Cryptographic proof that every update, patch, and release followed governance gates before reaching production. One Merkle tree per change pipeline.
Who this is for: SOC 2 auditors, C3PAO assessors, compliance officers, DORA examiners, release managers, platform engineers, and DevOps leads integrating change governance into CI/CD pipelines.
Organizations push changes to production every day. Some push hundreds of times per day. The engineering team knows the change went through a pipeline. The release manager approved it. The canary looked healthy. But when the auditor asks for evidence, the answer is a screenshot of a green CI badge and a chat message that says "LGTM."
The problem is not that governance did not happen. In most mature organizations, it did. The problem is that the evidence is scattered across tools that were never designed to produce audit-grade proof. The CI system logs a build. The approval lives in a pull request comment. The canary metrics sit in a monitoring dashboard with a 30-day retention window. The staging validation is a memory in someone's head.
Six months later, an auditor asks: "Can you prove this specific release passed integration testing, was validated in staging, survived canary deployment, and was explicitly approved before production rollout?" The answer is usually "we can reconstruct most of it." Reconstruction is not proof.
Change Management Witnessing closes this gap. Each governance gate in the deployment pipeline produces a cryptographic attestation. The attestations roll up into a hierarchical Merkle tree that represents the entire change lifecycle. The tree is tamper-evident. The proofs are offline-verifiable. And the structure root is a single hash that an auditor, a regulator, or an automated compliance system can verify in milliseconds.
Key distinction: SWT3 does not implement change management controls. It witnesses that your existing change management process produced the expected evidence at each gate. Your CI/CD pipeline remains yours. SWT3 provides the cryptographic receipt. One pipeline, one Merkle tree, nine framework requirements: CM-3, CM-4, CM-5 (NIST), Art.17 (EU AI Act), Cl.8.1 (ISO 42001), Art.8 (DORA), CM.L2-3.4.3, CM.L2-3.4.4 (CMMC), and CC8.1 (SOC 2).
A deployment pipeline is a phased process. Each phase has entry criteria, produces evidence, and must satisfy its gate before the next phase begins. This is the same pattern used in construction (foundation before framing), manufacturing (QC before shipping), and pharmaceutical production (sterilization before release). The domain labels change; the structure does not.
| Phase | Purpose | Gate Criteria |
|---|---|---|
| 0: Integration Testing | Verify the change does not break existing functionality | All unit tests pass. Integration tests pass. Security scan returns no critical findings. |
| 1: Staging Validation | Validate the change in a production-like environment | Staging deployment completes. Smoke tests pass. Drift check confirms the change behaves within expected thresholds. |
| 2: Canary Deployment | Expose a small percentage of production traffic to the change | Error rate stays below threshold. Latency remains within SLA. No anomalous behavior detected. |
| 3: Production Rollout | Deploy to the full fleet | Rollout completes. Parameter freeze attested. Post-deployment validation passes. |
Each gate is deterministic. The gate evaluates whether all milestones in the phase have terminal PASS verdicts. If any milestone is missing or has an unresolved FAIL, the gate blocks. The gate decision itself is recorded as a witnessed leaf in the Merkle tree.
Assessor Note
The four-phase model is a reference architecture. Organizations may have more phases (e.g., a separate security review phase) or fewer (e.g., no canary for internal tools). The protocol accommodates any number of phases. What matters is that each phase has at least one witnessed milestone and that the gate between phases is cryptographically recorded.
Each phase maps to specific SWT3 procedures. The procedure determines what evidence is captured and how it is evaluated.
Witnesses the baseline state of configurations, dependencies, and infrastructure. Used at the beginning and end of the pipeline to prove the change was deliberate and bounded. Factor A captures the configuration hash; Factor B records the change ticket identifier; Factor C holds the commit SHA or artifact version.
Detects behavioral drift between the pre-change and post-change state. Used during staging validation to prove the change did not introduce unexpected behavioral shifts. Records metrics evaluated, drift score, baseline hash, and drifted count.
Maps observed drift to business impact categories. Used during canary deployment to prove that any detected drift was within acceptable consequence boundaries. Records drift magnitude, consequence level, whether the threshold was breached, and the action taken.
Attests that model parameters, configuration, and weights are frozen before production deployment. Proves the version that was tested is the version that shipped. Factor A captures the configuration hash; Factor B records rollback plan status; Factor C holds the approval ticket.
The change pipeline maps to a hierarchical Merkle tree. Each milestone produces attestation fingerprints. Milestones roll up to phases. Phases roll up to a single structure root that represents the complete governance state of the change.
Change Pipeline Root (structure_id: "release-v2.14.0")
|
+-- Phase 0: Integration Testing
| +-- unit_tests_pass [CM-3.1] fingerprint: a1b2c3...
| +-- integration_tests_pass [CM-3.1] fingerprint: d4e5f6...
| +-- security_scan_clean [CM-3.1] fingerprint: 789abc...
|
+-- Phase 1: Staging Validation
| +-- staging_deploy_complete [CM-3.1] fingerprint: def012...
| +-- smoke_tests_pass [CM-3.1] fingerprint: 345678...
| +-- drift_check_within_threshold [AI-DRIFT.1] fingerprint: 9abcde...
|
+-- Phase 2: Canary Deployment
| +-- canary_5pct_deployed [CM-3.1] fingerprint: f01234...
| +-- error_rate_below_threshold [AI-DRIFT.2] fingerprint: 567890...
| +-- latency_within_sla [AI-DRIFT.2] fingerprint: abcdef...
|
+-- Phase 3: Production Rollout
+-- rollout_complete [CM-3.1] fingerprint: 012345...
+-- freeze_attestation [AI-FREEZE.1] fingerprint: 6789ab...
+-- post_deploy_validation [CM-3.1] fingerprint: cdef01...
Each leaf is a witnessed attestation. Each phase gate evaluates terminal verdicts. The structure root changes with every new leaf.
The structure root is a single hash. If any leaf changes, the root changes. An auditor who recorded the root at deployment time can verify months later that the evidence has not been altered. The proof is mathematical, not procedural.
Temporal ordering matters. The gate check for Phase 0 must have an earlier timestamp than the first leaf in Phase 1. This proves the sequence was honored. SWT3 records millisecond-precision timestamps in every attestation fingerprint, making temporal ordering independently verifiable.
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 = "release-v2.14.0"
acc.add_phase("test", "Integration Testing", sequence=0)
acc.add_phase("staging", "Staging Validation", sequence=1)
acc.add_phase("canary", "Canary Deployment", sequence=2)
acc.add_phase("production", "Production Rollout", sequence=3)
# Phase 0: Integration Testing
acc.add_milestone("unit-tests", "test", "Unit Test Suite", required_procedures=["CM-3.1"])
payload = witness.witness(
procedure_id="CM-3.1",
factor_a="tests_passed_847_of_847",
factor_b="coverage_94_percent",
factor_c="commit_sha_a1b2c3d4"
)
acc.add_leaf("unit-tests", payload.fingerprint, {"passed": 847, "coverage": 94.0})
acc.add_milestone("security-scan", "test", "Security Scan", required_procedures=["CM-3.1"])
payload = witness.witness(
procedure_id="CM-3.1",
factor_a="scanner_critical_0_high_0",
factor_b="scan_duration_42s",
factor_c="commit_sha_a1b2c3d4"
)
acc.add_leaf("security-scan", payload.fingerprint, {"critical": 0, "high": 0})
# Phase 1: Staging Validation
acc.add_milestone("staging-deploy", "staging", "Staging Deployment", required_procedures=["AI-DRIFT.1"])
payload = witness.witness_drift(
metrics_evaluated=42, drifted_count=0,
drift_type="deployment_comparison",
baseline_hash="staging_v2.13.0_hash",
drift_score=0.003
)
acc.add_leaf("staging-deploy", payload.fingerprint, {"environment": "staging", "drift_score": 0.003})
# Phase 2: Canary Deployment
acc.add_milestone("canary-metrics", "canary", "Canary Health", required_procedures=["AI-DRIFT.2"])
payload = witness.witness_drift_consequence(
drift_type="canary_error_rate",
drift_magnitude=0.02,
consequence_level="low",
threshold_breached=False,
action_taken="proceed_to_full_rollout"
)
acc.add_leaf("canary-metrics", payload.fingerprint, {"canary_pct": 5, "error_rate": 0.02})
# Phase 3: Production Rollout
acc.add_milestone("freeze-attest", "production", "Parameter Freeze", required_procedures=["AI-FREEZE.1"])
payload = witness.witness(
procedure_id="AI-FREEZE.1",
factor_a="config_hash_locked_xyz789",
factor_b="rollback_plan_verified",
factor_c="approval_ticket_CHG-2026-1847"
)
acc.add_leaf("freeze-attest", payload.fingerprint, {"change_ticket": "CHG-2026-1847"})
# Verify the full pipeline
gate = acc.check_gate("production")
print(gate["passed"]) # True
print(acc.structure_root) # Single hash: entire release governance state
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();
acc.structureId = "release-v2.14.0";
acc.addPhase("test", "Integration Testing", 0);
acc.addPhase("staging", "Staging Validation", 1);
acc.addPhase("canary", "Canary Deployment", 2);
acc.addPhase("production", "Production Rollout", 3);
// Phase 0: Integration Testing
acc.addMilestone("unit-tests", "test", "Unit Test Suite", ["CM-3.1"]);
let payload = await witness.witness({
procedureId: "CM-3.1",
factorA: "tests_passed_847_of_847",
factorB: "coverage_94_percent",
factorC: "commit_sha_a1b2c3d4"
});
acc.addLeaf("unit-tests", payload.fingerprint, { passed: 847, coverage: 94.0 });
acc.addMilestone("security-scan", "test", "Security Scan", ["CM-3.1"]);
payload = await witness.witness({
procedureId: "CM-3.1",
factorA: "scanner_critical_0_high_0",
factorB: "scan_duration_42s",
factorC: "commit_sha_a1b2c3d4"
});
acc.addLeaf("security-scan", payload.fingerprint, { critical: 0, high: 0 });
// Phase 1: Staging Validation
acc.addMilestone("staging-deploy", "staging", "Staging Deployment", ["AI-DRIFT.1"]);
payload = await witness.witnessDrift({
metricsEvaluated: 42, driftedCount: 0,
driftType: "deployment_comparison",
baselineHash: "staging_v2.13.0_hash",
driftScore: 0.003
});
acc.addLeaf("staging-deploy", payload.fingerprint, { environment: "staging", driftScore: 0.003 });
// Phase 2: Canary Deployment
acc.addMilestone("canary-metrics", "canary", "Canary Health", ["AI-DRIFT.2"]);
payload = await witness.witnessDriftConsequence({
driftType: "canary_error_rate",
driftMagnitude: 0.02,
consequenceLevel: "low",
thresholdBreached: false,
actionTaken: "proceed_to_full_rollout"
});
acc.addLeaf("canary-metrics", payload.fingerprint, { canaryPct: 5, errorRate: 0.02 });
// Phase 3: Production Rollout
acc.addMilestone("freeze-attest", "production", "Parameter Freeze", ["AI-FREEZE.1"]);
payload = await witness.witness({
procedureId: "AI-FREEZE.1",
factorA: "config_hash_locked_xyz789",
factorB: "rollback_plan_verified",
factorC: "approval_ticket_CHG-2026-1847"
});
acc.addLeaf("freeze-attest", payload.fingerprint, { changeTicket: "CHG-2026-1847" });
// Verify the full pipeline
const gate = acc.checkGate("production");
console.log(gate.passed); // true
console.log(acc.structureRoot); // Single hash: entire release governance state
Change management is one of the most universally regulated activities in information systems. The same deployment pipeline evidence satisfies requirements across multiple frameworks simultaneously.
| Framework | Control | What the Evidence Proves |
|---|---|---|
| NIST 800-53 | CM-3 (Configuration Change Control) | Changes are tracked, tested, and approved before implementation |
| NIST 800-53 | CM-4 (Impact Analysis) | Staging validation and drift checks demonstrate impact was assessed |
| NIST 800-53 | CM-5 (Access Restrictions for Change) | Freeze attestation proves only authorized changes reached production |
| EU AI Act | Art. 17 (Quality Management System) | Systematic governance of changes to high-risk AI systems |
| ISO 42001 | 8.1 (Operational Planning and Control) | Planned processes for managing AI system changes |
| DORA | Art. 8 (ICT Risk Management Framework) | Change management within ICT risk management for financial entities |
| CMMC | CM.L2-3.4.3 | Track, review, approve/disapprove, and log changes to organizational systems |
| CMMC | CM.L2-3.4.4 | Analyze the security impact of changes prior to implementation |
| SOC 2 | CC8.1 (Change Management) | Changes to infrastructure, data, software, and procedures are authorized and managed |
Assessor Note
The framework crosswalk is built into the SWT3 registry. When evaluating change management evidence, request the structure root for a recent release and verify the phase gate sequence. A single verified tree satisfies CM-3, CM-4, and CM-5 simultaneously because it proves the change was tested (CM-4), tracked through gates (CM-3), and frozen before production (CM-5).
SOC 2 Type II Note
For CC8.1, the auditor tests whether changes to infrastructure, data, software, and procedures are authorized and managed. The phase-gated Merkle tree maps directly to the SOC 2 testing procedure: select a sample of changes, request the structure root for each, and verify that all four gates opened cleanly. The freeze attestation in Phase 3 proves authorization preceded deployment.
DORA Art.8 Note
Financial entities must include change management procedures within their ICT risk management framework. The structure root for each release can be included in the entity's ICT risk management report. For examiners: the drift attestations in Phases 1 and 2 prove the entity assessed the impact of each change on ICT services before production deployment.
When evaluating change management evidence from a phase-gated Merkle tree, auditors should verify five things.
Every phase must have at least one leaf. A tree where Phase 0 and Phase 3 have leaves but Phase 1 is empty indicates the change skipped staging. The gate between phases records whether it opened cleanly or was overridden. If a phase was skipped, the override leaf (or its absence) tells the story.
Assessor Note
Query the phase gate for each phase in sequence. A GATE_OPEN result with zero override leaves means the phase completed normally. A GATE_OPEN with an override leaf means someone bypassed the gate. A phase with no leaves at all means the phase was never executed.
The canary phase should contain AI-DRIFT.2 attestations showing that drift magnitude stayed within acceptable limits. Check that threshold_breached is false on all canary leaves. If it is true, the action_taken field should show either a rollback or an explicit risk acceptance with justification.
The AI-FREEZE.1 leaf in the production phase must have an earlier timestamp than the rollout completion leaf. This temporal ordering proves the configuration was frozen before deployment began, not after. The fingerprint formula includes a millisecond timestamp, making this independently verifiable.
The freeze attestation factors should include evidence that a rollback plan exists. This is not about whether the rollback was needed. It is about whether the organization had a plan before committing to production.
Every leaf in the tree should correlate to a change ticket (Factor C or metadata). This links the cryptographic evidence chain back to the organization's change advisory board (CAB) or equivalent approval process. The ticket number does not need to be verified by SWT3; its presence in the witnessed factors means it was recorded at the time of attestation.
Not every change succeeds. When a canary deployment detects anomalies, the change must be rolled back. The rollback decision is itself a governance event that needs evidence.
When the canary error rate exceeds the threshold, the pipeline witnesses the failure and the rollback decision.
# Canary failure detected
payload = witness.witness_drift_consequence(
drift_type="canary_error_rate",
drift_magnitude=0.15,
consequence_level="high",
threshold_breached=True,
action_taken="rollback_initiated"
)
acc.add_leaf("canary-metrics", payload.fingerprint, {
"canary_pct": 5,
"error_rate": 0.15,
"threshold": 0.05,
"decision": "rollback"
})
# Witness the rollback completion
payload = witness.witness(
procedure_id="CM-3.1",
factor_a="rollback_to_v2.13.0_complete",
factor_b="production_stable_confirmed",
factor_c="incident_ticket_INC-2026-0442"
)
acc.add_leaf("canary-metrics", payload.fingerprint, {
"rollback_target": "v2.13.0",
"incident_ticket": "INC-2026-0442"
})
The tree now contains the full narrative: the canary deployment detected a problem, the drift exceeded the threshold, a rollback was initiated, and the rollback completed successfully. All timestamped. All tamper-evident.
After a rollback, the fix must re-enter the pipeline at Phase 0. A new structure is created for the remediated release (e.g., release-v2.14.1). The new structure can reference the original structure ID in its metadata, creating a traceable link between the failed attempt and the successful one.
Assessor Note
A release that failed canary and was rolled back tells a stronger governance story than one that passed on the first attempt. The failure, the decision, and the remediation are all witnessed. Look for structure pairs: a failed release linked to its successful remediation. That is evidence of a mature change management process.
CMMC C3PAO Note
CM.L2-3.4.4 requires analysis of security impact prior to implementation. The canary phase with AI-DRIFT.2 attestations is direct evidence of impact analysis: the drift magnitude, consequence level, and threshold evaluation are captured at the time of assessment, not reconstructed after the fact. A rollback triggered by threshold breach proves the control is operational, not just documented.
Large-scale update events present a specific challenge: hundreds of patches must be tested, validated, and deployed in a compressed timeframe. Each patch needs its own evidence trail, but the volume makes individual pipeline runs impractical for evidence collection.
The SWT3 batch endpoint accepts up to 500 attestations per request. For a monthly patch cycle, the pattern is straightforward.
# Batch witness for a monthly patch cycle
import requests
patches = [
{
"procedure_id": "CM-3.1",
"factor_a": "KB5034441_installed",
"factor_b": "priority_critical",
"factor_c": "patch_tuesday_2026_11"
},
{
"procedure_id": "CM-3.1",
"factor_a": "KB5034442_installed",
"factor_b": "priority_high",
"factor_c": "patch_tuesday_2026_11"
},
# ... up to 500 per batch
]
response = requests.post(
"https://your-instance.example.com/api/v1/witness/batch",
headers={"Authorization": "Bearer axm_live_..."},
json={"attestations": patches}
)
# Response includes fingerprints for all patches
for result in response.json()["results"]:
print(f"{result['procedure_id']}: {result['fingerprint']}")
Each fingerprint from the batch response can be added as a leaf in a monthly patch structure. The structure root for "patch-cycle-2026-11" becomes a single hash that proves every patch in the cycle was witnessed.
Assessor Note
For large patch cycles, request the batch structure root rather than individual attestations. Verify the root, then spot-check individual patches by requesting multi-tier proofs for specific fingerprints. This is the same logarithmic verification pattern used for fleet validation: full coverage, targeted verification.
All endpoints require Bearer authentication (axm_* API key or session token).
| Method | Endpoint | Purpose |
|---|---|---|
| POST | /api/v1/merkle/milestone | Create a change pipeline structure with phases and milestones |
| GET | /api/v1/merkle/proof?fingerprint=xxx | Generate a multi-tier inclusion proof for any attestation |
| POST | /api/v1/merkle/gate | Evaluate a phase gate (returns GATE_OPEN or GATE_BLOCKED) |
| POST | /api/v1/witness/batch | Bulk attestation (up to 500 per request) for large patch cycles |
POST /api/v1/merkle/milestone
Content-Type: application/json
Authorization: Bearer axm_live_...
{
"action": "create",
"name": "release-v2.14.0",
"domain": "change_management",
"phases": [
{
"name": "Integration Testing",
"sequence": 0,
"milestones": [
{"name": "Unit Tests", "required_procedures": ["CM-3.1"]},
{"name": "Security Scan", "required_procedures": ["CM-3.1"]}
]
},
{
"name": "Staging Validation",
"sequence": 1,
"milestones": [
{"name": "Drift Check", "required_procedures": ["AI-DRIFT.1"]}
]
},
{
"name": "Canary Deployment",
"sequence": 2,
"milestones": [
{"name": "Canary Health", "required_procedures": ["AI-DRIFT.2"]}
]
},
{
"name": "Production Rollout",
"sequence": 3,
"milestones": [
{"name": "Freeze Attestation", "required_procedures": ["AI-FREEZE.1"]},
{"name": "Post-Deploy Validation", "required_procedures": ["CM-3.1"]}
]
}
]
}
Response:
{
"ok": true,
"structure_id": "ST-...",
"phase_ids": ["PH-...", "PH-...", "PH-...", "PH-..."],
"milestone_count": 6
}
POST /api/v1/merkle/gate
Content-Type: application/json
Authorization: Bearer axm_live_...
{
"phase_id": "PH-...",
"structure_id": "ST-..."
}
Response:
{
"ok": true,
"passed": true,
"gate_status": "GATE_OPEN",
"gate_fingerprint": "...",
"failing_milestones": [],
"unresolved_fails": [],
"missing_procedures": []
}
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/witness/batch
Content-Type: application/json
Authorization: Bearer axm_live_...
{
"attestations": [
{
"procedure_id": "CM-3.1",
"factor_a": "KB5034441_installed",
"factor_b": "priority_critical",
"factor_c": "patch_tuesday_2026_11"
},
{
"procedure_id": "CM-3.1",
"factor_a": "KB5034442_installed",
"factor_b": "priority_high",
"factor_c": "patch_tuesday_2026_11"
}
]
}
Response:
{
"ok": true,
"results": [
{"procedure_id": "CM-3.1", "fingerprint": "...", "verdict": "PASS"},
{"procedure_id": "CM-3.1", "fingerprint": "...", "verdict": "PASS"}
],
"count": 2
}
Full SDK documentation: sovereign.tenova.io/docs
Questions or feedback: support@tenovaai.com