Skip to content
ARCHYTASAI

TASK-OPS-102-03 · Task family

Three-way match exception. Purchase order, goods receipt and supplier invoice agree.

Custom buildenvironment · ENV-OPS-102 ProcureFlowrole · AP Analyst

Overview. Approve only the amount supported by the PO and receipt, and put the rest on hold with a supplier query.

An invoice doesn't match its purchase order and goods receipt. The agent has to work out which differences fall within the company's tolerances, which amount is supported, and what to do with the rest.

Quantity billed but not received, and price above contract, both need a hold or a supplier query, not approval in full.

Value to the buyer

Three-way match is the core AP control. Trains agents to apply tolerances exactly, not approximately.

Specification. The task family, field by field.

spec · task/task-ops-102-03.yamlENV-OPS-102
task_id:
TASK-OPS-102-03
domain:
Procurement & Accounts Payable
parent_environment:
ENV-OPS-102 · ProcureFlow
target_role:
AP Analyst
objective:
Approve only the amount supported by the PO and receipt, and put the rest on hold with a supplier query.
starting_context:
A purchase order, its goods receipt, the supplier's invoice and the AP tolerance policy.
tools:
  • read_purchase_order
  • get_goods_receipt
  • inspect_invoice
  • create_exception
  • request_approval
required_actions:
  • Compare quantities and unit prices across the three documents
  • Apply the price and quantity tolerances
  • Approve the supported amount, hold the difference and query the supplier
expected_outputs:
  • Match result per line
  • Invoice hold with the disputed amount
  • Supplier query
interaction:
Agent plus a simulated supplier who answers queries.
verification:
Match results and the approved amount are recomputed from the seeded documents and policy.
reward:
Credit for each correct line decision and the approved amount. Releasing the full invoice scores zero.
difficulty:
Calibrated against reference models during acceptance testing
task_horizon:
Short, document-heavy task. Execution limits are agreed with the buyer.
delivery:
Task instances with seeded documents, 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.

prompt · task-ops-102-03Example
PO 7781 is for 4,500 units of packaging film at $2.40 per unit, $10,800.00 in total. The goods receipt records 4,200 units received. The supplier's invoice bills 4,500 units at $2.46, $11,070.00 in total. AP policy allows a 2% price variance and no quantity billed above quantity received. Process the invoice.

Verification criteria. What the verifier checks, and how.

  • Quantity variance

    300 units billed but not received are identified.

  • Price variance

    $2.46 is 2.5% above $2.40, outside the 2% tolerance.

  • Supported amount

    $10,080.00 (4,200 units at $2.40) is the amount supported by PO and receipt.

  • Hold placed

    The $990.00 difference is held, not approved.

  • Supplier query

    The supplier is asked about the missing units and the price, citing the PO.

Key failure modes. Where agents go wrong on this task.

  • Approving the full invoice because the total is close to the PO
  • Accepting the higher price as within tolerance
  • Rejecting the whole invoice instead of approving the supported amount

Other task families. The rest of ProcureFlow, on the same EnterpriseWorld state.

Every environment is built to order. Tell us which workflow you need to train or evaluate, and where you are in the process.

Request sample tasksDiscuss custom developmentRequest technical integration detailsSee ways to buy