DGF Consult
Policy-Anchored AI · DGF Consult

Public institutions need AI they can actually defend.

General-purpose AI is useful, but in compliance-sensitive domains it can produce confident answers with no traceable path back to the policy authority that should govern them. Refusing AI leaves officials worse-equipped. Policy-Anchored AI is the third path: AI assists the work, while controlled logic governs the decision path.

Policy-Anchored AI · architecture5 layers
  1. L1

    Human judgment

    An official brings the problem, the intent, and the accountability.

  2. L2

    Structured facts

    The model turns informal input into structured facts — and flags what is missing.

  3. L3

    Policy / authority model

    A versioned, traceable model of rules, thresholds, authorities, and their relationships.

  4. L4

    Controlled reasoning

    Facts are resolved against authority by deterministic logic — not by the model.

  5. L5

    Traceable output

    A conclusion in which every claim links back to a versioned policy element.

The model never enters the path of authority. It assists at the edges; controlled logic governs the core.
The design, in one line

AI assists the reasoning process. Controlled logic governs the decision path.

An architectural distinction, not a marketing one. A system either has this boundary or it does not — and the difference is visible to any reviewer who knows where to look.

01The shift

The new transformation service is not a report. It is a framework.

Traditional professional services produce advice, decks, and documents. The next generation will produce structured frameworks, decision systems, testable logic, and traceable outputs.

Policy-Anchored AI is my attempt to make that shift concrete — a defensible way to put AI to work in the decisions institutions cannot afford to get wrong. GC Procurement Advisor is the first place it runs.

02The problem

Useful most of the time. Defensible none of the time.

Public institutions are pushed to adopt AI in exactly the places it is hardest to defend — procurement, privacy, security authorization. The real risk is not hallucination. It is uncontrolled reasoning.

i

Uncontrolled reasoning

A fluent answer that quietly skips a required step, misses a threshold, or invents an obligation — in a way no reader can detect from the output alone.

ii

No traceable authority path

A confident conclusion with no link back to the policy, threshold, or authority that should govern it. Useful, perhaps. Defensible, no.

iii

No stable answer under review

Re-phrase the request and the answer moves. In a compliance domain, a system that can be talked into a different conclusion can be gamed.

03The framework

Policy-Anchored AI

Authority lives in policy, not the model — a versioned, traceable model of rules, thresholds, and authorities. The model does what it is best at. The boundary is enforced, not advised.

D1

Fact discipline

It separates what is known, inferred, missing, not applicable, and provisional — and is allowed to say “not enough information”.

D2

Traceable conclusion

Every conclusion links back to the policy elements, versions, and reasoning behind it — reproducible six months later.

D3

Identity-stable output

Identical situations produce identical conclusions, regardless of who asks, when, or how they phrase it.

04Behind the framework

Built from transformation experience.

Policy-Anchored AI comes from decades of work inside public-sector transformation — IT modernization, procurement and governance reform, and large multi-stakeholder delivery across the Government of Canada.

The same pattern recurs: major initiatives fail when facts, authority, governance, delivery, procurement, and accountability are not connected early enough. The framework is that lesson, turned into an architecture.

25+ years
Public-sector transformation and IT-enabled change
38+ / 80+
Departments and agencies — Canada.ca Managed Web Service transition planning
Chief Negotiator
Former — internal trade, NAFTA, and WTO-related files
First implementation
GC Procurement Advisor — working prototype of Policy-Anchored AI
05 — The proof

GC Procurement Advisor

The first working application prototype of the Policy-Anchored AI framework — tested against Government of Canada procurement scenarios, and available for structured walkthroughs and pilot discussions.

Procurement is the proving ground because it stress-tests the framework against policies, thresholds, vehicles, trade agreements, privacy, security, accessibility, official languages, governance triggers — and human accountability.

06The paper

The positioning paper

A public-facing paper on compliance-grade decision support: the problem, the three operating disciplines, a worked procurement example, and how the pattern should be evaluated.

Working draft · May 2026
“A confident answer with no traceable path back to the policy that should govern it. In a compliance domain, that is the wrong trade.”

Policy-Anchored AI — DGF Consult

07Engagement

Five ways to engage.

Engagements are framed around the framework — not a menu of generic services.

  • 01

    AI transformation advisory

    For leaders deciding where AI can safely change professional services, decision support, intake, compliance, and governance.

  • 02

    Policy-anchored system design

    For organizations that need AI-assisted tools in domains where accuracy, auditability, and defensibility matter.

  • 03

    Compliance-grade prototype development

    For teams that want to turn an ambiguous advisory process into a structured, testable, policy-anchored prototype.

  • 04

    Executive review and challenge

    For leaders who need an independent read on whether an AI idea is real, safe, governable, and worth funding.

  • 05

    Public-sector procurement & delivery strategy

    Available as one application of the broader capability — not the centre of the work.

Suited to senior advisory work, structured discovery, pilot design, and implementation planning.

Start a conversation

Bring the hard problem.

If the work involves policy, judgment, governance, and AI, that is exactly the kind of conversation worth having.