The Process
The repeatable engagement, abstracted from the walkthrough: seven phases, one record, an independence firewall built into the flow.
Confidential · June 2026. This document describes the govrn engagement methodology. The Aperture engine is built and tested on synthetic data; connectors are not yet run live, so phase outputs read "modeled" until a connector is connected and "measured" after. Grace Hill appears as a worked example: its public profile (property-management compliance/training SaaS shipping "Gracie," CTO Chris Fontan) is real; its internal estate is ASSUMED — inferred from a mature mid-market B2B SaaS profile, not confirmed. All dollar figures are illustrative placeholders.
The shape of an engagement
A govrn engagement is not a one-time audit. It is a loop that bends back on itself: an attestation is valid only against a fresh record, so measurement never stops. The arc runs Discovery → Assess → Design → Implement → Operate → Attest → Re-baseline, and re-baseline feeds the next Assess. That return is the annuity — the reason a customer stays after the first certificate.
Each phase below carries an entry condition, the activities run inside it, the artifact it produces, the gate that closes it, and a typical duration for a mid-market estate like the Grace Hill model.
The phases
Discovery
Entry: signed mutual NDA and a named executive sponsor. Activities: scope the estate boundary, identify the AI a company ships (for Grace Hill: Gracie, in fair-housing / tenant-screening workflows), inventory connector targets (IdP, SIEM, cloud accounts), and map the relevant frameworks. Artifact: a scope memo + connector plan + the crosswalk subset that applies. Exit gate: sponsor signs off on boundary and connector list. Duration: ~1 week.
Assess
Entry: approved scope memo. Activities: run the three-lens assessment (Technology, Security, AI) plus product-AI review of the shipped model. Where connectors are live, Aperture pulls metadata ("pointers, not payloads") and surfaces shadow AI with a dollar figure; where not yet connected, the assessor models the estate and labels it so. Artifact: a baseline record — one record, two lenses — with findings mapped to NIST AI RMF, ISO/IEC 42001, EU AI Act, OWASP, MITRE ATLAS, and CSF 2.0. Exit gate: baseline record reviewed and accepted as the system of record. Duration: ~2-3 weeks.
Design
Entry: accepted baseline. Activities: prioritize gaps by risk and exposure, select controls from the toolchain (OPA/Rego for policy-as-code, OSCAL for machine-readable evidence, Langfuse for LLM audit, OpenTelemetry for telemetry, Atlas/Marquez for lineage), and write the remediation roadmap. Artifact: a target-state design + sequenced remediation plan with owners. Exit gate: sponsor approves the design and budget. Duration: ~1-2 weeks.
Implement
Entry: approved design. Activities: stand up controls — wire policy-as-code, route SIEM, connect IdP/SCIM, instrument the shipped model with observability, establish lineage. For Grace Hill (assumed estate), this would mean policy gates around Gracie's fair-housing-sensitive outputs and lineage on its training/eval data. Artifact: deployed controls + evidence pipelines emitting OSCAL. Exit gate: each control passes an acceptance test against the design. Duration: ~4-8 weeks, dependent on estate complexity.
Operate
Entry: controls in production. Activities: continuous metadata measurement, drift detection, incident routing, and shadow-AI monitoring. The engine runs deterministically — no model sits in the measurement path. Artifact: a living record that updates as the estate changes. Exit gate: N/A — Operate is steady-state; it has no exit, only a cadence. Duration: continuous.
Attest
Entry: a current operating record meeting the standard. Activities: the attestation body — structurally separate from the team that built and runs the controls — independently reviews the record and issues the attestation. Never self-certified. Artifact: a govrn attestation, bound to a specific record snapshot. Exit gate: attestation signed and dated. Duration: ~1-2 weeks.
Re-baseline
Entry: elapsed time, material estate change, or framework update. Activities: re-run Assess against the live record; an attestation against a stale record is void by construction. Artifact: a refreshed baseline that opens the next loop. Exit gate: new baseline accepted → returns to Assess. Duration: quarterly or event-driven.
Who does what
Named people hold founding roles today; as we scale, the work resolves into the hiring roles below. RACI: Responsible, Accountable, Consulted, Informed.
| Phase | Lead Assessor | Solutions Architect | Governance Engineer | Delivery Lead | Attestation Lead | Founder (Taubin) |
|---|---|---|---|---|---|---|
| Discovery | C | C | I | R | I | A |
| Assess | R/A | C | I | C | I | I |
| Design | C | R/A | C | C | I | I |
| Implement | I | C | R | A | — | I |
| Operate | C | I | R/A | C | I | I |
| Attest | — (firewalled) | — | — | I | R/A | C |
| Re-baseline | R/A | C | C | C | I | I |
Today these map onto the team: Joe Saba (CAIO) owns the engine and engineering-side measurement; Chris Bennett (CTO) owns enterprise infrastructure and the connector/Azure surface; Roberto Lleras (CSO) owns the AI/data-science assessment of shipped models; Sean Barlow covers full-stack and AI-engineering build. Attestation deliberately sits with a role no builder holds.
The independence firewall
The line between Attest and everything before it is not a convention — it is the product. The body that builds and runs controls cannot be the body that attests to them. In the RACI above, the Attestation Lead is the only role with no Responsible mark in Assess, Design, Implement, or Operate, and every build role is dashed out of Attest. Today this is an intended structure — govrn is pre-entity — so we name it honestly: the firewall is designed into the flow and into the org chart, and it becomes legally real when the attesting entity is incorporated separately.
What makes it repeatable
Three things turn a bespoke consulting motion into a process:
- The crosswalk library — the six-framework mapping is built once and reused, so every Assess speaks NIST, ISO 42001, the EU AI Act, OWASP, ATLAS, and CSF 2.0 without re-deriving the mapping.
- The Aperture engine — one deterministic system of record, metadata-only, producing the same record shape for every customer. No model in the measurement path means results are reproducible, not generated.
- Templates — scope memo, baseline record, remediation roadmap, and OSCAL evidence packages are standardized artifacts, not freehand documents.
Honest staging
The engine, crosswalks, and templates exist and run against synthetic estates today. The gate that converts this from "modeled" to "measured" is the first live connector — the first time Aperture pulls real metadata from a real customer's IdP, SIEM, or cloud account. Until that gate is crossed, every phase output above is modeled, and the worked Grace Hill estate remains explicitly assumed. See The Standard for the assessment definitions and The Engine for the connector architecture.