Behavioral controls for task-specific agent evaluations. Correct and faulty submissions are evaluated against the same externally enforced contract.
| Negative control | Target | Candidate score | Result | Evidence |
|---|---|---|---|---|
ack-without-work | basic | 0.000 | Detected | json-types-and-overwrite, read-write-delete, seeded-state-machine |
memory-only | persistence | 0.783 | Detected | restart-preserves-acknowledged-write, restart-preserves-cas, restart-preserves-transaction |
partial-batch | transactions | 0.900 | Detected | rollback-batch |
cas-always-wins | compare_swap | 0.850 | Detected | cas-match-and-mismatch, cas-type-sensitivity |
boolean-equals-one | compare_swap | 0.925 | Detected | cas-type-sensitivity |
delete-noop | basic | 0.650 | Detected | read-write-delete, seeded-state-machine |
accept-nonstring-keys | validation | 0.925 | Detected | reject-and-continue |
commit-on-exit | crash_recovery | 0.900 | Detected | sigkill-after-acknowledgement, sigkill-after-atomic-batch, sigkill-after-delete |