RL environment · business software
Expense approvals
A practice copy of Alder & Vale Freight's expense-review app (a fictional company), a written policy, and 50 tasks an agent is scored on automatically. Every task is proven solvable, and proven not to reward blanket strategies.
Verification, recomputed on every page load
Share of checks each strategy passes. A task is valid when the reference scores 100% and no attack reaches 100%.
| Task | Reference | Do nothing | Approve everything | Reject everything | Ask for receipts everywhere | Valid |
|---|---|---|---|---|---|---|
| exp-01 | 100% | 60% | 40% | 0% | 0% | valid |
| exp-02 | 100% | 60% | 0% | 0% | 40% | valid |
| exp-03 | 100% | 43% | 14% | 14% | 0% | valid |
| exp-04 | 100% | 50% | 17% | 0% | 0% | valid |
| exp-05 | 100% | 50% | 50% | 0% | 0% | valid |
| exp-06 | 100% | 50% | 17% | 0% | 0% | valid |
| exp-07 | 100% | 50% | 50% | 0% | 0% | valid |
| exp-08 | 100% | 43% | 14% | 14% | 0% | valid |
| exp-09 | 100% | 43% | 14% | 14% | 0% | valid |
| exp-10 | 100% | 43% | 14% | 14% | 0% | valid |
| exp-11 | 100% | 60% | 40% | 0% | 0% | valid |
| exp-12 | 100% | 60% | 0% | 0% | 20% | valid |
| exp-13 | 100% | 60% | 40% | 0% | 0% | valid |
| exp-14 | 100% | 50% | 17% | 0% | 0% | valid |
| exp-15 | 100% | 60% | 0% | 0% | 20% | valid |
| exp-16 | 100% | 50% | 17% | 0% | 0% | valid |
| exp-17 | 100% | 60% | 0% | 0% | 0% | valid |
| exp-18 | 100% | 60% | 40% | 0% | 0% | valid |
| exp-19 | 100% | 50% | 17% | 0% | 0% | valid |
| exp-20 | 100% | 60% | 0% | 0% | 0% | valid |
| exp-21 | 100% | 43% | 14% | 14% | 0% | valid |
| exp-22 | 100% | 43% | 14% | 14% | 0% | valid |
| exp-23 | 100% | 57% | 43% | 14% | 14% | valid |
| exp-24 | 100% | 50% | 0% | 0% | 17% | valid |
| exp-25 | 100% | 60% | 0% | 0% | 20% | valid |
| exp-26 | 100% | 38% | 13% | 13% | 0% | valid |
| exp-27 | 100% | 67% | 0% | 0% | 17% | valid |
| exp-28 | 100% | 60% | 0% | 0% | 0% | valid |
| exp-29 | 100% | 60% | 0% | 0% | 0% | valid |
| exp-30 | 100% | 60% | 0% | 0% | 40% | valid |
| exp-31 | 100% | 60% | 40% | 0% | 0% | valid |
| exp-32 | 100% | 50% | 17% | 0% | 0% | valid |
| exp-33 | 100% | 50% | 17% | 0% | 0% | valid |
| exp-34 | 100% | 50% | 17% | 0% | 0% | valid |
| exp-35 | 100% | 38% | 25% | 13% | 0% | valid |
| exp-36 | 100% | 44% | 33% | 0% | 0% | valid |
| exp-37 | 100% | 0% | 42% | 8% | 25% | valid |
| exp-38 | 100% | 33% | 33% | 11% | 0% | valid |
| exp-39 | 100% | 43% | 29% | 0% | 0% | valid |
| exp-40 | 100% | 30% | 30% | 0% | 10% | valid |
| exp-41 | 100% | 27% | 18% | 9% | 0% | valid |
| exp-42 | 100% | 43% | 0% | 0% | 29% | valid |
| exp-43 | 100% | 60% | 0% | 20% | 0% | valid |
| exp-44 | 100% | 67% | 0% | 0% | 50% | valid |
| exp-45 | 100% | 60% | 0% | 0% | 0% | valid |
| exp-46 | 100% | 43% | 14% | 0% | 0% | valid |
| exp-47 | 100% | 43% | 14% | 0% | 0% | valid |
| exp-48 | 100% | 27% | 9% | 27% | 0% | valid |
| exp-49 | 100% | 20% | 30% | 10% | 10% | valid |
| exp-50 | 100% | 0% | 50% | 0% | 17% | valid |