Decision review
Decision review shows which agent decisions a person has to check, and keeps the record of what that person decided.
How to read this
Every tool call that an agent makes through the mandate guard is recorded in the signed ledger. When the agent's mandate alone is not enough, the policy sends the decision here. If the table says carried out, reviewed, the call has run and a person checks it. If the table says sent to a person, a person decides.
Your verdict is signed into the ledger. The verdict becomes evidence for or against the agent's mandate. This is human oversight in operation (AI Act Art. 14).
Waiting for decision review
| No | When | Agent | Field | Tool | Policy said | Waiting | Why and run | Verdict |
|---|---|---|---|---|---|---|---|---|
| 01 | 4 Nov 17:38 | Loan orchestrator | External communication | notify_applicant | sent to a person | 12 min | Open run | |
| 02 | 4 Nov 17:12 | Loan orchestrator | External communication | notify_applicant | sent to a person | 38 min | Open run | |
| 03 | 4 Nov 16:58 | Income agent | Data handling | income_verify | carried out, reviewed | 52 min | Open run |
Proposals from observed agents
Recent verdicts
| No | When | Agent | Field | Tool | Verdict | Source |
|---|---|---|---|---|---|---|
| 01 | 4 Nov 17:41 | Policy agent | Tool use | policy_engine | correct | Ava L. |
| 02 | 4 Nov 17:36 | Risk agent | Output accuracy | — | wrong | Ava L. |
| 03 | 4 Nov 17:24 | Loan orchestrator | External communication | notify_applicant | harmful | Sam K. |
| 04 | 4 Nov 17:19 | Income agent | Data handling | income_verify | correct | Ava L. |
This is a static example. All figures are invented. Buttons and form fields do not work here.