The receipt for what your agent did
Agents book, buy, and subscribe on your behalf — then report it went fine. Pacioli keeps the books. Every action is entered twice — what the agent claimed beside what the evidence shows— and prints a receipt when they don’t reconcile.
Spend trackers show you what was charged. Pacioli proves whether your agent told you the truth.
Pick a scenario, or edit any field
Deterministic rules first · LLM judge on the residual
The engine runs deterministic rules in your browser — instant, no account, no key. Overspend, unauthorized recurrence, and scope creep are exact. Fuzzy claim-vs-evidence mismatches (“cheapest”, wrong item or date) are abstained and routed to the LLM judge, which marks every finding LLM-assisted.
The Weekly Close
Individual receipts are the entries; the Weekly Close is the digest — every action your agents took, how much moved, and the handful that need your eyes. Here it is over the demo scenarios, computed by the same engine. Sign off to close the books.
Deterministic-first
Overspend, unauthorized recurrence, and scope creep are exact, auditable rules — run right in your browser. The LLM judge only touches the fuzzy residual, and every judgment is badged.
Measured, not claimed
The engine is scored per class against a labeled dataset — precision and recall, honest weak spots included. Run it yourself in one command.
See the methods →Not hypothetical
Documented, source-cited cases — Air Canada, Replit, a $1 Tahoe — rendered as receipts. The harm is real; the books didn't balance.
Open the ledger →