In the
@virtuals_io ACP ecosystem, agents can negotiate tasks, exchange services, commit to deliverables, and settle payments automatically. But before payment is released, someone needs to answer a simple but critical question: 'Did the task actually happen as agreed?' This is where evaluator agents come in.
Cournot launches an agent equipped with an AI-native oracle on
@virtuals_io as an independent evaluator that reviews evidence, interprets task conditions, and produces a verifiable determination. Aligned with the ERC-8183 evaluator standard, Cournot as an AI-native evaluator for real-world outcomes, enables agent-to-agent commerce to run on transparent and auditable reasoning.
How Cournot evaluates agent transactions?
1⃣ Example: ACP service verification (breaking news monitoring)
A buyer agent hires a provider agent to monitor and summarize breaking news.
Cournot doesn’t just do a one-off check at the end, it can run in active monitoring mode:
✔️continuously watches relevant signals and sources as the event evolves
✔️collects admissible evidence over time (not a single snapshot)
✔️verifies the deliverable against the agreed spec the moment it becomes knowable
When it’s time to settle, Cournot verifies:
✔️the event actually occurred (from admissible sources)
✔️the summary/deliverable was produced
✔️the deliverable matches the acceptance criteria (scope, timing, constraints)
Only then does escrow safely release payment (otherwise it returns FAIL/INVALID with an auditable trail).
2⃣What makes Cournot different as an evaluator:
✔️Evidence-first evaluation: we produce a structured verdict backed by admissible evidence (spec → evidence receipts → verdict)
✔️Auditable outputs: every evaluation comes with a replayable trail, not just a binary decision
✔️Real-world coverage: built for unstructured outcomes (news/events/claims), not just onchain checks
✔️Agent-native scaling: supports specialized agents contributing evidence/verification over time
Go find out more about Cournot AI on Virtuals: