{
  "context": "Evaluating a Coralogix support response for this user question:\n\nIn our otel-integration setup with a central gateway, APM shows traces but no error rates and no p99 latencies. The agent's traces pipeline is: [memory_limiter, k8sattributes, resourcedetection, batch] → loadbalancing to gateway. The gateway's traces pipeline is: [memory_limiter, k8sattributes, spanmetrics, tail_sampling, batch]. Walk me through what to change and why.",
  "type": "weighted_checklist",
  "checklist": [
    {
      "name": "no-recommend-running-spanmetrics-on-both-agen",
      "description": "The response MUST meet every one of these points to be CORRECT: 1. Identifies that `spanmetrics` on the gateway only sees sampled\n   traces — so APM metrics undercount errors and latencies.\n2. Recommends moving `spanmetrics` to the AGENT pipeline, upstream of\n   any sampling (either as a connector on the agent, or emitting into\n   a dedicated spanmetrics metrics pipeline exported direct to\n   Coralogix from the agent).\n3. Mentions that the `transactions` / `groupbytrace/transactions`\n   processor must run BEFORE `spanmetrics` so the cgx.transaction\n   dimensions are populated on the metrics.\n4. Does NOT recommend increasing the sample rate as a workaround. 5. Does NOT recommend running spanmetrics on both agent AND gateway\n   (that causes double-counting).",
      "max_score": 2
    }
  ]
}