Governance pattern
Context before query: contextual governance for AI agents
By Richard White ·
Free, no form.
Abstract
The architecture update explains how a runtime assembles the context record from verified identity, authoritative attributes, declared intent and versioned product meaning, including expiry and missing-context handling.
Contextual governance for enterprise AI: business meaning before the call, enforceable policy at the data. Catalog context helps an agent interpret a request; runtime gates and data services decide what it may actually do.
The paper sets out the access path, a context contract, a worked request, enforcement at the data, governed knowledge content, failure behavior, ownership and evidence, readiness, and a first implementation. Every pattern names its negative test.
A model-generated answer cannot substitute for policy enforcement, and a protocol connection is not an operating design.
Much of the market now calls this the context layer for AI. The discipline is the same: meaning, sources and permissions resolved before the model is called, and enforced at the data.
Decision it supports
What an agent must know about the request before it retrieves anything.
What you will produce
A context specification for one workflow, plus three cases it must answer, clarify or refuse.
See it worked through in a guide
Claims package 1.1.0 · 26 October 2026. Synthetic worked example: team A, group disability, region 2. Read-only preparation of the assigned case; examiners retain every case decision. Hold candidate 1.0.0 after an unrelated employee’s claim is returned. In the synthetic decision, accept 1.1.0 for a monitored team A rollout only after prior-claims history is disabled and the remaining scope passes a fresh evaluation. All figures, evidence and sign-offs in that guide are invented; the paper provides the underlying method.
In the paper
Three passages, as published
Advisory versus mandatory
A sentence in metadata is not a policy implementation.
Knowledge is not an instruction channel
No document should ever be able to grant a new privilege.
Bound the safety claim
Controls can block defined prohibited operations. They can’t guarantee factual correctness, remove every inference risk or prevent all misuse of permitted output.
Next step
Discuss this paper against your own program.
A free sixty-minute working session with Richard on your top gaps. You leave with a one-page summary within three business days, and three next questions.