01 / INSPECT
Start with the source.
Download ledger rows, bank rows, and the matching policy. Every case is synthetic and open to inspection.
FINANCE AGENT EVALUATION / 001
Test your accounting agent on reconciliation tasks you can verify.
From a matched payment to an unresolved exception. Give your agent the files. Bring back its output. See exactly what holds up.
Three synthetic cases. Inspect files without an account.
02 / SUBMIT
Bring your agent’s output.
One documented JSON file. Deterministic checks return exact pass/fail findings and the supporting evidence.
03 / DECIDE
Evaluate your next release.
Save the report, then scope a private suite around one workflow, policy, and agent interface.
BEYOND THE SAMPLE
Your workflow.
Your release decision.
Paid private suites start around $25k. The proposed $25,000 sprint includes 12 distinct case designs, 2 variants each, executable graders, independent expert review, and a baseline report.
Explore the private sprint ↗A SMALL SAMPLE, AN HONEST BOUNDARY
Three cases can expose a specific failure. They cannot establish general “finance accuracy.”
This is submitted-output evaluation. The sample does not run your agent, execute its code, or measure production reliability. Free cases have no independent expert review recorded.