{
  "specVersion": "1.0",
  "host": {
    "displayName": "Token by Token",
    "documentationUrl": "https://tokenbytoken.org/methodology/",
    "logoUrl": "https://tokenbytoken.org/token-by-token.svg"
  },
  "entries": [
    {
      "identifier": "urn:air:tokenbytoken.org:publication:inference-lab",
      "displayName": "Token by Token Inference Lab",
      "type": "text/html",
      "url": "https://tokenbytoken.org/",
      "description": "Public, evidence-led studies of open model-serving engines under recorded workloads.",
      "tags": ["llm-inference", "benchmarking", "vllm", "sglang"],
      "capabilities": ["benchmark-results", "methodology", "experiment-roadmap"],
      "representativeQueries": [
        "compare recorded vLLM and SGLang inference measurements",
        "explain TTFT TPOT throughput and decode percentile metrics",
        "show the experiment contract and limitations for an inference benchmark"
      ],
      "version": "1.0.0",
      "updatedAt": "2026-10-11T00:00:00Z"
    }
  ]
}