Cryptographic proof that your DR test executed correctly through all phases. Not just that it happened, but that it happened in the right order, with the right outcomes.
Who this is for: IT auditors, C3PAO assessors, DORA compliance officers, business continuity managers, infrastructure leads running DR exercises, and developers integrating DR test evidence into compliance pipelines.
"Show me the evidence that your last DR test actually ran."
Most organizations can produce a DR plan. Many can produce a calendar entry showing a test was scheduled. Some can produce a summary report written after the fact. Almost none can prove the test executed in the correct sequence with verifiable timing between phases.
The gap is not whether the test happened. The gap is whether it happened correctly. Did the backup integrity check complete before the restore began? Did functional validation finish before someone authorized the cutover? Was the RTO actually met, or was the timestamp adjusted after the fact?
Traditional DR test evidence is a narrative document. Someone writes "the test was successful" in a document and attaches it to a ticket. That document could have been written before the test, after the test, or instead of the test. There is no cryptographic binding between the evidence and the execution.
Phased process witnessing changes this. Each phase of the DR test produces a witness anchor with a timestamp, a fingerprint, and a clearing level. The phases form a Merkle tree. The tree proves sequence, completeness, and timing. An auditor does not need to trust the narrative. The math is the evidence.
Assessor Note
When reviewing DR test evidence, ask for the structure root hash. If the organization can produce one, the entire test lifecycle is verifiable from that single value. If they cannot, the evidence is narrative-only and requires traditional corroboration.
One hash. Four phases. Six frameworks. The structure root of a witnessed DR test is a single value that an assessor can independently verify. It proves sequence, completeness, and timing for CP-4 (NIST), Art.11 (DORA), A.6.2.6 (ISO 42001), Art.9 (EU AI Act), IV.C (SR 11-7), and CP.L2-3.6.3 (CMMC) simultaneously.
A well-executed DR test moves through four distinct phases. Each phase depends on the successful completion of the one before it. Skipping a phase or executing them out of order invalidates the test.
Before restoring anything, verify that backups exist and are intact. This means checking backup checksums, confirming retention windows, and validating that the backup set covers all critical systems. A restore from a corrupted backup is not a DR test. It is an incident.
Restore systems from verified backups into the DR environment. Each service restoration is an independent event. Track which services restored successfully, which failed, and whether the overall restore completed within the recovery time objective (RTO).
Verify that restored systems actually work. Run regression tests. Compare DR environment behavior against production baselines. A system that restores but produces incorrect results is not recovered. Functional parity is the measure, not just service availability.
The final authorization to declare the DR environment production-ready. This is a human decision backed by evidence from the three preceding phases. The cutover decision records who authorized it, what evidence they reviewed, and whether RTO and RPO targets were met.
Phase 1 Phase 2 Phase 3 Phase 4
Backup System Functional Cutover
Verification Restore Validation Decision
+-----------+ +-----------+ +-----------+ +-----------+
| Checksums | | Service A | | Regression| | RTO met? |
| Retention | | Service B | | Baseline | | RPO met? |
| Coverage | | Service C | | Parity % | | Authorize |
+-----+-----+ +-----+-----+ +-----+-----+ +-----+-----+
| | | |
v v v v
GATE 1 ---------> GATE 2 ---------> GATE 3 ---------> GATE 4
(backup ok?) (restore ok?) (validated?) (go/no-go)
Four phases, four gates. Each gate must open before the next phase begins. The sequence is cryptographically enforced.
Key distinction: SWT3 witnesses that each phase completed and records the evidence. It does not implement the backup, restore, or validation processes themselves. The protocol is the evidence layer, not the execution layer.
Each DR test phase maps to one or more SWT3 procedures. These procedures define what evidence is captured and how it is structured in the witness anchor.
The overarching attestation that a DR test was conducted. Covers the full test lifecycle, including scope, participants, and overall outcome. This is the anchor that auditors look for first.
Phase: All (overall attestation) • Factors: test scope, participant count, pass/fail outcome
System validation under stress conditions. In a DR context, this captures whether services survived the restore process and operated correctly under degraded conditions. Each service restoration is a perturbation; each successful restoration is a survival.
Phase: Backup Verification, System Restore • Factors: perturbations tested, perturbations survived, perturbation type
Comparing DR environment behavior against production baselines. The "champion" is the production system; the "challenger" is the restored DR environment. Functional parity percentage is the key metric.
Phase: Functional Validation • Factors: production baseline hash, DR environment hash, parity percentage
The formal authorization to return to production. Records who approved the cutover, how many validators signed off, and whether the validation passed. This is the final gate in the DR test lifecycle.
Phase: Cutover Decision • Factors: system ID, recommission type, validator count, validation result
| Phase | Procedure | Evidence Captured |
|---|---|---|
| Backup Verification | CP-4.1, AI-ROBUST.1 | Backup checksums, retention confirmation, coverage scope |
| System Restore | AI-ROBUST.1 | Services restored, failures encountered, RTO elapsed time |
| Functional Validation | AI-ASSESS.1 | Production vs. DR parity, regression test results, baseline hash |
| Cutover Decision | AI-RECOMM.1 | Authorization identity, validator count, RTO/RPO attestation |
The four DR test phases form a hierarchical Merkle tree. Individual evidence items (backup checksums, service statuses, test results) are leaves. Leaves roll up to milestones. Milestones roll up to phases. Phases roll up to a single structure root that represents the entire DR test.
DR Test Structure Root
(single hash = entire test)
|
+---------------+---------------+
| | |
Phase 0 Phase 1 Phase 2 Phase 3
Backup Restore Validate Cutover
| | | |
+---+---+ +---+---+ +---+ +---+
| | | | | |
MS:backup MS:ret MS:svcA MS:svcB MS:parity MS:cutover
integrity ention restore restore check approval
| | | | | |
leaves leaves leaves leaves leaves leaves
(checksums)(days) (status)(status)(parity %) (approver,
RTO hours)
Three-tier Merkle tree. One structure root proves the complete DR test. Any individual leaf is verifiable via inclusion proof against the root.
The structure root changes every time a new leaf is added. After the final cutover milestone, the root hash is frozen. That hash is the DR test's cryptographic identity. Share it with auditors, store it in your GRC tool, reference it in your POA&M. One hash, complete provenance.
Assessor Note
Request the structure root and use the proof API to verify any individual milestone. A valid inclusion proof confirms that the specific evidence item (e.g., backup checksum verification) is part of the overall DR test record. Tampering with any leaf invalidates the root.
from swt3_ai import Witness
from swt3_ai.milestone_merkle import MilestoneAccumulator
witness = Witness(tenant_id="acme-corp", endpoint="https://your-instance.example.com/api/v1/witness")
acc = MilestoneAccumulator()
acc.structure_id = "dr-test-q4-2026"
acc.add_phase("backup", "Backup Verification", sequence=0)
acc.add_phase("restore", "System Restore", sequence=1)
acc.add_phase("validate", "Functional Validation", sequence=2)
acc.add_phase("cutover", "Cutover Decision", sequence=3)
# Phase 1: Backup Verification
acc.add_milestone("backup-integrity", "backup", "Backup Integrity Check",
required_procedures=["CP-4.1"])
payload = witness.witness_robustness(
perturbations_tested=1, perturbations_survived=1,
perturbation_type="backup_restore_cycle"
)
acc.add_leaf("backup-integrity", payload.fingerprint,
{"procedure": "CP-4.1", "backup_size_gb": 847})
# Phase 2: System Restore
acc.add_milestone("restore-complete", "restore", "System Restore Verified",
required_procedures=["AI-ROBUST.1"])
payload = witness.witness_robustness(
perturbations_tested=12, perturbations_survived=12,
perturbation_type="service_restoration"
)
acc.add_leaf("restore-complete", payload.fingerprint,
{"services_restored": 12, "rto_met": True})
# Phase 3: Functional Validation
acc.add_milestone("validation-pass", "validate", "Functional Validation",
required_procedures=["AI-ASSESS.1"])
payload = witness.witness(
procedure_id="AI-ASSESS.1",
factor_a="production_baseline_hash_abc123",
factor_b="dr_environment_hash_def456",
factor_c="functional_parity_98.7_percent"
)
acc.add_leaf("validation-pass", payload.fingerprint,
{"parity_pct": 98.7, "test_suite": "full_regression"})
# Phase 4: Cutover Decision
acc.add_milestone("cutover-approved", "cutover", "Cutover Authorization",
required_procedures=["AI-RECOMM.1"])
payload = witness.witness_recommission(
system_id="prod-cluster-east",
recommission_type="dr_recovery",
validator_count=3,
validation_passed=True
)
acc.add_leaf("cutover-approved", payload.fingerprint,
{"approver": "ops-lead", "rto_hours": 2.3})
# Check all gates
for phase_id in ["backup", "restore", "validate", "cutover"]:
gate = acc.check_gate(phase_id)
print(f"{phase_id}: {'OPEN' if gate['passed'] else 'BLOCKED'}")
# The structure root is now the cryptographic identity of this DR test
print(f"DR test root: {acc.structure_root}")
import { Witness } from "@tenova/swt3-ai";
import { MilestoneAccumulator } from "@tenova/swt3-ai/milestone-merkle";
const witness = new Witness({
tenantId: "acme-corp",
endpoint: "https://your-instance.example.com/api/v1/witness"
});
const acc = new MilestoneAccumulator("dr-test-q4-2026");
acc.addPhase("backup", "Backup Verification", 0);
acc.addPhase("restore", "System Restore", 1);
acc.addPhase("validate", "Functional Validation", 2);
acc.addPhase("cutover", "Cutover Decision", 3);
// Phase 1: Backup Verification
acc.addMilestone("backup-integrity", "backup", "Backup Integrity Check",
["CP-4.1"]);
const p1 = await witness.witnessRobustness({
perturbationsTested: 1, perturbationsSurvived: 1,
perturbationType: "backup_restore_cycle"
});
acc.addLeaf("backup-integrity", p1.fingerprint,
{ procedure: "CP-4.1", backupSizeGb: 847 });
// Phase 2: System Restore
acc.addMilestone("restore-complete", "restore", "System Restore Verified",
["AI-ROBUST.1"]);
const p2 = await witness.witnessRobustness({
perturbationsTested: 12, perturbationsSurvived: 12,
perturbationType: "service_restoration"
});
acc.addLeaf("restore-complete", p2.fingerprint,
{ servicesRestored: 12, rtoMet: true });
// Phase 3: Functional Validation
acc.addMilestone("validation-pass", "validate", "Functional Validation",
["AI-ASSESS.1"]);
const p3 = await witness.witness({
procedureId: "AI-ASSESS.1",
factorA: "production_baseline_hash_abc123",
factorB: "dr_environment_hash_def456",
factorC: "functional_parity_98.7_percent"
});
acc.addLeaf("validation-pass", p3.fingerprint,
{ parityPct: 98.7, testSuite: "full_regression" });
// Phase 4: Cutover Decision
acc.addMilestone("cutover-approved", "cutover", "Cutover Authorization",
["AI-RECOMM.1"]);
const p4 = await witness.witnessRecommission({
systemId: "prod-cluster-east",
recommissionType: "dr_recovery",
validatorCount: 3,
validationPassed: true
});
acc.addLeaf("cutover-approved", p4.fingerprint,
{ approver: "ops-lead", rtoHours: 2.3 });
// Check all gates
for (const phaseId of ["backup", "restore", "validate", "cutover"]) {
const gate = acc.checkGate(phaseId);
console.log(`${phaseId}: ${gate.passed ? "OPEN" : "BLOCKED"}`);
}
console.log(`DR test root: ${acc.structureRoot}`);
DR testing evidence is required by virtually every compliance framework that governs critical infrastructure. The following table maps each framework's DR testing requirement to the SWT3 procedures used in this guide.
| Framework | Requirement | What It Demands | SWT3 Procedures |
|---|---|---|---|
| NIST 800-53 | CP-4 | Contingency plan testing at defined frequency; test results documented | CP-4.1, AI-ROBUST.1 |
| DORA | Art. 11 | ICT response and recovery testing; scenario-based exercises with measurable outcomes | CP-4.1, AI-ASSESS.1, AI-RECOMM.1 |
| ISO 42001 | A.6.2.6 | AI system continuity planning; evidence that continuity controls are tested | CP-4.1, AI-ROBUST.1 |
| EU AI Act | Art. 9 | Risk management system must include resilience measures; testing evidence required | AI-ROBUST.1, AI-ASSESS.1 |
| SR 11-7 | IV.C | Outcomes analysis for model risk; production vs. recovery environment comparison | AI-ASSESS.1, AI-RECOMM.1 |
| CMMC | CP.L2-3.6.3 | Test the organizational contingency plan; evidence of execution required | CP-4.1, AI-ROBUST.1 |
Assessor Note
A single DR test witnessed through this phase-gate pattern satisfies the evidence requirements of all six frameworks simultaneously. The structure root hash can be referenced in NIST SSP artifacts, DORA ICT reports, ISO 42001 audit packages, and CMMC assessment documentation. One test, one root, six frameworks.
DORA Art.11 Note
Financial entities must test ICT business continuity plans at least annually (Art.11(6)). The ECB supervisory team expects scenario-based exercises with measurable outcomes. The structure root is the artifact: it links directly to the RTO/RPO attestations in the cutover milestone. Include the structure ID in your ICT risk management report.
CMMC C3PAO Note
For CP.L2-3.6.3, the assessor needs evidence that the contingency plan was tested and that the test results were documented. The phase-gated Merkle tree satisfies both: the structure proves the test ran, and the milestone leaves contain the documented results. Reference the structure root in the assessment object evidence column.
When evaluating DR test evidence produced through phase-gated witnessing, auditors should verify four properties.
Phases must be sequential. The backup verification anchor must precede the system restore anchor. The system restore must precede functional validation. Functional validation must precede cutover. If any phase's timestamp violates the expected sequence, the test evidence is compromised. The Merkle tree's phase gate records capture this ordering: a gate cannot open until the preceding phase is complete.
Assessor Note
Check the gate fingerprints. Each GATE_OPEN leaf contains a timestamp. Compare the GATE_OPEN timestamp for Phase N against the first leaf timestamp in Phase N+1. If Phase N+1 has evidence dated before Phase N's gate opened, the sequence was violated. This is the "time travel" test.
All four phases must be present. A DR test that stops after system restore but never validates functional parity is incomplete. The structure root will reflect the missing phases: the validate and cutover phase roots will be empty, and their gates will show GATE_BLOCKED status.
The DR environment must be separate from production. The AI-ASSESS.1 anchor captures two distinct hashes: the production baseline and the DR environment. If both hashes are identical, the "test" may have been run against production itself, which is not a valid recovery exercise.
Recovery time and recovery point objectives are the measurable outcomes of a DR test. The cutover milestone records the elapsed time. Compare the witnessed RTO against the organization's stated objectives. A DR test that meets its functional targets but exceeds its time targets is a partial success, and the evidence should reflect that.
EU AI Act Conformity Note
Under Art.9(2)(a), the risk management system must be a continuous iterative process that includes testing. During an Annex VI conformity assessment, the NB can request the DR test structure root and verify the phase sequence independently. The Ed25519 signature on each receipt is issued by a DID the NB can resolve without contacting the provider.
DR tests fail. That is partly the point: a test that can never fail is not testing anything. The witness protocol captures failure evidence with the same fidelity as success evidence. FAIL verdicts are not errors in the system. They are the system working correctly.
Backup verification succeeds, but the restore of Service C fails. The restore milestone shows 11 of 12 perturbations survived. The gate for the restore phase shows GATE_BLOCKED with failing_milestones listing the incomplete service.
# Restore partially fails
payload = witness.witness_robustness(
perturbations_tested=12, perturbations_survived=11,
perturbation_type="service_restoration"
)
acc.add_leaf("restore-complete", payload.fingerprint,
{"services_restored": 11, "service_c_status": "FAILED",
"rto_met": False})
gate = acc.check_gate("restore")
print(gate["passed"]) # False
print(gate["failing_milestones"]) # ["restore-complete"]
The partial failure is now part of the permanent record. The organization can remediate (fix Service C, re-run the restore, add another leaf with 12/12) or document the failure as a finding in their POA&M. Either way, the evidence exists.
The restore completes, but functional validation reveals data integrity issues. The AI-ASSESS.1 anchor records low parity (e.g., 72.3% instead of the expected 98%+). The validation gate blocks. The corruption is discovered during the test, not in production. The witness record proves the test caught the problem.
All services restore and pass validation, but the total elapsed time exceeds the stated RTO. The cutover milestone records the actual time. The recommission witness captures validation_passed=True but the metadata shows rto_hours: 6.2 against a 4-hour target. The test succeeded functionally but failed its time objective. Both facts are in the tree.
FAIL verdicts are evidence, not defects. A DR test that produces a clean FAIL verdict with detailed factor data is more valuable than a test with no evidence at all. The FAIL anchor proves the test ran, the failure was detected, and the organization has a documented basis for remediation. Many auditors consider a well-documented failure more credible than an undocumented success.
The most common failure mode is the absence of evidence entirely. No structure root. No milestone leaves. No gate checks. When an auditor asks for DR test evidence and receives nothing, that is a finding against CP-4. The phase-gate approach makes this gap immediately visible: either the structure exists or it does not.
All endpoints require Bearer authentication (axm_* API key or session token).
POST /api/v1/merkle/milestone
Content-Type: application/json
Authorization: Bearer axm_live_...
{
"action": "create",
"name": "DR Test Q4 2026",
"domain": "disaster_recovery",
"phases": [
{
"name": "Backup Verification",
"milestones": [
{"name": "Backup Integrity", "required_procedures": ["CP-4.1"]}
]
},
{
"name": "System Restore",
"milestones": [
{"name": "Service Restoration", "required_procedures": ["AI-ROBUST.1"]}
]
},
{
"name": "Functional Validation",
"milestones": [
{"name": "Parity Check", "required_procedures": ["AI-ASSESS.1"]}
]
},
{
"name": "Cutover Decision",
"milestones": [
{"name": "Authorization", "required_procedures": ["AI-RECOMM.1"]}
]
}
]
}
Response:
{
"ok": true,
"structure_id": "ST-...",
"phase_ids": ["PH-...", "PH-...", "PH-...", "PH-..."],
"milestone_count": 4
}
GET /api/v1/merkle/proof?fingerprint=a1b2c3d4e5f6&structure_id=ST-...
Authorization: Bearer axm_live_...
Response:
{
"ok": true,
"fingerprint": "a1b2c3d4e5f6",
"structure_root": "...",
"proof": {
"milestone_proof": ["..."],
"phase_proof": ["..."],
"structure_proof": ["..."]
},
"verified": true
}
POST /api/v1/merkle/gate
Content-Type: application/json
Authorization: Bearer axm_live_...
{
"structure_id": "ST-...",
"phase_id": "PH-..."
}
Response:
{
"ok": true,
"passed": true,
"gate_status": "GATE_OPEN",
"gate_fingerprint": "...",
"failing_milestones": [],
"unresolved_fails": [],
"missing_procedures": []
}
A gate_status of GATE_BLOCKED means one or more milestones have unresolved failures or missing required procedures. The response includes the specific blockers so the caller can diagnose and remediate.
Questions or feedback: support@tenovaai.com