RAG Evaluation Loop

{
  "nodes": [
    { "id":"q","type":"rect","label":"Eval questions\n+ ground truth","x":40,"y":40,"fill":"#dbeafe" },
    { "id":"sys","type":"rect","label":"RAG system","x":260,"y":40,"fill":"#ede9fe" },
    { "id":"a","type":"rect","label":"Generated answers","x":480,"y":40,"fill":"#fef3c7" },
    { "id":"j","type":"diamond","label":"LLM-as-judge","x":700,"y":40,"fill":"#fce7f3" },
    { "id":"h","type":"rect","label":"Human spot-check","x":480,"y":180,"fill":"#dbeafe" },
    { "id":"m","type":"rect","label":"Metrics:\nfaithfulness, context recall,\nanswer relevancy","x":700,"y":180,"fill":"#dcfce7" },
    { "id":"i","type":"rect","label":"Improve\nretriever / prompt / model","x":260,"y":180,"fill":"#fce7f3" },
    { "id":"n","type":"note","label":"Re-run nightly on the held-out eval set","x":40,"y":260,"fill":"#fef9c3" }
  ],
  "edges": [
    { "id":"e1","source":"q","target":"sys" },
    { "id":"e2","source":"sys","target":"a" },
    { "id":"e3","source":"a","target":"j" },
    { "id":"e4","source":"a","target":"h" },
    { "id":"e5","source":"j","target":"m" },
    { "id":"e6","source":"h","target":"m" },
    { "id":"e7","source":"m","target":"i" },
    { "id":"e8","source":"i","target":"sys","style":"dashed" }
  ]
}