Developer Governance Instrumentation Guide -- Witnessing How You Govern

Who this is for: Engineering teams, DevOps, platform teams, and MLOps engineers preparing AI systems for assessment under CMMC, EU AI Act, ISO 42001, SR 11-7, or NIST AI RMF. No changes to your governance structure are required.

Key principle

You do not need to change how you govern. You need to witness how you govern. Your governance policies, review cadences, and approval workflows stay exactly as they are. SWT3 makes them provable.

1. Quick Start 2. Core Governance Events 3. Human Oversight Events 4. Governance Infrastructure 5. Framework-Specific Checklists 6. Integration Patterns 7. Related Guides

1. Quick Start -- Your First Governance Anchor

Five lines of code to prove your governance framework is operational:

Python

from swt3_ai import Witness

witness = Witness(tenant_id="your-tenant-id")
witness.witness_governance_framework(
    controls_defined=24,
    controls_active=22
)

TypeScript

import { Witness } from "@tenova/swt3-ai";

const witness = new Witness({ tenantId: "your-tenant-id" });
witness.witnessGovernanceConfig({
    rules: [{ id: "GOV-001", expression: "active" }],
    governanceVersion: 1
});

That call mints an AI-GOV.1 anchor recording that your governance framework has 24 controls defined and 22 active. When an assessor reviews your evidence chain, they see cryptographic proof that governance was operational -- not a screenshot, not a self-attestation, but a tamper-evident record with a verifiable fingerprint.

SDK note

Python SDK class is Witness (not SWT3Witness). Import: from swt3_ai import Witness. The flush_interval parameter is in seconds (float), not milliseconds. All governance methods are available in Python SDK v0.6.0+. TypeScript SDK has a subset; check the SDK docs for availability.

2. Core Governance Events

These methods witness the governance decisions your organization already makes. Call them at the point where the decision is finalized -- after approval, not before.

2.1 Policy Framework Attestation (AI-GOV.1)

Witness that your governance framework is active and operational. Call periodically (daily or weekly) or after any framework change.

witness.witness_governance_framework(
    controls_defined=24,      # Total controls in your framework
    controls_active=22,       # Controls currently enforced
    framework_version="2.1"   # Your internal version identifier
)

When to call: After framework updates, during scheduled governance reviews, or as a periodic heartbeat.

2.2 Governance Review Completion (AI-GOV.2)

Witness that scheduled governance reviews were conducted.

witness.witness_governance_review(
    reviews_scheduled=4,      # Reviews scheduled in period
    reviews_completed=4,      # Reviews actually completed
    review_period_days=90     # Review cycle length
)

When to call: After each governance review meeting concludes.

2.3 Policy Update (AI-GOV.4)

Witness that a governance policy was updated through your change control process.

witness.witness_governance_update(
    previous_version=3,       # Version before change
    current_version=4,        # Version after change
    change_summary="Updated data retention thresholds per Q3 risk review"
)

When to call: After a policy update is approved and published. Not during drafting.

2.4 Accountability Assignment (AI-GOV.5)

Witness that accountability roles are assigned and acknowledged by responsible personnel.

witness.witness_governance_accountability(
    roles_assigned=6,         # Total accountability roles
    roles_acknowledged=6,     # Roles with signed acknowledgment
    responsible_party="Chief AI Officer"
)

When to call: After role assignments are updated or acknowledged. Typically quarterly or after organizational changes.

2.5 Delegation Tree (AI-DEL.1)

Witness authority delegation from one entity to another with defined scope and depth.

witness.witness_delegation_tree(
    delegator_id="platform-admin",
    scope="model-deployment:production",
    delegation_depth=2        # How many levels deep
)

When to call: When permissions are delegated between teams, agents, or services. Especially important for multi-agent architectures.

2.6 Governance Change Authorization (AI-METAGOV.5)

Witness that a governance configuration change was authorized by an identified operator.

import hashlib

credential_hash = hashlib.sha256(b"operator-session-token").hexdigest()

witness.authorize_governance_change(
    scope_domain="global",           # Scope: global, model, tenant, agent
    permission_level="admin",        # Level: viewer, operator, admin, superadmin
    operator_id="jane.doe@company.com",
    change_description="Enable production guardrails v2",
    operator_credential_hash=credential_hash
)

When to call: Before applying any governance configuration change. The anchor proves who authorized what.

2.7 Emergency Override (AI-METAGOV.6)

Witness that an emergency governance override was invoked. Creates a mandatory review obligation.

witness.witness_emergency_override(
    override_reason="safety_incident",  # Reason code
    review_window_hours=24,             # Mandatory review deadline
    operator_id="oncall@company.com",
    change_description="Disabled model output pending safety review"
)

When to call: Immediately when an emergency override occurs. The review window creates a provable obligation to follow up.

2.8 Consent Collection (AI-CONSENT.1)

Witness that data subject consent was collected under a specified legal basis.

witness.witness_consent(
    subjects_covered=1500,           # Number of subjects
    legal_basis_type="explicit",     # GDPR legal basis
    withdrawal_available=True        # Withdrawal mechanism exists
)

When to call: After consent collection campaigns complete or when consent status changes materially.

3. Human Oversight Events

Human oversight anchors are typically minted as part of the inference witnessing pipeline rather than standalone calls. When a human reviews, overrides, or approves an AI decision, the witness captures it automatically if the right parameters are set.

The key procedures:

AI-HITL.1 -- Human review occurred before output release. Captured when the inference pipeline includes a human approval step.
AI-HITL.2 -- Human override of AI recommendation. Captured when a human changes the AI's output.
AI-HITL.3 -- Overseer identity cryptographically bound to review event. Captured when the reviewer authenticates through a verified identity method.

For standalone human oversight witnessing outside the inference pipeline, use the governance escalation method:

witness.witness_governance_escalation(
    escalation_paths_defined=3,    # Paths to human decision-makers
    escalation_paths_tested=3,     # Paths verified as functional
    last_test_date="2026-08-01"
)

See the SDK documentation for full HITL integration with the wrap() proxy pattern.

4. Governance Infrastructure (Platform Teams)

These methods are for platform teams managing governance infrastructure itself -- the systems that enforce governance policies.

4.1 Governance Configuration Attestation (AI-METAGOV.1)

Self-attest that governance rules are loaded and active. This is recursive governance -- the governance system witnessing its own state.

witness.witness_governance_config(
    rules=[
        {"id": "GRD-001", "expression": "toxicity < 0.8"},
        {"id": "GRD-002", "expression": "pii_detected == False"},
    ],
    governance_version=5,
    operator_id="platform-service"
)

4.2 Governance Layer Registration (AI-METAGOV.2)

witness.register_governance_layer(
    layer_id="content-filter-v3",
    config_hash="a1b2c3d4e5f6...",
    stack_position=2             # Position in the enforcement stack
)

4.3 Policy Downgrade Detection (AI-METAGOV.3)

Detect and witness when a policy version decreases. This provides monotonic enforcement -- downgrades are recorded even if they were intentional.

import hashlib

content_hash = hashlib.sha256(policy_content.encode()).hexdigest()
result = witness.check_policy_downgrade(
    policy_version=4,
    policy_content_hash=content_hash,
    strict=True                  # Raise exception on downgrade
)

4.4 Federation Sync (AI-METAGOV.7)

Witness policy divergence between federated governance configurations.

witness.witness_governance_sync(
    divergence_type="version_mismatch",
    local_policy_hash="abc123...",
    remote_policy_hash="def456...",
    remote_tenant_id="partner-org"
)

5. Framework-Specific Checklists

Preparing for a specific assessment? These checklists identify the governance events you should be witnessing. Check each box as you instrument the corresponding call in your codebase.

Preparing for CMMC Level 2? Witness these 8 governance events:

AI-GOV.1 Governance framework active with controls defined and enforced
AI-GOV.5 Accountability roles assigned and acknowledged
AI-ACC.1 Access control scopes enforced (least privilege)
AI-AUDIT.1 Audit log integrity verified
AI-HITL.1 Human review before critical decisions
AI-DEL.1 Authority delegation with defined scope
AI-METAGOV.5 Governance changes authorized by identified operators
AI-GOV.4 Policy updates tracked with version control

Preparing for EU AI Act conformity assessment? Witness these 12 governance events:

AI-GOV.1 Risk management system active (Art. 9)
AI-GOV.2 Governance reviews completed on schedule (Art. 9)
AI-HITL.1 Human oversight measures operational (Art. 14)
AI-HITL.3 Overseer identity bound to review events (Art. 14)
AI-DRIFT.1 Continuous monitoring against performance thresholds (Art. 15)
AI-CONSENT.1 Data subject consent with withdrawal mechanism (Art. 10)
AI-EXPL.1 Explainability mechanisms active (Art. 13)
AI-FAIR.1 Bias detection and mitigation operational (Art. 10)
AI-AUDIT.1 Audit log integrity for technical documentation (Art. 12)
AI-METAGOV.6 Emergency override with mandatory review (Art. 14)
AI-GOV.4 Post-market monitoring policy updates (Art. 72)
AI-EMRG.1 Emergency override lifecycle management (Art. 14)

Preparing for ISO/IEC 42001 certification? Witness these 10 governance events:

AI-GOV.1 AI management system framework active (Clause 5 -- Leadership)
AI-GOV.2 Management reviews completed (Clause 9.3)
AI-GOV.5 Roles and responsibilities assigned (Clause 5.3)
AI-GOV.6 Escalation paths defined and tested (Clause 6.1)
AI-METAGOV.1 Governance configuration attested (Annex A.2)
AI-DRIFT.1 Performance monitoring operational (Clause 9.1)
AI-FAIR.1 Bias and fairness controls active (Annex A.10)
AI-CONSENT.1 Data management and consent (Annex A.7)
AI-GOV.4 Continual improvement records (Clause 10.1)
AI-AUDIT.1 Internal audit integrity (Clause 9.2)

Preparing for SR 11-7 examination? Witness these 6 governance events:

AI-GOV.1 Model risk management framework active (Section III)
AI-HITL.3 Effective challenge -- reviewer identity and independence (Section V)
AI-DRIFT.1 Model performance monitoring (Section VI)
AI-GOV.2 Model validation reviews completed (Section V)
AI-GOV.5 Board and senior management oversight documented (Section IV)
AI-METAGOV.6 Override governance with mandatory review (Section VII)

6. Integration Patterns

Pattern A: Inline Witnessing

Call witness methods directly in your governance workflows. Best for teams that own the governance code.

# In your policy approval handler:
def approve_policy(policy_id, approver, version):
    # Your existing approval logic
    db.update_policy_status(policy_id, "approved")

    # Add governance witnessing
    witness.witness_governance_update(
        previous_version=version - 1,
        current_version=version,
        change_summary=f"Policy {policy_id} approved by {approver}"
    )

Pattern B: Event-Driven Witnessing

Hook into existing ITSM or GRC tool webhooks. Best for teams using ServiceNow, Jira, or similar platforms.

# Webhook handler for your GRC platform events:
@app.post("/webhooks/grc")
def handle_grc_event(event):
    if event["type"] == "risk_acceptance":
        witness.witness_governance_framework(
            controls_defined=event["total_controls"],
            controls_active=event["accepted_controls"]
        )
    elif event["type"] == "delegation_change":
        witness.witness_delegation_tree(
            delegator_id=event["delegator"],
            scope=event["scope"],
            delegation_depth=event["depth"]
        )

Pattern C: Batch Witnessing

Retroactively witness existing governance records. Best for initial onboarding when you have historical governance data.

# Import historical governance reviews from your records:
for review in historical_reviews:
    witness.witness_governance_review(
        reviews_scheduled=review["scheduled"],
        reviews_completed=review["completed"],
        review_period_days=review["period"]
    )

# Flush all records at once:
witness.flush()

Clearing level guidance

Governance events should typically use Clearing Level 0 (Analytics) or Level 1 (Standard). Governance metadata is organizational, not personally identifiable. Reserve Level 2 (Sensitive) and Level 3 (Classified) for inference events containing PII or protected data.

Temporal proof and non-repudiation

Every governance witness anchor you mint is included in the daily Merkle rollup. That rollup is automatically submitted to an independent RFC 3161 Timestamping Authority (DigiCert, Sectigo, or FreeTSA) and signed with ML-DSA-65 (FIPS 204) post-quantum signatures. This means your governance evidence chain has four layers of temporal proof: (1) per-anchor epoch timestamp, (2) daily Merkle root binding all anchors, (3) RFC 3161 external timestamp from a trusted third party, and (4) ML-DSA-65 post-quantum signature on the root. For AU-10 non-repudiation and EU AI Act Art. 12 logging requirements, this provides independently verifiable proof that your governance events occurred when claimed -- without relying on the SWT3 platform itself.

7. Related Guides

For Your Assessor

Governance Assessment Workflow -- The 4-step methodology your assessor will follow when reviewing your governance evidence
Assessor Evidence Matrix -- What coverage SWT3 provides per procedure
Assessor Onboarding Kit -- Share this with your assessor before assessment day

SDK and Integration

SDK Documentation -- Full API reference for Python, TypeScript, and MCP
AI Witness Quick Start -- Getting started with inference witnessing
CI/CD Integration Guide -- Automated witnessing in deployment pipelines

Governance Instrumentation Guide

Contents

1. Quick Start -- Your First Governance Anchor

Python

TypeScript

2. Core Governance Events

2.1 Policy Framework Attestation (AI-GOV.1)

2.2 Governance Review Completion (AI-GOV.2)

2.3 Policy Update (AI-GOV.4)

2.4 Accountability Assignment (AI-GOV.5)

2.5 Delegation Tree (AI-DEL.1)

2.6 Governance Change Authorization (AI-METAGOV.5)

2.7 Emergency Override (AI-METAGOV.6)

2.8 Consent Collection (AI-CONSENT.1)

3. Human Oversight Events

4. Governance Infrastructure (Platform Teams)

4.1 Governance Configuration Attestation (AI-METAGOV.1)

4.2 Governance Layer Registration (AI-METAGOV.2)

4.3 Policy Downgrade Detection (AI-METAGOV.3)

4.4 Federation Sync (AI-METAGOV.7)

5. Framework-Specific Checklists

Preparing for CMMC Level 2? Witness these 8 governance events:

Preparing for EU AI Act conformity assessment? Witness these 12 governance events:

Preparing for ISO/IEC 42001 certification? Witness these 10 governance events:

Preparing for SR 11-7 examination? Witness these 6 governance events:

6. Integration Patterns

Pattern A: Inline Witnessing

Pattern B: Event-Driven Witnessing

Pattern C: Batch Witnessing

7. Related Guides

For Your Assessor

SDK and Integration

Framework-Specific