Observe
Keep the origin request unchanged
Accept normalized evidence and record the result without calling customer-facing execution or enforcement paths.
Safe policy simulation
Test Agent-Commerce Policy Before Enforcement
Shadow Mode evaluates what a policy would decide while the merchant's application remains in control. It creates evidence for review without blocking checkout, approving payment, or modifying an order.
Current product boundaryShadow Mode is the current operating model. All results are non-enforcing.
THE OPERATING MODEL
A policy should earn operational trust before it gains authority.
Observed Agent Events are evaluated through the same deterministic policy semantics intended for later decision support. Results are explicitly labelled WOULD_ALLOW, WOULD_BLOCK, WOULD_REVIEW, or WOULD_REQUIRE_AUTH.
Operators compare those counterfactual outcomes with real workflows, investigate false positives, refine evidence requirements, and preserve the reviewed policy version. Activation remains a separate human-owned decision.
WHAT THE RECORD MUST SHOW
Shadow Mode separates learning from customer impact.
Compare
Measure counterfactual outcomes
Review which actions would pass, require authority, need human attention, or be blocked under a policy version.
Decide later
Require explicit activation
Evidence from Shadow Mode informs a later scoped activation; it never activates enforcement automatically.
EVIDENCE FLOW
A safe adoption sequence
- 01Integrate one read-only event source
- 02Normalize and validate evidence
- 03Run deterministic policy in OBSERVE
- 04Inspect Decision Receipts and Timeline
- 05Review gaps and counterfactual outcomes
- 06Make a separate activation decision later
CLEAR PRODUCT BOUNDARY
WOULD_BLOCK is not a blocked checkout.
Shadow Mode language describes a counterfactual policy result. AgentGateFee performs zero customer action in this mode, and the merchant's checkout, payment processor, and order system remain unchanged.
COMMON QUESTIONS
Understand the evidence before changing the workflow.
Will Shadow Mode slow down checkout?
The current public workflow is observational and does not sit in a live enforcement path. Production latency claims require separate evidence for each integration architecture.
Can a merchant activate enforcement from this page?
No. ENFORCE remains separately gated and requires customer-controlled adapters, fallback, rollback evidence, and explicit scoped activation.
What can teams learn in Shadow Mode?
They can inspect evidence quality, policy outcomes, authority gaps, review volume, and differences between observed workflows and proposed rules.
START WITH EVIDENCE