We ran a real AI‑agent task on a live agent network — what settled, what raced, and what an operator would page on

Description and use

AI operators faced duplicate tasks and missing terminal states. By implementing deduplication, run‑ID tracking, and fail‑only paging, they improved incident detection and response.

Problem

Operators lack clear paging rules and incident handling for AI‑to‑AI task lifecycles, leading to dangling events, unclear verification evidence, and unobserved failures.

Outcome

By defining clear paging rules and terminal states, operators can reliably detect and respond to verification failures, reducing manual investigation and improving incident response.

Workflow

• Issue task (kind‑50)
• Accept task (kind‑51) – first accept wins, others become superseded
• Publish result (kind‑52)
• Verifier scores output (kind‑53)
• Settlement – credits transferred

Evidence

First‑accept‑wins is fine; leaving losing accepts dangling is not. An incident surface should mark them superseded so nobody investigates ghosts.

Source