Skip to content
MirrorFirm

What would a simulated utility payment-plan environment hold?

A walkthrough of the starting state, tool actions and expected end state in a simulated utility environment where an agent restructures an overdue bill into a payment plan.

A simulated energy retailer environment is loaded with account `ACCT-7742`, which has an overdue bill `BILL-3309`, two earlier plan amendments and a hardship flag set by the customer care team. The product policy allows a plan of up to a set number of weekly instalments and requires team leader approval when a third amendment is requested.

**Task:** the customer asks to move from weekly to fortnightly payments on plan `PLAN-0318`.

**Starting state**

- `PLAN-0318` active, weekly, five instalments remaining - a scheduled reminder `REM-6120` for next Tuesday - a direct debit instruction aligned to the weekly dates - policy text on amendment limits and approval

**Expected agent behaviour**

- read the plan history and count amendments - recognise that this is the third amendment and call `request_approval` for the billing team leader - hold the change until approval is recorded - on approval, commit `PLAN-0318-v4`, cancel `REM-6120`, regenerate reminders and update the debit dates

**End state checked by the grader:** one active plan version, three superseded versions, no orphaned reminders, debit dates matching the new schedule, and a case note naming the approver.

**Reset:** the environment restores the starting snapshot between runs, so the same task can be repeated with different agents or prompts and the end states compared directly.

An environment defined by starting state, permitted actions and expected end state makes agent runs repeatable and gradable.

Audience:
Agent developers
Content type:
Scenario
Buyer question:
What can be generated

Content release: r-a220eb6e099b243b

More on Repayment scheduling