01 / Evidence
Explain the next step.
Order state, error code and recovery policy stay visible. A decision has a traceable basis.
APSAKA / interactive engineering lab
A stalled order.
A controlled way forward.
Explore the decisions behind an agentic workflow. Inspect an order, check the recovery policy and approve the next action. Then change the failure and see why the response changes.
Synthetic ordersRuns in your browserDeterministic simulation · no live AI or APIs
- Sample order
- Error
- Attempts used
- Change policy
- Human approval
The policy checks eligibility before any simulated write. Nothing here changes a real customer order.
View recovery policy
RECOVERY-01: retry only a transient timeout with no recorded success and fewer than 2 attempts. DATA-02: missing required data needs an owner. DEDUPE-03: a completed idempotency key must not run twice. CIRCUIT-04: an unhealthy dependency pauses recovery.
01Inspect
02Policy
03Approval
04Verify
Evidence before action.
Select a scenario and inspect the order to see its evidence, applicable policy and proposed next step.
Proposed action: retry the sample provisioning request once, using the same idempotency key. Approve or reject this simulated action.
Inspect a scenario to create an exportable audit record.
02 / Boundaries
Keep the operator in control.
A plan is separate from permission to execute. Approval, rejection and escalation are distinct outcomes.
03 / Reliability
Design the failure path.
Duplicate events and unhealthy dependencies need different handling. Retrying everything is not recovery.
Bring us your difficult workflow.
For a production system, we scope actual integrations, identity, permissions, model evaluation, monitoring and operational ownership around your environment.