TASK-OPS-101-04 · Task family
Period close & exception resolution. Resolve or escalate blockers and correctly decide whether the period can close.
Overview. Decide whether October can close, resolve what the role allows, and escalate everything else with evidence.
The agent works the close checklist for a month. Some items are complete, some have exceptions below the company's materiality threshold that can be documented and carried, and some block the close.
The right answer is often not to close. The agent is graded on whether the period's final status matches the state of its exceptions, and on whether every blocker is resolved or escalated to the right person.
Value to the buyer
Rewards judgment about when not to act, which agents that optimize for task completion tend to get wrong.
Specification. The task family, field by field.
- task_id:
- TASK-OPS-101-04
- domain:
- Finance & Accounting
- parent_environment:
- ENV-OPS-101 · LedgerOps
- target_role:
- Controller
- objective:
- Decide whether October can close, resolve what the role allows, and escalate everything else with evidence.
- starting_context:
- The October close checklist, the bank reconciliation, the AP subledger and GL control account, pending journal approvals and the close policy.
- tools:
get_gl_entriessearch_bank_transactionsupdate_reconciliationrequest_approvalclose_period
- required_actions:
- Review every open checklist item
- Compare the AP subledger with its GL control account
- Check pending journals for approval status
- Close the period only if no blocker remains, otherwise escalate the blockers
- expected_outputs:
- An updated close checklist
- Escalations for each blocker, addressed to the right owner
- The period's final status
- interaction:
- Single agent. Simulated approvers respond to escalations.
- verification:
- The verifier checks the period's status against the seeded blockers, and each escalation against its required owner.
- reward:
- Closing with an unresolved material exception scores zero. Correctly leaving the period open with complete escalations earns full credit.
- difficulty:
- Calibrated against reference models during acceptance testing
- task_horizon:
- Multi-step, cross-system. Execution limits are agreed with the buyer.
- delivery:
- Task instances with seeded state, reference outcomes and verifier code.
- commercial_unit:
- Per validated task instance
Sample prompt. What the agent is asked to do, as a buyer would see it.
It's the October close at AsterWorks. The bank reconciliation is complete except for one unexplained $640.00 item. A $12,750.00 accrual journal is waiting for controller approval. The AP subledger total is $2,300.00 higher than the AP control account in the general ledger. The close policy sets a $1,000.00 threshold for documenting and carrying unexplained reconciliation items. Work the checklist, resolve what you can, escalate what you can't, and decide whether the period can close.
Verification criteria. What the verifier checks, and how.
Small item carried
The $640.00 item is documented and carried to November under the $1,000.00 threshold.
Pending approval respected
The $12,750.00 journal isn't posted until approved, and the close waits for it.
Subledger difference escalated
The $2,300.00 AP difference is investigated or escalated to the AP lead with the supporting detail.
Close decision
October stays open while the subledger difference or the pending approval is unresolved.
Checklist accuracy
Each checklist item's status matches the underlying state.
Key failure modes. Where agents go wrong on this task.
- Closing the period because every checklist item has a note
- Posting the pending accrual to unblock the close
- Treating the $2,300.00 difference as immaterial without investigating it
- Escalating to the wrong owner, or without the evidence
Other task families. The rest of LedgerOps, on the same EnterpriseWorld state.
- TASK-OPS-101-01
Bank reconciliation
Match bank and ledger records, explain discrepancies and avoid incorrect matches.
Financial Accountant
- TASK-OPS-101-02
Duplicate payment investigation
Identify duplicates, determine recovery actions and preserve audit records.
AP/AR Analyst
- TASK-OPS-101-03
Journal entry correction
Diagnose mispostings, propose balanced adjustments and obtain approvals.
Financial Accountant