Pilot Evaluation FAQ

SWT3 provides supplementary evidence for conformity assessment -- it does not replace the conformity assessment process. The protocol creates continuous, cryptographically anchored records of AI system behavior between formal assessment cycles. These records can demonstrate ongoing compliance with technical requirements (Art. 9, 10, 11, 12, 14, 15) in a format that is independently verifiable.

The Notified Body retains full authority over all conformity decisions. SWT3 evidence is recommended as supporting documentation, not as a substitute for the NB's own evaluation procedures.

SWT3 procedures map directly to EU AI Act articles. The current mapping covers Art. 9, 10, 11, 12, 14, 15, 16, and 50. When harmonised standards are published under the AI Act, crosswalk updates will be provided to map SWT3 procedures to the corresponding standard clauses.

The protocol's procedure-based architecture is designed to accommodate new standards without changes to the underlying evidence format. Each SWT3 Witness Anchor references a specific UCT procedure, which can be crosswalked to any regulatory framework.

Yes. SWT3 evidence is recommended as supplementary documentation for record-keeping and monitoring requirements within a quality management system audit. The NB can reference specific anchor fingerprints in their assessment report to establish traceability between the audit finding and the underlying evidence.

Each anchor includes a 12-character SHA-256 fingerprint, a Unix epoch timestamp, the procedure identifier, and the verdict. This provides the granularity needed for audit trail reconstruction.

Navigate to sovereign.tenova.io/verify and paste the anchor string. Verification runs entirely in the browser using client-side SHA-256. No login is required, no data is sent to any server, and the page functions without network connectivity after initial load.

For programmatic verification, you can also use the CLI: axiom verify SWT3-E-VULTR-... or the Python/TypeScript SDKs.

FULL: SWT3 produces complete evidence for the requirement. The anchor and its factors contain sufficient information for an assessor to make a determination without additional documentation.

PARTIAL: SWT3 produces supporting evidence, but the NB must assess surrounding processes, organizational controls, or documentation that falls outside the protocol's scope. For example, Art. 14 (human oversight) requires both technical evidence (which SWT3 captures) and organizational evidence (which the NB must verify independently).

See nb-pilot-scope.html for an article-by-article coverage matrix.

No. SWT3 is an evidence protocol. It creates witness records that document AI system behavior. Conformity decisions, certificates, and CE marking remain exclusively within the authority of the Notified Body. The platform does not make compliance determinations of any kind.

The clearing level determines where witness data is stored and what the platform receives:

• Level 0 (Analytics): Hashes, factors, model name, and provider transmitted. Dashboard access available.
• Level 1 (Standard): Hashes, factors, and model name transmitted. Raw prompts and responses never leave the customer's infrastructure.
• Level 2 (Sensitive): Hashes and factors only. Model identity and provider metadata stripped.
• Level 3 (Classified): Only numeric factors and hashed model identifiers are transmitted to the platform. No plaintext model names or provider metadata. For fully air-gapped Sovereign deployments, anchors remain entirely in the customer's infrastructure via .pulse bundles and the platform receives nothing.

Yes. The protocol specification and core SDKs are published under the Apache 2.0 license. SDKs are available in Python, TypeScript, Rust, C#, Ruby, Swift, Go, and Kotlin. The Axiom platform (dashboard, adjudicator, APIs) is proprietary. Patent pending.

SWT3 maps to CMMC practices via the NIST 800-53 crosswalk. AI-specific procedures map to control families including AC (Access Control), AU (Audit and Accountability), CM (Configuration Management), IA (Identification and Authentication), and SC (System and Communications Protection).

See the CMMC Overlay guide for practice-level mapping with specific procedure-to-practice relationships.

Yes. The platform exports OSCAL-formatted Assessment Results validated against the NIST oscal-cli reference implementation. Anchors can be referenced by fingerprint in SSP control descriptions to provide cryptographic traceability between the narrative and the underlying evidence.

Traditional (non-OSCAL) SSP exports are also available in HTML format for C3PAOs that prefer narrative-style documentation.

Yes. The platform tracks POA&M items with the following capabilities:

• Auto-close on remediation: When a previously failing control passes, the corresponding POA&M item is closed automatically.
• Severity-based milestones: CRITICAL = 7 days, HIGH = 30 days, MEDIUM = 90 days.
• Overdue flagging: Items that exceed their milestone date are flagged for escalation.
• Export: Available as OSCAL POA&M or CSV.

Two mechanisms provide CA-7 evidence:

• Drift monitor: Detects behavioral changes between scans and classifies severity. Any drift event is anchored with its own SWT3 Witness Anchor for auditability.
• Posture trend API: Provides daily snapshots of pass/fail/inherited counts, score, and direction (improving, declining, or stable). Historical data is available for up to 365 days on Enclave tier.

Both provide timestamped, cryptographically anchored records suitable for CA-7 continuous monitoring documentation.

Yes. 25 controls are mapped to Ubuntu 22.04 STIG Rule IDs. Axiom supports ingestion of DISA STIG artifact formats (.ckl, .cklb, XCCDF) via the axiom ingest command, allowing you to anchor external STIG evidence into the SWT3 ledger.

Yes. Assessor mode provides read-only dashboard access with full export capability. Assessors can view controls, verdicts, the compliance ledger, anchor verification, and all export functions. Assessors cannot modify attestations, configurations, controls, or any other platform state. All assessor actions are logged in the SI-12 audit trail.

The pilot covers one AI system. This focused scope ensures that the evaluation produces meaningful, actionable evidence rather than shallow coverage across multiple systems. Additional AI systems require an Enclave subscription or a separate pilot engagement.

Not currently. Authentication is via email and password with optional TOTP multi-factor authentication. SSO (SAML/OIDC) is on the roadmap for enterprise deployments. If SSO is a hard requirement for your evaluation, contact engineering@tenovaai.com to discuss timeline.

Yes. Pro and Enclave tiers include Auditor Share Links -- read-only, time-limited URLs that give assessors direct access to compliance posture without requiring credentials. Links can be revoked at any time. During the pilot, Enclave-tier access includes this capability.

Row-Level Security (RLS) at the database layer. Each tenant's data is isolated by tenant_id on every query. No tenant can access another tenant's anchors, verdicts, configuration, or any other data. The scoped client pattern enforces this isolation at the application layer as well, providing defense in depth.

• OSCAL: SSP, Assessment Results, POA&M (validated against NIST oscal-cli)
• Compliance Passport: HTML (Pro+), HMAC-signed JSON (Pro+), W3C Verifiable Credential (Enclave+)
• CSV: Ledger export, POA&M export
• HTML Reports: Gap-to-Green remediation roadmap, Executive Summary, CVE report
• Air-gap: .pulse bundles for offline transfer and ingestion

Yes. POST /api/v1/witness/batch accepts up to 500 anchors per request. The SDK handles batching automatically via the configurable flush interval, so most integrations do not need to call the batch endpoint directly. Individual anchors can be submitted via POST /api/v1/witness.

Near zero. Witnessing happens asynchronously after the inference returns. Your application is never blocked by the witness pipeline. The SDK buffers anchors in memory and flushes them on a configurable interval (default: 5 seconds). The inference response is returned to the caller immediately.

The SDK includes adapters or verified compatibility for:

• OpenAI, Anthropic, AWS Bedrock, Google ADK
• vLLM, Ollama, NVIDIA Dynamo, Cerebras
• LiteLLM (100+ providers via unified interface)
• Vercel AI SDK
• Any OpenAI-compatible endpoint

The wrap(client) pattern works with any provider that exposes a standard chat completion interface. For non-standard pipelines, the individual witness methods (witness_drift, witness_hardware, etc.) can be called directly.

SDKs are available in 8 languages: Python, TypeScript, Rust, C#, Ruby, Swift, Go, and Kotlin. Additionally, an MCP (Model Context Protocol) server provides 33 tools for MCP-native integration. All SDKs produce identical fingerprints -- cross-language parity is verified against shared test vectors.

Yes. Both the Python and TypeScript SDKs handle streaming transparently. The wrap(client) proxy intercepts streaming responses, accumulates the content for witnessing, and yields chunks to your application in real time. The witness anchor is minted after the stream completes.

The SDK buffers anchors locally and flushes them when connectivity resumes. No evidence is lost. Your AI system continues operating normally regardless of platform availability. The SDK is designed so that a witness infrastructure outage never affects inference latency or availability.

Yes. The SDK can operate in local-only mode, storing anchors as .pulse bundles on the local filesystem. These bundles can be transferred via sneakernet or secure file transfer and ingested into the platform when connectivity is available. Air-gapped deployment is available under Sovereign engagements. Note that air-gapped mode is a deployment configuration, not a clearing level -- you can run any clearing level (0-3) in either connected or air-gapped mode.

No. At Clearing Level 1 and above, raw prompts and responses never leave your infrastructure. The platform receives only one-way SHA-256 hashes and numeric factors (e.g., token count, latency, confidence score). These hashes cannot be reversed to recover the original content. At Clearing Level 3, anchor content is further minimized to numeric factors and hashed model names. In air-gapped Sovereign deployments, the platform receives no witness data at all.

Account-level data only: email address, organization name, and IP addresses for rate limiting. Witness anchors are one-way SHA-256 hashes that cannot identify any natural person. The numeric factors (token counts, latency measurements, confidence scores) are aggregate statistics, not personal data.

See the Data Processing Agreement for the complete sub-processor list and processing activities.

• Application: Vultr, United States (New Jersey)
• Database: Supabase (AWS us-east-1)
• Payment processing: Stripe

The full sub-processor list with data categories and locations is published in the Data Processing Agreement.

Yes. Transfer mechanisms include Standard Contractual Clauses (Module 2: Controller to Processor) and the EU-US Data Privacy Framework. Witness anchors are one-way SHA-256 hashes that fall outside the scope of personal data under GDPR Recital 26 (information that cannot reasonably be used to identify a natural person).

Full details are provided in the Data Processing Agreement and the Clearing Addendum.

EU-resident deployment is available as part of Sovereign engagements. This includes EU-based application hosting and EU-region database deployment. Contact engineering@tenovaai.com to discuss requirements and timeline.

$10,000 USD, single invoice, net-30. The pilot covers 10 working days, Enclave-tier platform access, three structured sessions (kickoff, midpoint review, final readout), and all deliverables including the evidence mapping report and integration guidance.

No. The pilot fee is a standalone engagement fee that covers the dedicated evaluation period, structured sessions, and deliverables. Subscription pricing is separate and begins after the pilot concludes.

• Pro: $499/month or $4,990/year
• Enclave: $9,500/month or $102,000/year
• Sovereign: $125,000 for a full ATO sprint engagement

All tiers include the full SDK, dashboard access, and guide library. Tier differences are primarily in retention duration, export formats, and support level. See the pricing page for a detailed comparison.

Extensions are available by mutual agreement at $1,000 per additional working day. Extension requests should be made before the original 10-day period concludes so that continuity of access and support is maintained.

Yes. The OPEN tier is free and includes:

• 30-day anchor retention
• Full SDK access (all 8 languages + MCP server)
• Dashboard access with controls, ledger, and verification
• Complete guide library (219 guides)
• No credit card required

The OPEN tier is recommended for technical evaluation before committing to a formal pilot. Sign up at sovereign.tenova.io/signup.

Your Enclave-tier tenant remains active with 365-day retention. Data is not deleted unless you request it. If you transition to a paid subscription, your existing anchors, verdicts, and configuration carry forward with no migration required. If you do not subscribe, your data remains accessible for export during the retention period.

Yes. Your tenant ID, API keys, anchors, and all configuration carry forward to any paid tier with no migration required. The transition from pilot to subscription is a tier change on your existing tenant -- not a new account.

All witness anchors are independently verifiable using the open-source SWT3 protocol. The verification algorithm is a deterministic SHA-256 computation that can be reproduced by anyone with the anchor string and the protocol specification. Your data remains exportable via JSON, CSV, and OSCAL formats at any time. The Pilot Agreement guarantees 90 days of export access following any cessation notice.

Two recommended resources:

• Guide library: Share the public guide library at sovereign.tenova.io/guides. All 219 guides are accessible without login.
• Assessor Hot Sheet: A 2-page leave-behind designed for assessment meetings. Covers the evidence model, verification steps, and regulatory mapping in a format suitable for handing to a client during an engagement. Available at assessor-hot-sheet.html.