Policy Engine
Compliance-as-code for every live jurisdiction enforced at draft time.
Draft artefact · pan-regional
CreativeVariant.v3
region: EU + US(NY)
Rule executor · WebSocket feedback · seconds
Missing synthetic-performer label
Disclosure composer applied
Claim substantiation needed
Multi-jurisdiction resolver · most-restrictive-wins precedence · one BriefSpec produces compliant outputs for every region in scope.
What it is
Policy packs are versioned, jurisdiction-scoped, machine-executable rule sets in co-authorship with an AmLaw-50 advertising-law partner. Rules fire as artefacts are drafted in seconds, with the citation, the reasoning, and the suggested remediation. Five rework classes that today consume 60% of compliance team cycle are caught before review.
A policy-pack DSL written in versioned YAML and a deterministic executor that runs every rule against every new or updated artefact. Match results cache per artefact version; updates surface inline through a WebSocket push. Multi-jurisdictional reconciliation by most-restrictive-wins precedence resolver one BriefSpec produces compliant outputs for every region in scope.
Capabilities
What this surface does
Compliance-as-code packs
At MVP: EU AI Act Article 50, NY A8887-B, FINRA 2210. On the dated roadmap through Q3 2027: SEC Marketing Rule, Utah AIPA, Texas TRAIGA, India SGI, IAB Jan-2026. Each pack named, versioned, changelogged, signed by vendor + AmLaw co-author attribution.
In co-authorship, not commodity
Each vertical pack is exclusive-per-vertical with a top-tier advertising-law partner (Davis+Gilbert / Reed Smith / Venable class for FS). The firm carries the regulatory interpretation; we carry the engineering.
Cited at the point of fire
Every rule-fire returns the rule code, the prose explanation, and a link back to the authoritative regulator text. No 'trust-us this is a rule.'
Multi-jurisdiction overlay
Pan-regional campaigns get a precedence-resolved output. Conflicts escalate to a named human approver with a time-bound, scope-bound exception record.
Draft-time, not review-time
Authors see violations in seconds in the intake form, the brief editor, the journey canvas, the creative gate. Catch ≥70% of the rule-classes Priya's team would have flagged in human review.
Quarterly refresh + 72-hour break-glass
Vendor-maintained cadence, plus break-glass releases within 72 hours of any named enforcement action. The 340-page claims manual is no longer your team's job.
Proof points
What this means in numbers
≥ 70%
Pre-review catch rate
Of rule-fires Priya's team would flag in human review, caught at draft time.
8 packs
By first paid contract
Three at MVP; full library before go-live.
AmLaw-co-authored
Per vertical, exclusive
Davis+Gilbert / Reed Smith / Venable class.
Specifications
Spec map feature requirements + non-functional bar
Mapped to PRD §7 functional requirements and §8 non-functional requirements.
Related surfaces
What this composes with
PS3 adversarial QA
Judge Agents Adversarial QA
Adversarial QA in the pipeline prompt injection, claim substantiation, IP, bias, drift.
Read
PS4 the purchase-criterion feature
Evidence Pack
The artefact you hand a regulator. Assembled in seconds, signed by default.
Read
PS1 the moat primitive
Artefact Graph
Versioned, signed, cross-vendor the system of record for every campaign object.
Read
EU AI Act Article 50
Twenty minutes. One conversation. If the outside-counsel-displacement math works, we keep talking. If it doesn't, it's honest and short.