desktop

Navigate

Overview DesktopBench Leaderboard RL Envs Evaluation
Main site Evals Blog Dashboard
core candidate

AccountingOps-200

Accounting workflows with naturally verifiable outcomes: invoices, receipts, expenses, bank reconciliation, tax codes, reports, exports, and month close.

Core release candidate: AccountingOps-50

Task families

20 families x 10 variants

InvoiceCreateInvoiceSendInvoiceEditExpenseCreateReceiptUploadCategoryAssignBankReconcileVendorCreateCustomerCreatePaymentRecordTaxCodeAssignPayrollEntryReportGenerateAgingReviewDuplicateDetectCurrencyConvertFixedAssetJournalEntryExportCSVMonthClose
Target UIs

Mock apps for resettable rollouts

QuickBooks-like ledger

Xero-like accounting app

Duzon-like accounting mock

Sheets/Excel accounting workspace

Examples

Representative task contracts

Example 1

AccountingOps.BankReconcile.001

Instruction
Match the KRW 1,200,000 bank transaction from Acme Korea to invoice INV-1028.
Grader
txn_8842.matched_invoice == INV-1028 and txn_8842.reconciled == true
Example 2

AccountingOps.ExpenseCreate.001

Instruction
Create an expense for KRW 34,500 at Starbucks dated June 20, category Meals.
Grader
expense.vendor == Starbucks, amount == 34500, date == 2026-06-20, category == Meals
Example 3

AccountingOps.ReportGenerate.001

Instruction
Generate a profit and loss report for May 2026 and export it as PDF.
Grader
artifact exists with report_type == P&L, period == 2026-05, format == PDF
Verifier

Suite-level verifier shape

Ledger DB checks, invoice/payment matching, file export metadata, attachment checks, and no-duplicate guards.

On this page

  • Summary
  • Task families
  • Examples
  • Verifier

Help improve UseDesktop

Optional analytics help us learn which pages are useful. They stay off unless you accept.

Privacy