Audience: Notified Bodies, conformity assessment bodies, C3PAOs, third-party auditors, certification body leads. Proficiency: Expert. Time: 12 minutes.

Contents

1. The Infrastructure Evidence Problem 2. Four Governance Domains 3. Change Management Evidence 4. Disaster Recovery Test Evidence 5. Integrity Monitoring Evidence 6. Supply Chain Assurance Evidence 7. Regulatory Framework Mapping 8. Verifying Witness Evidence 9. Assessment Protocol 10. Related Guides

1. The Infrastructure Evidence Problem

Most organizations can describe their infrastructure governance policies. Fewer can prove those policies were followed for a specific change, a specific recovery test, or a specific supply chain event.

The gap is not in policy. It is in evidence. When an assessor asks "show me proof that your last disaster recovery test completed all four phases before you declared success," the typical answer is a spreadsheet with manual timestamps and a sign-off email. When they ask "prove this patch went through staging before production," the answer is a ticket with status transitions.

Neither of these is independently verifiable. Neither is tamper-evident. Neither proves sequence or completeness.

SWT3 witness anchors close this gap. Each governance decision, each phase completion, each test result mints a cryptographic receipt. These receipts are organized into hierarchical Merkle trees that prove temporal ordering: Phase 2 cannot exist without Phase 1. The tree root is independently timestamped via RFC 3161, creating evidence that is verifiable without trusting the organization that produced it.

Independent verification without trust. Every witness receipt is Ed25519-signed by a resolvable DID (did:web:sovereign.tenova.io) and rolled into a daily Merkle tree with an RFC 3161 timestamp. The assessor can recompute any fingerprint, walk any inclusion proof, and verify any signature without access to the organization's systems. The evidence is self-proving.

Assessor note

This guide maps infrastructure governance to SWT3 procedures and regulatory requirements. It does not prescribe how organizations should implement their governance. SWT3 is a witness layer: it records what happened, it does not decide what should happen.

2. Four Governance Domains

Infrastructure governance evidence falls into four domains. Each domain has distinct regulatory drivers, distinct evidence requirements, and distinct failure modes that assessors encounter during conformity assessment.

DomainCore QuestionPrimary ProceduresEvidence Type
Change ManagementDid this change follow governance gates before reaching production?CM-3.1, AI-DRIFT.1, AI-DRIFT.2, AI-FREEZE.1Phase-gated Merkle tree
Disaster RecoveryDid this DR test complete all phases successfully?CP-4.1, AI-ROBUST.1, AI-ASSESS.1, AI-RECOMM.1Phase-gated Merkle tree
Integrity MonitoringCan you prove system integrity was continuously monitored?SI-7.1, SI-7.2, AI-DRIFT.1, AI-DRIFT.2Continuous witness chain
Supply ChainCan you prove component provenance and trust verification?AI-SUPPLY.1, AI-SBOM.1, AI-HW.1, AI-TRUST.1Per-component attestation

3. Change Management Evidence

A change management pipeline witnessed by SWT3 produces a Merkle tree with one branch per phase. The assessor can verify that each phase completed before the next began, and that drift thresholds were respected at each gate.

What to request from the provider

  1. Structure ID for the change (e.g., release-v2.14.0). This identifies the Merkle tree.
  2. Phase gate proof for each phase. The /api/v1/merkle/gate endpoint returns whether all required milestones were satisfied before the gate opened.
  3. Drift attestation between staging and production. The AI-DRIFT.1 witness anchor should show drift score below the organization's stated threshold.
  4. Freeze attestation before production deployment. The AI-FREEZE.1 anchor should precede the production rollout anchor in the tree.

Assessor note

The most common finding: organizations skip the canary phase under time pressure. If the Merkle tree has leaves in Phase 0 (test) and Phase 3 (production) but nothing in Phase 2 (canary), that is a nonconformity against Art.17 QMS requirements.

Regulatory drivers

DORA Art.8 note

For financial entities, the ICT risk management framework must include change management procedures (Art.8(4)). The structure root for each deployment can be included in the entity's ICT risk management report submitted to competent authorities. The drift attestations between phases prove that impact assessment occurred before production deployment, not after.

4. Disaster Recovery Test Evidence

A DR test witnessed by SWT3 produces a phase-gated Merkle tree covering backup, restore, validation, and cutover. Each phase contains milestones with specific evidence leaves.

What to request from the provider

  1. DR test structure ID (e.g., dr-test-q4-2026). One tree per test execution.
  2. Backup verification leaf: proves backup integrity was checked before restore began.
  3. Restore completion leaf: AI-ROBUST.1 anchor with service count and success rate.
  4. Champion-challenger comparison: AI-ASSESS.1 anchor comparing DR environment to production baseline. Look for functional parity percentage.
  5. Cutover authorization: AI-RECOMM.1 anchor with validator count and RTO attestation. This leaf must be the last in the tree (temporal ordering).

Assessor note

Request both PASS and FAIL anchors. A DR test that never fails is suspicious. Legitimate DR programs generate FAIL verdicts when RTO is exceeded or services fail to restore. The absence of any FAIL history across multiple test cycles is itself a finding worth investigating.

Regulatory drivers

EU AI Act Annex VI note

During conformity assessment of high-risk AI systems, the NB evaluates whether the provider's risk management system (Art.9) includes tested resilience measures. The DR test structure root is a single artifact the assessment team can verify independently. Resolve the DID, check the Ed25519 signature, walk the phase proofs. No reliance on the provider's internal systems required.

5. Integrity Monitoring Evidence

Continuous integrity monitoring produces a chain of witness anchors over time, not a single tree. The assessor verifies that the chain has no gaps longer than the organization's stated monitoring interval.

What to request from the provider

  1. SI-7.1 witness chain: File integrity monitoring attestations. Each anchor records the number of files checked and any modifications detected.
  2. SI-7.2 witness chain: Package verification attestations. Each anchor records installed packages verified against known-good hashes.
  3. AI-DRIFT.1 continuous anchors: Model and system drift detection over time. Look for regularity: if the organization claims hourly monitoring, the anchors should appear approximately hourly.
  4. Chain gap analysis: Query the ledger for time gaps exceeding the stated monitoring interval. Gaps indicate monitoring failures.

Assessor note

Integrity monitoring evidence is the easiest domain to verify at scale. Query the witness ledger for procedure IDs SI-7.1 and AI-DRIFT.1, sort by timestamp, and check for gaps. The daily Merkle rollup at /api/v1/merkle provides a per-day summary: if a day has zero leaves for integrity procedures, that is a gap.

Regulatory drivers

6. Supply Chain Assurance Evidence

Supply chain governance produces per-component attestations. Each hardware component, software dependency, and AI model has its own provenance record.

What to request from the provider

  1. AI-SBOM.1 attestation: The AI bill of materials with component hashes. Compare against the organization's declared component inventory.
  2. AI-SUPPLY.1 attestation: Supply chain risk assessment with supplier compliance scores. Look for the suppliers_assessed and suppliers_compliant factors.
  3. AI-HW.1 attestation: Hardware runtime attestation. For AI-specific infrastructure, this proves the hardware matches the declared configuration (GPU type, firmware version, driver version).
  4. AI-TRUST.1 verification: Trust verification between components. In multi-vendor environments, each vendor's components should have trust attestations linking them to the overall system.

Assessor note

Supply chain evidence is the most framework-dense domain. AI-SBOM.1 alone maps to 11 regulatory frameworks. When assessing EU AI Act compliance, cross-reference the SBOM attestation against Art.60(a) registration database requirements and the G7 AI-SBOM minimum elements.

Regulatory drivers

7. Regulatory Framework Mapping

The following table maps infrastructure governance domains to specific regulatory articles. This is the reference for scoping a conformity assessment.

FrameworkChange MgmtDR TestingIntegritySupply Chain
EU AI ActArt.17(1)(d)Art.9(2)(a)Art.15(1)Art.60(a)
ISO 42001Cl.8.1A.6.2.6A.6.2.3A.6.2.5
EU Machinery RegArt.18Annex III 1.2.1Art.18Art.10(2)
DORAArt.8(4)Art.11(6)Art.9(2)Art.28
NIST 800-53CM-3CP-4SI-7SA-12
CMMC v2.0CM.L2-3.4.3CP.L2-3.6.3SI.L2-3.14.6SA.L2-3.13.11
SR 11-7IV.AIV.CIII.BV.B
NIS 2Art.21(2)(e)Art.21(2)(c)Art.21(2)(d)Art.21(2)(d)

8. Verifying Witness Evidence

All SWT3 witness anchors are independently verifiable. The assessor does not need to trust the organization's systems to confirm the evidence.

Verification steps

  1. Anchor format check: Every anchor follows SWT3-{TIER}-{PROVIDER}-{UCT}-{PROCEDURE}-{VERDICT}-{EPOCH}-{SHA256_12}. Parse and validate structure.
  2. Fingerprint recomputation: Using the factors from the witness payload, recompute SHA256("WITNESS:{tenant}:{procedure}:{fa}:{fb}:{fc}:{ts_ms}").hex()[:12] and compare against the anchor's hash suffix.
  3. Merkle inclusion proof: For phase-gated evidence, request an inclusion proof from /api/v1/merkle/proof?fingerprint=xxx. Walk the proof to the daily root.
  4. Ed25519 signature verification: Every witness receipt is signed by did:web:sovereign.tenova.io. The public key is in the DID document at /.well-known/did.json.
  5. RFC 3161 timestamp: The daily Merkle root is independently timestamped. Verify the timestamp token against the TSA's certificate.

Public verification tool: sovereign.tenova.io/verify accepts any SWT3 anchor for client-side verification. No account required. The verification runs entirely in the browser.

9. Assessment Protocol

A suggested protocol for assessing infrastructure governance evidence during conformity assessment.

Pre-assessment (desk review)

  1. Request the organization's infrastructure governance policy and procedure inventory.
  2. Identify which SWT3 procedures are in scope for the assessment framework.
  3. Request witness ledger export filtered by infrastructure procedures (CM-3.1, CP-4.1, SI-7.1, SI-7.2, AI-DRIFT.1, AI-SUPPLY.1, AI-SBOM.1, AI-HW.1).
  4. Check for temporal gaps in the witness chain. Flag any gap exceeding the organization's stated monitoring interval.

On-site assessment

  1. Select one recent change and one recent DR test for deep verification.
  2. Request the Merkle tree structure ID for each.
  3. Walk the phase gate proofs. Verify that phases were sequential and complete.
  4. For the change: verify drift scores at each phase gate. For the DR test: verify RTO/RPO attestations.
  5. Independently recompute at least three anchor fingerprints using the raw factors.

Common findings

Questions or feedback: support@tenovaai.com