eval_752
  • English

      • Runs + Browser Harness REST API

      The /runs and /browser-harness endpoints power the active runs board, Browser Harness import flow, archive, comparison, and drill-down views.

      Endpoint Summary

      MethodPathDescription
      POST/runsCreate a new run
      GET/runsList runs for archive and analysis views
      GET/runs/activeReturn active-run snapshots for active and recent runs
      GET/runs/{run_id}Return run detail summary
      GET/runs/{run_id}/logsReturn run logs
      GET/runs/{run_id}/itemsReturn grouped item results with rich metadata
      POST/runs/{run_id}/retryDispatch a retry
      POST/runs/{run_id}/cancelCancel a pending or running run
      POST/browser-harness/packsBuild a signed Browser Harness pack
      POST/browser-harness/importsImport Browser Harness captures as runs

      GET /runs/active

      This endpoint returns the read model used to hydrate the active runs board before SSE begins.

      Query Parameters

      NameTypeDefaultNotes
      limitinteger 1-2412Maximum number of active snapshots returned

      Response Shape

      [
  {
    "run": {
      "id": "run-123",
      "providerName": "OpenAI",
      "datasetName": "MMLU-Pro",
      "modelName": "gpt-4.1",
      "status": "running"
    },
    "progress": {
      "completed": 14,
      "total": 100,
      "correct": 11,
      "incorrect": 3,
      "awaiting": 1,
      "faulted": 0,
      "pending": 85
    },
    "lifecycle": {
      "resumedCount": 1,
      "lastResumedAt": "2026-03-06T19:43:11Z",
      "cancelRequestedAt": null
    },
    "currentItem": {
      "sequence": 15,
      "state": "running",
      "promptText": "Which option best describes...",
      "choices": ["A", "B", "C", "D"]
    },
    "recentItems": []
  }
]

      Notes

      • progress.awaiting counts answered items that have not been judged yet
      • progress.faulted counts terminal items that failed or were canceled
      • lifecycle is persisted server-side and used to rebuild the active board after refresh
      • currentItem.activity may include streaming previews and chunk metadata when available

      Client Integration Notes

      1. useRunsQuery drives archive and comparison selectors
      2. useActiveRunsQuery hydrates the active board before SSE begins
      3. useRunEvents merges run_status, run_progress, and run_item payloads into local state
      4. on transient SSE errors, clients should invalidate active caches but keep reconnection possible
      5. the Runs page treats /runs/active as the primary board source and fetches full run detail on demand