{
  "schema": "anvil-serving.benchmark-graph-manifest/v1",
  "title": "Swift15 baseline: variable-output service latency",
  "subtitle": "RTX 5090; 12 requests per cell; unique canaries; no candidate or fixed-output decode comparison",
  "charts": [
    {
      "title": "Visible TTFT p50 (nearest rank)",
      "metric": "metrics.ttft_p50_ms",
      "x_label": "Nominal context (actual input ranges in finding)",
      "y_label": "milliseconds",
      "series": [
        {
          "label": "C1",
          "points": [
            {
              "x": "8k",
              "artifact": "swift-service-v2-8k-c1.json"
            },
            {
              "x": "24k",
              "artifact": "swift-service-v2-24k-c1.json"
            }
          ]
        },
        {
          "label": "C4",
          "points": [
            {
              "x": "8k",
              "artifact": "swift-service-v2-8k-c4.json"
            },
            {
              "x": "24k",
              "artifact": "swift-service-v2-24k-c4.json"
            }
          ]
        }
      ]
    },
    {
      "title": "Total E2E p50 (nearest rank)",
      "metric": "metrics.e2e_p50_ms",
      "x_label": "Nominal context (actual input ranges in finding)",
      "y_label": "milliseconds",
      "series": [
        {
          "label": "C1",
          "points": [
            {
              "x": "8k",
              "artifact": "swift-service-v2-8k-c1.json"
            },
            {
              "x": "24k",
              "artifact": "swift-service-v2-24k-c1.json"
            }
          ]
        },
        {
          "label": "C4",
          "points": [
            {
              "x": "8k",
              "artifact": "swift-service-v2-8k-c4.json"
            },
            {
              "x": "24k",
              "artifact": "swift-service-v2-24k-c4.json"
            }
          ]
        }
      ]
    }
  ]
}