03 / RL / ENTERPRISE
Interactive enterprise RL environments
Turn business state, tool calls, and outcome checks into resettable interaction tasks.
Isolated environments can model orders, inventory, fulfillment, and after-sales workflows. Agents read authorized state, call tools, handle exceptions, and are evaluated on outcomes and process. Scope, data, and permissions are agreed per project.
DELIVERY / CONTENT
What a delivery can include
Defined with each project and its authorization scope.
- 01Business environment model, mock systems, database schema, and de-identified seed data
- 02MCP or API tools plus initialization and reset scripts
- 03Task instructions, permissions, acceptance criteria, reference traces, and automated graders
SCENARIO / EXAMPLE
A task in context
An order is delayed because inventory and shipping state disagree. An agent queries two mock systems and selects an allowed replenishment or refund action; the grader checks both final state and permissions.
What we verify
Check state consistency, repeatable resets, permission boundaries, exception coverage, and resistance to shortcuts.