting rules (vmalert), email to SecurityAlerts@theprotocol.cloud Continuous Alert rule configs
Agent Health Checker + Federation Sync Agent endpoint health; federation peer availability and card sync Background task cycles background_tasks/agent_health_checker.py, background_tasks/federation_sync.py
Velocity Snapshot Transaction velocity for dynamic fee scaling Background task cycle background_tasks/velocity_snapshot.py (referenced in CLAUDE.md)
API Test Suite 650+ endpoints across ~89 categories On-demand API Tester
Stress Test ~275 tx/s intra-registry, ~114 tx/s end-to-end, 16 concurrent simulations On-demand PRODUCTION_DOCS/06_OPERATIONS/Stress_Test_And_Simulation.md

Status: IMPLEMENTED

9.2 Internal Audit

Internal audit procedures are documented in Compliance_Audit_Procedures.md. Current practice:

  • Supply invariant audit runs continuously (60s cycle) with zero-tolerance breach detection
  • The full API test suite (650+ endpoints across ~89 categories) is required to pass
  • Stress tests verify system behavior under load (~275 tx/s intra-registry, ~114 tx/s end-to-end)
  • Session logs provide complete audit trail of all engineering changes
  • CrowdSec, fail2ban, and Lynis provide security audit capability

Status: IMPLEMENTED. The audit programme and schedule are documented in the published Compliance_Audit_Procedures.md (§2 continuous automated, §3 daily, §4 weekly, §5 monthly, §6 quarterly) and practiced: recent deep-audit records include the user-funds conservation audit (2026-06-09), the shadow-bookkeeping audit (2026-06-10) and the compliance-documentation accuracy audit (2026-06-12), all in sessionlogs2026/.

9.3 Management Review

Review Activity Frequency Mechanism Evidence
Supply audit review Continuous (automated) Supply auditor background task Alertmanager breach notifications
Governance proposal review Per-proposal Commander reviews all proposals, veto power routers/governance.py
Federation health review Continuous (automated) Peer health checker + federation sync Background task logs
Security review On-demand CrowdSec alerts, fail2ban status, Lynis audit monitor-security.sh
Performance review On-demand Grafana dashboards, Prometheus queries grafana.theprotocol.cloud

Status: IMPLEMENTED. Automated reviews active; formal management review meeting cadence not applicable (sole maintainer).


Clause 10: Improvement

10.1 Nonconformity and Corrective Action

Incident classification and response procedures are documented in Incident_Response_Procedures.md. Summary:

Priority Definition Response Time Example
P1 (Critical) Supply invariant breach, data loss, complete service outage Immediate Supply audit BREACH status
P2 (High) AI component failure, federation desync, payment system error < 1 hour EigenTrust convergence failure
P3 (Medium) Single agent service outage, non-critical background task failure < 4 hours Service agent container crash
P4 (Low) Documentation gap, cosmetic UI issue, non-critical metric anomaly < 24 hours Grafana dashboard misconfiguration

Corrective action workflow: Detection (automated monitoring) -> Investigation (container logs, DB queries) -> Fix (code change, migration) -> Verification (live API call, test suite) -> Documentation (session log).

Status: IMPLEMENTED. Procedures are documented in the published Incident_Response_Procedures.md (P1-P4 classification, containment, post-incident review) and exercised in practice (incident records in sessionlogs2026/).

10.2 Continual Improvement

Improvement Mechanism Implementation Evidence
Stress testing Validated to ~275 tx/s intra-registry and ~114 tx/s end-to-end under 16 concurrent simulations PRODUCTION_DOCS/06_OPERATIONS/Stress_Test_And_Simulation.md
Governance proposals Parameter-change proposals recorded in the immutable event ledger, with agent voting routers/governance.py
Migration versioning Incremental schema evolution (the registry’s Alembic migrations) Alembic versions directory
Feature flags Emission policies (32 types), projection snapshots toggle event_emission_policies table, ENABLE_PROJECTION_SNAPSHOTS env var
Session logging Every engineering session documented with changes and outcomes sessionlogs2026/

Status: IMPLEMENTED. Improvement mechanisms are active and the review cycle is documented (Compliance_Audit_Procedures.md §6 quarterly audits and §12 policy review). Practice exceeds the documented cadence: every engineering session produces logged improvements (sessionlogs2026/, CLAUDE.md changelog).


Gap Analysis Summary

Clause Requirement Status Evidence Gap
4.1 Organization context IMPLEMENTED CLAUDE.md, this document None
4.2 Interested parties IMPLEMENTED Stakeholder analysis above None
4.3 AIMS scope IMPLEMENTED 6 AI components documented None
4.4 AIMS documentation IMPLEMENTED PRODUCTION_DOCS/, 11_COMPLIANCE/ suite published None
5.1 Leadership commitment IMPLEMENTED Commander sole authority None
5.2 AI policy IMPLEMENTED Runtime invariants, emission policies None
5.3 Roles and responsibilities IMPLEMENTED 4-tier auth model None
6.1 Risk assessment IMPLEMENTED Published risk register (Risk_Assessment_Framework.md) + active mitigations None
6.2 AI objectives IMPLEMENTED Measurable metrics with targets None
6.3 Planning of changes IMPLEMENTED Published Change_Management_Policy.md + gated canary-first rollouts None
7.1 Resources IMPLEMENTED Hetzner dedicated server, production frames and cloud operators None
7.2 Competence IMPLEMENTED Sole maintainer’s full-stack expertise None
7.3 Awareness IMPLEMENTED 90+ documentation files, public pages None
7.4 Communication IMPLEMENTED Alertmanager, WebSocket, Prometheus None
7.5 Documented information IMPLEMENTED Version-controlled, versioned, published in-app None
8.1 Operational planning IMPLEMENTED Docker Compose, migrations, leader election None
8.2 AI system lifecycle IMPLEMENTED Published AI_System_Lifecycle.md; all components operational None
8.3 Data management IMPLEMENTED Per-category protections documented None
8.4 Third-party AI IMPLEMENTED Minimal: Barretenberg, networkx only None
9.1 Monitoring IMPLEMENTED 9 monitoring systems active None
9.2 Internal audit IMPLEMENTED Published audit programme (Compliance_Audit_Procedures.md §§2-6) + audit records None
9.3 Management review IMPLEMENTED Automated + Commander review None
10.1 Nonconformity IMPLEMENTED Published Incident_Response_Procedures.md + incident records None
10.2 Continual improvement IMPLEMENTED Documented review cycle + per-session improvement records None

Summary: 22 IMPLEMENTED, 0 PARTIAL, 0 MISSING (self-assessed). Outstanding items sit outside the clause table: independent third-party certification (roadmap) and the disk-encryption-at-rest hardening gap (risk register R-OP-08, documented under 8.3).


Cross-References

Companion Documents (11_COMPLIANCE/)

Document Purpose Status
ISO_42001_Alignment.md This document: master alignment Published v2.3.2
Risk_Assessment_Framework.md AI-specific risk register (Clause 6.1) Published v2.2.3
AI_System_Lifecycle.md Lifecycle management for 6 AI components (Clause 8.2) Published v2.2.1
Data_Protection_Policy.md Data classification and protection (Clause 8.3) Published v2.2.3
Change_Management_Policy.md Change advisory and deployment process (Clause 6.3) Published v2.1.2
Compliance_Audit_Procedures.md Internal audit schedule and procedures (Clause 9.2) Published v2.2.2
Incident_Response_Procedures.md P1-P4 incident classification and response (Clause 10.1) Published v2.1.2
Jurisdiction_Profiles.md Declared, enforced and attested feature scope per deployment Published (revised 2026-10-03)

All eight documents are rendered in-app at /compliance (public).

Key PRODUCTION_DOCS References

Document Relevance
PRODUCTION_DOCS/01_SYSTEM_OVERVIEW/ Organization context (Clause 4)
PRODUCTION_DOCS/02_ARCHITECTURE/ AI component architecture (Clauses 4.3, 8.2)
PRODUCTION_DOCS/03_SERVICES/ Service documentation (Clauses 7.3, 8.1)
PRODUCTION_DOCS/04_DEPLOYMENT/ Infrastructure resources (Clause 7.1)
PRODUCTION_DOCS/05_API_REFERENCE/ API documentation, 90+ files (Clause 7.5)
PRODUCTION_DOCS/06_OPERATIONS/ Operational procedures, stress testing (Clauses 8.1, 10.2)
PRODUCTION_DOCS/07_MONITORING/ Monitoring configuration (Clause 9.1)
PRODUCTION_DOCS/08_SECURITY/ Security controls (Clauses 5.2, 6.1)
PRODUCTION_DOCS/CREDENTIALS_REFERENCE.md Auth tier documentation (Clause 5.3)

Key Source Files

File Relevance
CLAUDE.md Operational doctrine: rules, architecture, credentials (Clauses 4-5)
Each frame’s and cloud operator’s compose file Production container manifests (Clause 8.1)
AgentVault/agentvault_registry/src/agentvault_registry/event_emitter.py Emission policy gate (Clause 5.2)
AgentVault/agentvault_registry/src/agentvault_registry/background_tasks/supply_auditor.py Supply invariant audit (Clause 9.1)
AgentVault/agentvault_registry/src/agentvault_registry/background_tasks/eigentrust_worker.py EigenTrust AI component (Clause 8.2)
AgentVault/agentvault_registry/src/agentvault_registry/background_tasks/collusion_worker.py Collusion detection AI component (Clause 8.2)
AgentVault/agentvault_registry/src/agentvault_registry/routers/governance.py Governance AI component (Clause 8.2)
agentvault-event-store/src/services/projections.py EventStore projections, supply audit endpoint (Clause 9.1)

Key API Endpoints

Endpoint Relevance
GET /api/v1/projections/supply-audit Supply invariant verification (Clause 9.1)
GET /api/v1/admin/event-emission-policies Emission policy management (Clause 5.2)
POST /api/v1/admin/projection-snapshot Projection baseline snapshots (Clause 9.1)
GET /api/v1/governance/proposals Governance monitoring (Clause 9.3)

End of ISO/IEC 42001:2023 Alignment Document TheProtocol: Sovereign AI Agent Orchestration Platform Assessment: 22/22 sub-clauses IMPLEMENTED (self-assessed; independent certification pending; known hardening gap: disk encryption at rest, R-OP-08)


v2.0.0 (2026-06-12): accuracy overhaul — corrected topology (16 federated registries, ~280 containers), API/stress/alert figures, file paths, the ZKP and lock mechanisms, and the self-assessment framing to the live system.

v2.1.0 (2026-06-12): false-gap audit — all 8 former PARTIAL sub-clauses upgraded to IMPLEMENTED: each cited as its gap a “formal document” that this suite itself published (6.1, 6.3, 8.2, 10.1), an audit schedule that Compliance_Audit_Procedures.md §§2-6 already defines (9.2, 10.2), or conflated independent certification with a documentation requirement (4.4, 7.5). Also corrected stale infrastructure figures (utilization, container/instance counts, worker model — gunicorn-supervised since V0041/V0049) and a non-existent file reference (peer_health_checker.py → agent_health_checker.py/federation_sync.py). The remaining honest caveats: self-assessed (no independent audit yet) and no block-level disk encryption at rest (risk register R-OP-08). Not legal advice.

v2.2.0 (2026-10-03): wording follows the live network and the agreed public terms: the platform’s units named as units (ECU on EU, ACU on ASIA), the production network named as the EU and ASIA frames with four cloud operators, votes counted by membership or by locks as each registry is set, and every surface that is switched off on the production frames named as switched off; the companion table carries the current versions and lists the eighth document. Self-assessed; not independently audited.

v2.3.0 (2026-10-13): the ZKP rows name UltraHonk, the four v2.0.0 policy circuits and the real endpoints. Self-assessed; not independently audited.

v2.3.1 (2026-10-16): the text without em dashes (the gap table’s empty cells read None and its empty sub-clause column is gone); the maintainer named as the sole maintainer, not a founder; the production topology, manifests, worker counts and secrets named as the EU and ASIA frames and their four cloud operators run them, in place of the retired first frames’ files; the Sybil mitigation named as volume-weighted trust (EigenTrust weights by transaction volume, and locks are switched off on the production frames); the retired PPAO dashboard removed from the public channels; the companion table carries the current versions. Self-assessed; not independently audited.

v2.3.2 (2026-10-17): the Bayesian trust component described as the released software runs it (a discount for each peer registry, weighting its federation votes and read by the peer trust policy), and new-agent fairness as EigenTrust++'s uniform pre-trust; the operational doctrine named as kept outside the public repository; schema versions named as the registry’s Alembic migrations; the credential hashes as the released software keeps them; the encryption-at-rest reference points at the Data Protection Policy’s §7.3; the companion table carries the current versions. Self-assessed; not independently audited.

v2.3.3 (2026-10-17): the alerting rules counted as vmalert evaluates them (more than 90; Prometheus only scrapes) in the stakeholder table and the two monitoring rows. Self-assessed; not independently audited.