GateHouse
All posts
EngineeringJun 4, 2026 · 8 min read

Why we chose Cedar for policy-as-code.

We wanted governance claims that are provable, not merely tested. Cedar's formal-verification properties let the boundary make machine-checkable guarantees.

PT
Paul Turner
Co-founder

When your core promise is that raw data never crosses a boundary, the policy engine that guards that boundary is not a feature. It is the product. We chose Cedar to author it, and the reason is a single word: provable.

Assertion vs. proof

Most governance tooling lets you assert a policy and test it against cases you thought of. That is fine until a regulator, a board, or a counterparty asks a question you did not test. We wanted governance claims that hold by construction — statements about what a policy will and will not release that are machine-checkable, not merely well-intentioned.

Cedar is a policy language with formal-verification properties. Those properties let GateHouse make machine-checkable governance claims rather than assertions — a direct fit with our attestation model, where the same policy runs in two modes.

Enforcement mode

At the boundary, the policy decides: this release is permitted, this one is blocked. Attribute-based, evaluated against jurisdiction × purpose × buyer classification × consent state.

Attestation mode

After the fact — or ahead of a deal — the same policy proves what a release did or would disclose. Not a log that says “trust us,” but a checkable statement derived from the policy itself.

“With Cedar, the boundary’s guarantees are formally provable, not merely tested.”

— GateHouse engineering

What we gave up, and why we didn’t mind

OPA/Rego was the main alternative, and it is excellent for general-purpose policy. But Rego’s expressiveness is also its cost: the more a language can say, the harder it is to prove things about what it says. Cedar deliberately constrains expressiveness to keep decidable analysis on the table. For a boundary whose whole value is that its guarantees are checkable, that trade is not close.

Selected for now. We revisit if the expressiveness ceiling starts to bite on real customer policies — the token contract and policy templates are written to make a migration survivable.

Governance as a first-class surface

Because policy is code with provable properties, we can do something the guided-UI-only tools cannot: show the commercial consequence of a rule as you write it. “With this policy, these datasets become invisible to buyers.” Governance and monetisation get edited in the same view, against the same engine that enforces them.


PT
Paul Turner
Co-founder, GateHouse
Talk to the team →

Move permissions, not data.

See how the thesis becomes a platform, on your own estate.