Module XIII — compliance & regulatory

Module XIII opens enterprise doors by mapping what the control plane already observes and audits onto regulatory frameworks, and by producing auditor-consumable evidence derived from the append-only, hash-chained ledger. It is an intelligence-layer module: it captures nothing new — it aggregates and transforms what the core and the other modules already record, and it never claims certification.

What it is

Module XIII has five surfaces, all read-and-derive over existing data:

• A versioned control catalog held in the repo as the deterministic source of truth — EU AI Act, NIST AI RMF, ISO/IEC 42001, SOC 2 / ISO 27001 and GDPR (plus GenAI/agentic cross-walks), modeled as versioned controls, each with its requirement and its satisfaction criterion. It is a technical mapping, not legal advice, and a control whose obligation the platform cannot evidence carries an explicit note so partial coverage never reads as total.
• A declarative control → evidence map. Every control maps to control-plane capabilities. A capability is either operational — present only when real tenant data exists (a ledger that verifies, observed access edges, security findings, eval results, deployments, a risk classification, a residency attestation) — or architectural — a platform-design guarantee cited to the design docs and labelled as such, never as telemetry.
• Exportable audit evidence — a sealed, append-only evidence package derived from the ledger.
• Agent risk classification into an EU AI Act tier cross-mapped to NIST AI RMF functions, from observed attributes — governed and audited.
• Data residency — a per-region attestation of where the deployment and its stores actually run, plus a scan that turns existing egress signals into a residency finding.

Control status & entities

Status is computed honestly, never asserted. A control is satisfied only when every mapped capability is present and at least one is operational; by_design when all present capabilities are architectural (design-ready, never satisfied); partial when some are present; gap when none are; unmapped when no capability backs it at all. satisfied never rests on design evidence alone.

The module declares four append-only / audited entities in the shared data model: a sealed evidence package (recording the chain head sequence and hash and the live hash-chain verify result), a per-control result inside that package, a risk classification per subject, and a per-region residency attestation. The evidence package references the ledger by sequence and hash and proves its body tamper-evident with a deterministic manifest hash — it never copies the ledger and never holds payloads or PII.

What it consumes & produces

Risk classification reads attributes already recorded by other modules — outbound read/write access edges, high/critical security findings, and an optional autonomy signal — and produces a suggested tier that is governed: a human must review and approve it, and the suggestion engine can never assign the unacceptable tier (that is a legal determination). The residency scan correlates existing egress lineage against self_hosted attestations and, per violation, raises a core finding and publishes an internal bus signal for the notifications module (XV) to deliver to SIEM/Slack/PagerDuty. Reading or exporting an evidence package, sealing one, classifying or reviewing risk, and attesting residency are privileged, tenant-scoped actions that self-audit to the ledger in the caller’s own transaction.

Honest limits

• Designed-for-audit, never certified. Every reporting response carries the disclaimer that it is not a certification and not legal advice. The output speaks of control status and evidence; it never says “compliant” or “certified”. Opt-in guarantees (such as at-rest encryption) default to absent until attested.
• No actuation. This module maps controls and exports evidence — it does not remediate, enforce or change anything. Its only side effect is the residency finding and bus signal, which other modules act on.
• Evidence is only as good as its sources. A control with no backing tenant data is an honest gap, not a faked pass; an absent operational capability lowers a framework’s status rather than inflating it. The least-privilege-drift evidence consumes module III’s reconciled drift (not the raw store path), so it inherits module III’s tiered coverage limits — an absent edge is not proof an access did not happen.
• Architectural evidence is design, not proof. Capabilities cited to the design docs attest how the platform is built, not that a control ran in your tenant; they produce by_design, which is deliberately distinct from satisfied.

Related

• Modules catalog — where module XIII sits and the honest govern/observe-vs-actuate split.
• Module III — access & resource map — the drift signal the risk classifier and the drift capability consume.
• Honesty & limits — why status, not certification.
• Govern and approve — reviewing a suggested risk tier.
• Forward audit to Splunk — the continuous ledger feed the auditor re-verifies against.
• Architecture overview — the intelligence layer and the event bus.