External Benchmark Adapter
Results submitted by harnesses the arena does not run itself — a Browser Agent Lab, a CI job, a human panel. Aggregated here, but kept as their own category and never blended into a first-party run's score.
| Source | Model | Benchmark | Submissions | Success | Mean score | Steps | Retries | Latency |
|---|---|---|---|---|---|---|---|---|
| browser-agent-lab | Yuu v1.2 (simulated) | browser-tasks | 3 | 67% | 67% | 7.7 | 1.3 | 5.1s |
Submit results
POST a submission (or array of submissions) to /api/external. The payload is validated and normalised; a success flag without an explicit score is scored 1 or 0. There is no hard dependency on the external system — the arena simply aggregates what it receives.
curl -X POST /api/external -H 'content-type: application/json' -d '{
"source": "browser-agent-lab",
"modelId": "yuu-sim-1.2",
"benchmarkId": "browser-tasks",
"benchmarkVersion": "1.0.0",
"taskId": "checkout-flow",
"success": true,
"steps": 7,
"retries": 1,
"latencyMs": 4200
}'