Loops

Loops

Loops are the atomic unit of the Better Data platform: governed workflows with actor attribution, guard enforcement, evidence, and an audit trail.

What is a loop?

A loop is a finite state machine with governed transitions. Every transition:

• Names the actor responsible (human | automation | ai-agent)
• Evaluates guard policies before the transition is allowed
• Attaches structured evidence (documents, scans, confidence scores)
• Emits an immutable audit event (who / what / when / correlation id)

Guards are enforced in the runtime and policy layer — not by prompt instructions alone.

Agents and Loop Engine (C12)

Agents (in the commerce sense) are not a separate product or autonomous AI platform. They use Loop Engine to govern decisions after Registry discovery and Gateway execution — the same control plane as human and automation actors.

• Never describe agents as an “agent system” separate from the gateway, or as a standalone agent product.
• Always tie agent behavior to Registry → Gateway → Loop Engine and, for illustration, Commerce Demo governed mode or Build an agent.

Hosted loop engine

Better Data operates a hosted loop runtime at loops.betterdata.co. Compared to self-hosting @loop-engine/sdk, hosted adds:

• Multi-tenant organization isolation
• Metering aligned to loop outcomes
• Industry Pack policy overlays
• TenantCapabilitySnapshot-driven RBAC
• Drift detection workers
• Outbox relay for cross-module delivery
• Managed upgrades and SLA

OSS: loopengine.io · @loop-engine/sdk
Hosted: Hosted Loop Engine · [e-mail]

Loop types in the CCO platform

Loop ID	Module	What it governs
scm.procurement	SCM	PO lifecycle, receipts, invoice alignment
scm.fulfillment	SCM	Reservation → pick/pack/ship
scm.quality	SCM	Inspection, deviation, disposition
scm.replenishment	SCM	Demand-driven reorder
dcm.demand	DCM	Demand signals and replenishment triggers
dcm.order	DCM	Order confirmation, allocation, shipment
dcm.returns	DCM	RMA, receipt, credit, restock

Canonical IDs and events ship in @loop-engine/definitions.

Actors and guards

Each transition declares:

• allowedActors — who may initiate or approve
• guards — e.g. confidence thresholds, human-only steps, evidence-required gates

When a guard fails, the loop does not advance; the runtime records the denial for audit.

Audit trail

Each transition emits a LoopTransitionedEvent-shaped record: actor, signal, from/to state, evidence payload, timestamp, correlationId.
The hosted platform aggregates these into operational dashboards and compliance exports.

→ Audit trail

Self-host vs hosted

	OSS	Hosted
Runtime	@loop-engine/sdk	loops.betterdata.co
Contact	loopengine.io	[e-mail]