[
  {
    "source": "original_nsys_worker_flush_v2",
    "kernel_count": 1091749,
    "counts": {
      "unique_runtime_match": 1091749
    },
    "matched_kernel_counts_by_api": {
      "cudaLaunchKernel_v7000": 371513,
      "cuLaunchKernelEx": 19776,
      "cudaLaunchKernelExC_v11060": 18184,
      "cuLaunchKernel": 16996,
      "cudaGraphLaunch_v10000": 665280
    },
    "links": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908/analysis/nsys_worker_flush_v2_physical/kernel_runtime_links.jsonl.gz",
    "key": "Full encoded process prefix (globalPid>>24 == globalTid>>24) plus correlationId; no cross-process correlation join",
    "scope": "All recorded kernels including warmup/initialization. One runtime API may launch many graph kernels. API duration is host call duration, not GPU time, queue wait, or critical-path latency. Missing runtime match may have driver/graph association outside this linker; not declared missing GPU execution."
  }
]