{
  "schema": "lingbot-evidence-v1",
  "checked_at": "2026-09-09T21:25:49.555642+00:00",
  "analysis_status": "Incomplete \u2014 evidence collection and validation gaps remain",
  "counter_restriction": "1",
  "baseline": [
    {
      "run": "original_r0",
      "steady_fps": 14.425322709972287,
      "rollout_fps": 12.224036573638378,
      "n": 80,
      "updated_at": "2026-09-08T09:35:21.832258+00:00"
    },
    {
      "run": "original_r1",
      "steady_fps": 14.003105677118704,
      "rollout_fps": 11.918680828242685,
      "n": 80,
      "updated_at": "2026-09-08T09:58:35.073708+00:00"
    },
    {
      "run": "original_r2",
      "steady_fps": 14.391931904541057,
      "rollout_fps": 12.200480308730132,
      "n": 80,
      "updated_at": "2026-09-08T10:06:47.572414+00:00"
    },
    {
      "run": "rebased_r0",
      "steady_fps": 14.404761015425771,
      "rollout_fps": 12.2582831962986,
      "n": 80,
      "updated_at": "2026-09-08T09:43:10.067034+00:00"
    },
    {
      "run": "rebased_r1",
      "steady_fps": 14.430399912803567,
      "rollout_fps": 12.282970681392925,
      "n": 80,
      "updated_at": "2026-09-08T09:50:56.686765+00:00"
    },
    {
      "run": "rebased_r2",
      "steady_fps": 14.351260677260294,
      "rollout_fps": 12.209826846345216,
      "n": 80,
      "updated_at": "2026-09-08T10:15:45.598630+00:00"
    }
  ],
  "profiles": [
    {
      "source": "original",
      "artifact_exists": true,
      "updated_at": "2026-09-08T10:25:06.160323+00:00"
    },
    {
      "source": "rebased",
      "artifact_exists": true,
      "updated_at": "2026-09-08T10:34:01.556861+00:00"
    }
  ],
  "captures": [
    {
      "config": "480x832_tp1_sp1",
      "workload_status": "completed",
      "tensor_files": 0,
      "metadata_logs": 0,
      "capture_status": "missing",
      "updated_at": "2026-09-08T10:53:20.526509+00:00"
    },
    {
      "config": "480x832_tp1_sp2",
      "workload_status": "completed",
      "tensor_files": 0,
      "metadata_logs": 0,
      "capture_status": "missing",
      "updated_at": "2026-09-08T10:45:34.207990+00:00"
    },
    {
      "config": "480x832_tp1_sp4",
      "workload_status": "completed",
      "tensor_files": 0,
      "metadata_logs": 0,
      "capture_status": "missing",
      "updated_at": "2026-09-08T10:37:41.541628+00:00"
    },
    {
      "config": "480x832_tp2_sp1",
      "workload_status": "completed",
      "tensor_files": 0,
      "metadata_logs": 0,
      "capture_status": "missing",
      "updated_at": "2026-09-08T11:03:39.662542+00:00"
    },
    {
      "config": "480x832_tp2_sp2",
      "workload_status": "completed",
      "tensor_files": 0,
      "metadata_logs": 0,
      "capture_status": "missing",
      "updated_at": "2026-09-08T11:19:58.650405+00:00"
    },
    {
      "config": "480x832_tp4_sp1",
      "workload_status": "completed",
      "tensor_files": 0,
      "metadata_logs": 0,
      "capture_status": "missing",
      "updated_at": "2026-09-08T11:12:01.059058+00:00"
    },
    {
      "config": "704x1280_tp1_sp1",
      "workload_status": "failed",
      "tensor_files": 0,
      "metadata_logs": 0,
      "capture_status": "missing",
      "updated_at": "2026-09-08T10:59:55.729498+00:00"
    },
    {
      "config": "704x1280_tp1_sp2",
      "workload_status": "completed",
      "tensor_files": 0,
      "metadata_logs": 0,
      "capture_status": "missing",
      "updated_at": "2026-09-08T10:50:18.482037+00:00"
    },
    {
      "config": "704x1280_tp1_sp4",
      "workload_status": "completed",
      "tensor_files": 0,
      "metadata_logs": 0,
      "capture_status": "missing",
      "updated_at": "2026-09-08T10:42:02.610336+00:00"
    },
    {
      "config": "704x1280_tp2_sp1",
      "workload_status": "completed",
      "tensor_files": 0,
      "metadata_logs": 0,
      "capture_status": "missing",
      "updated_at": "2026-09-08T11:08:29.481148+00:00"
    },
    {
      "config": "704x1280_tp2_sp2",
      "workload_status": "completed",
      "tensor_files": 0,
      "metadata_logs": 0,
      "capture_status": "missing",
      "updated_at": "2026-09-08T11:24:18.584862+00:00"
    },
    {
      "config": "704x1280_tp4_sp1",
      "workload_status": "completed",
      "tensor_files": 0,
      "metadata_logs": 0,
      "capture_status": "missing",
      "updated_at": "2026-09-08T11:16:26.092907+00:00"
    },
    {
      "config": "cross_480x832_tp1_sp4_v1",
      "workload_status": "completed",
      "tensor_files": 48,
      "metadata_logs": 5,
      "capture_status": "recorded_needs_validation",
      "updated_at": "2026-09-09T09:24:50.045962+00:00"
    },
    {
      "config": "dit_only_704x1280_tp1_sp1_v1",
      "workload_status": "completed",
      "tensor_files": 3,
      "metadata_logs": 2,
      "capture_status": "recorded_needs_validation",
      "updated_at": "2026-09-09T13:23:03.432256+00:00"
    },
    {
      "config": "dit_only_704x1280_tp1_sp1_v2",
      "workload_status": "completed",
      "tensor_files": 12,
      "metadata_logs": 2,
      "capture_status": "recorded_needs_validation",
      "updated_at": "2026-09-09T13:42:07.724372+00:00"
    },
    {
      "config": "evening_480x832_tp1_sp1_v1",
      "workload_status": "completed",
      "tensor_files": 12,
      "metadata_logs": 2,
      "capture_status": "recorded_needs_validation",
      "updated_at": "2026-09-09T11:59:40.313602+00:00"
    },
    {
      "config": "evening_480x832_tp2_sp1_v1",
      "workload_status": "completed",
      "tensor_files": 24,
      "metadata_logs": 3,
      "capture_status": "recorded_needs_validation",
      "updated_at": "2026-09-09T10:56:10.105150+00:00"
    },
    {
      "config": "evening_480x832_tp2_sp2_v1",
      "workload_status": "completed",
      "tensor_files": 48,
      "metadata_logs": 5,
      "capture_status": "recorded_needs_validation",
      "updated_at": "2026-09-09T10:30:32.910078+00:00"
    },
    {
      "config": "evening_480x832_tp4_sp1_v1",
      "workload_status": "completed",
      "tensor_files": 48,
      "metadata_logs": 5,
      "capture_status": "recorded_needs_validation",
      "updated_at": "2026-09-09T11:35:02.264586+00:00"
    },
    {
      "config": "evening_704x1280_tp1_sp1_v1",
      "workload_status": "failed",
      "tensor_files": 4,
      "metadata_logs": 2,
      "capture_status": "recorded_needs_validation",
      "updated_at": "2026-09-09T12:17:03.628714+00:00"
    },
    {
      "config": "evening_704x1280_tp1_sp2_v1",
      "workload_status": "completed",
      "tensor_files": 24,
      "metadata_logs": 3,
      "capture_status": "recorded_needs_validation",
      "updated_at": "2026-09-09T10:02:17.453471+00:00"
    },
    {
      "config": "evening_704x1280_tp2_sp1_v1",
      "workload_status": "completed",
      "tensor_files": 24,
      "metadata_logs": 3,
      "capture_status": "recorded_needs_validation",
      "updated_at": "2026-09-09T11:06:48.105316+00:00"
    },
    {
      "config": "evening_704x1280_tp2_sp2_v1",
      "workload_status": "completed",
      "tensor_files": 48,
      "metadata_logs": 5,
      "capture_status": "recorded_needs_validation",
      "updated_at": "2026-09-09T10:38:54.463818+00:00"
    },
    {
      "config": "evening_704x1280_tp4_sp1_v1",
      "workload_status": "completed",
      "tensor_files": 48,
      "metadata_logs": 5,
      "capture_status": "recorded_needs_validation",
      "updated_at": "2026-09-09T11:43:13.385717+00:00"
    },
    {
      "config": "fixed2_480x832_tp1_sp4",
      "workload_status": "completed",
      "tensor_files": 24,
      "metadata_logs": 5,
      "capture_status": "recorded_needs_validation",
      "updated_at": "2026-09-09T08:44:50.231164+00:00"
    },
    {
      "config": "fixed_480x832_tp1_sp4",
      "workload_status": "failed",
      "tensor_files": 0,
      "metadata_logs": 5,
      "capture_status": "missing",
      "updated_at": "2026-09-09T08:39:44.160861+00:00"
    },
    {
      "config": "retry_480x832_tp1_sp2_v1",
      "workload_status": "completed",
      "tensor_files": 24,
      "metadata_logs": 3,
      "capture_status": "recorded_needs_validation",
      "updated_at": "2026-09-09T12:56:54.409136+00:00"
    },
    {
      "config": "retry_704x1280_tp1_sp4_v1",
      "workload_status": "completed",
      "tensor_files": 48,
      "metadata_logs": 5,
      "capture_status": "recorded_needs_validation",
      "updated_at": "2026-09-09T13:07:11.351679+00:00"
    }
  ],
  "items": [
    {
      "id": "baseline",
      "title": "Clean baseline timing",
      "status": "partial",
      "recorded": "6/6 runs; raw chunk/session times, FPS, TTFC and saved outputs.",
      "missing": "80 steady samples per run; final requirement is at least 1,000 per run.",
      "impact": "Mean throughput comparisons are possible. Final P95/P99 or a service SLA is not established.",
      "next": "Expand independent samples after accepting the configuration.",
      "evidence": "matched_baselines/*/out/summary.json"
    },
    {
      "id": "timeline",
      "title": "CUDA activity timeline",
      "status": "partial",
      "recorded": "Worker\u663e\u5f0f\u5199\u51fa\u8865\u91c7\u5df2\u5b8c\u6210\uff1a\u6700\u540e\u57574\u4e2aDiT\u8fdb\u7a0b\u54048692\u6b21kernel\u3001\u5176\u4e2d400\u6b21\u5177\u540dAttention\uff1b\u65f6\u949f\u951a\u70b9\u543b\u5408\u3002",
      "missing": "\u65e7\u65f6\u95f4\u7ebf\u5c3e\u90e8\u7f3a\u5931\u4ecd\u4fdd\u7559\uff1b\u65b0\u8bb0\u5f55\u5168\u90e8\u5c42/shape\u5173\u8054\u53ca\u5173\u952e\u8def\u5f84\u5f52\u56e0\u5c1a\u672a\u5b8c\u6210\u3002",
      "impact": "\u65b0\u8865\u91c7\u901a\u8fc7\u6700\u540e\u5757\u8986\u76d6\u68c0\u67e5\uff0c\u4f46\u4e0d\u8bc1\u660e\u5168\u8c03\u7528\u5b8c\u6574\uff0c\u4e5f\u4e0d\u8ffd\u6eaf\u4fee\u590d\u65e7\u7edf\u8ba1\u3002\u8bca\u65ad\u63d2\u6869\u8017\u65f6\u4e0d\u80fd\u66ff\u4ee3\u5e72\u51c0\u57fa\u7ebf\u3002",
      "next": "\u6309\u7269\u7406GPU\u5206\u6790\u65b0\u8c03\u7528\u660e\u7ec6\uff0c\u6838\u5bf9\u5404\u5757\u53cashape\u8986\u76d6\u3002",
      "evidence": "analysis/original_nsys_worker_flush_v2_anchor_audit.json; analysis/nsys_worker_flush_v2_physical"
    },
    {
      "id": "counters",
      "title": "GPU hardware counters",
      "status": "partial",
      "recorded": "Root NCU captured real saved LingBot FA3 QKV replay: preparation and attention kernels, 11 replay passes each; raw report and metrics exported.",
      "missing": "Additional NCU collection paused: guard kills root SSH workloads under normal-user scheduling, including --user root.",
      "impact": "Earlier FA3 raw counters remain valid. Further root profiling needs an approved scheduler identity; existing timeline/replay work continues.",
      "next": "Clarify exception to normal-user scheduling and test root-owned gpu run; do not modify guard or driver settings.",
      "evidence": "raw/ncu_attention_fa3_pilot_v1/attention.ncu-rep; metrics.csv; selected_metrics.json"
    },
    {
      "id": "cpu",
      "title": "CPU sampling & scheduling",
      "status": "not_recorded",
      "recorded": "OS runtime/API trace requested.",
      "missing": "CPU statistical samples and scheduler context-switch trace explicitly disabled.",
      "impact": "API waits can be inspected, but CPU hotspot attribution and scheduler/preemption causes are limited.",
      "next": "Enable appropriate host perf permissions if these investigations are needed.",
      "evidence": "--sample=none --cpuctxsw=none"
    },
    {
      "id": "qkv",
      "title": "Exact attention shapes & tensors",
      "status": "partial",
      "recorded": "12\u79cd\u914d\u7f6e\u7684\u4ee3\u8868\u8f93\u5165\u5df2\u5b8c\u6210\u91c7\u96c6\u5ba1\u6838\uff1b1008/1008\u4e2a\u56de\u653e\u70b9\u5df2\u5c1d\u8bd5\uff0c1004\u70b9\u5ba1\u6838\u901a\u8fc7\uff0c4\u70b9\u6267\u884c\u5931\u8d25\u3002",
      "missing": "704\u00d71280 TP1/SP1\u76844\u4e2aSDPA math\u7ec4\u5408\u663e\u5b58\u4e0d\u8db3\uff1b\u4ee3\u8868\u8f93\u5165\u4e0d\u7b49\u4e8e\u5168\u90e8\u5c42\u5b8c\u6574\u5f20\u91cf\u3002",
      "impact": "\u6709\u6548\u70b9\u53ef\u8fdb\u884c\u540c\u8f93\u5165\u8017\u65f6\u3001\u8bef\u5dee\u548c\u5b9e\u9645kernel\u6bd4\u8f83\uff1bOOM\u70b9\u65e0\u6709\u6548\u8ba1\u65f6\u3002\u72ec\u7acb\u56de\u653e\u4e0d\u8bc1\u660e\u6574\u6a21\u578b\u7b49\u4ef7\u3002",
      "next": "\u4fdd\u7559\u5931\u8d25\u8bb0\u5f55\uff0c\u5c06\u6709\u6548\u7ed3\u679c\u7528\u4e8e\u5019\u9009\u5206\u6790\uff1b\u6574\u6a21\u578b\u63a5\u53d7\u95e8\u69db\u5355\u72ec\u9a8c\u8bc1\u3002",
      "evidence": "analysis/matrix_coverage.json; analysis/attention_execution_failures.json; analysis/attention_evening_*_audit.json"
    },
    {
      "id": "correctness",
      "title": "Numerical correctness",
      "status": "partial",
      "recorded": "Decoder validation and saved ten-chunk output comparisons. Same-source launch variation observed.",
      "missing": "Full DiT equivalence, multi-seed quality, long rollouts and action-response acceptance.",
      "impact": "Cannot label rebased outputs equivalent or promote an optimized backend based on speed alone.",
      "next": "Complete same-/cross-source comparisons and predeclare acceptance criteria.",
      "evidence": "analysis/*_vs_*.json; out/decode_validation.json"
    },
    {
      "id": "inventory",
      "title": "Every-kernel inventory",
      "status": "partial",
      "recorded": "Historical windows: 56,458 events, 1,422 provisional groups; 99.30% host-op linkage.",
      "missing": "Historical exact kernel-input coverage is 0%; current full-timeline grouping unfinished.",
      "impact": "Historical host-op shapes cannot be treated as exact inputs of fused kernels or current-source attribution.",
      "next": "Build a current invocation ledger with shape/context evidence and unresolved groups.",
      "evidence": "raw/historical_inventory_D/coverage.json"
    },
    {
      "id": "mfu",
      "title": "FLOPs, MFU & calibration",
      "status": "partial",
      "recorded": "\u5355\u5361\u5927GEMM\u4e09\u6b21\u72ec\u7acb\u542f\u52a8\u517127\u9879\u901a\u8fc7\u5ba1\u6838\uff1b\u5206\u522b\u8bb0\u5f55BF16\u3001FP32\u7981\u7528TF32\u548c\u5141\u8bb8TF32\u3002Attention\u4ee3\u8868\u8f93\u5165\u5df2\u8ba1\u7b97\u6709\u6548\u7b97\u6cd5FLOPs\u3002",
      "missing": "\u5168\u90e8\u6a21\u578bFLOPs\u53ca\u9010shape\u4e0a\u9650\u3001DRAM/L2\u6d41\u91cf\u548c\u5e26\u5bbd\u6821\u51c6\u672a\u5b8c\u6210\u3002",
      "impact": "\u5355\u5361\u5927\u77e9\u9635\u63d0\u4f9b\u5c40\u90e8\u8ba1\u7b97\u53c2\u8003\uff0c\u4e0d\u80fd\u5f53\u4f5c\u901a\u7528\u5cf0\u503c\u3001\u5168\u6a21\u578bMFU\u6216\u5b8c\u6574roofline\u3002",
      "next": "\u8865\u9f50\u786c\u4ef6\u6d41\u91cf\u548c\u5168\u6a21\u578bshape\u5f52\u56e0\u540e\u518d\u8ba1\u7b97\u5bf9\u5e94\u5229\u7528\u7387\u3002",
      "evidence": "analysis/local_gemm_calibration_audit.json; analysis/graph_points.jsonl"
    },
    {
      "id": "service",
      "title": "Encoding & client delivery",
      "status": "not_recorded",
      "recorded": "Current endpoint: CPU-ready uint8 in decoder rank 0.",
      "missing": "Downstream RGB IPC, encoding, network, playable-client delivery and backpressure tests.",
      "impact": "14 FPS CPU-ready does not prove 14 FPS at the client.",
      "next": "Add separately measured fixed-codec delivery and bounded-queue tests.",
      "evidence": "Case D measurement contract"
    },
    {
      "id": "archive",
      "title": "Raw data & reproduction",
      "status": "recorded",
      "recorded": "Source archives, patches, input copies, checkpoint hashes, timing logs, tensors and Nsight artifacts retained locally.",
      "missing": "Full artifact checksum catalog still pending; large raw files are not mirrored to this web host.",
      "impact": "Existing data can be reanalysed. Public dashboard is a summary; original evidence stays on the GPU workspace.",
      "next": "Complete checksums and preserve each new run without overwriting.",
      "evidence": "runs/analysis_report_20260908/raw/provenance/"
    },
    {
      "id": "optimization",
      "title": "Optimization conclusions",
      "status": "partial",
      "recorded": "1004\u4e2a\u72ec\u7acbAttention\u70b9\u5ba1\u6838\u901a\u8fc7\uff1bcuDNN cross\u6574\u6a21\u578b\u8bca\u65ad\u5df2\u4fdd\u5b58\u5b9e\u9645dispatch\u4e0e40\u4e2a\u8f93\u51fa\u6bd4\u8f83\u3002",
      "missing": "\u6574\u6a21\u578b\u8f93\u51fa\u6bd4\u8f83\u672a\u901a\u8fc7\u9884\u8bbe\u95e8\u69db\uff1b\u540c\u7248\u672c\u8de8\u542f\u52a8\u5dee\u5f02\u5c1a\u672a\u5b9a\u4f4d\u3002H1\u2013H5\u53d7\u63a7\u6d88\u878d\u672a\u5b8c\u6210\u3002",
      "impact": "\u4e0d\u80fd\u8ba4\u9886\u6574\u6a21\u578b\u52a0\u901f\u6216\u6570\u503c\u7b49\u4ef7\uff1b\u5df2\u6709QKV\u878d\u5408\u548cuint8\u8f6c\u6362\u4e0d\u80fd\u91cd\u590d\u8ba1\u4e3a\u65b0\u4f18\u5316\u3002",
      "next": "\u9694\u79bb\u8de8\u542f\u52a8\u6570\u503c\u5dee\u5f02\uff0c\u4fdd\u6301\u9884\u8bbe\u95e8\u69db\uff0c\u518d\u5f00\u5c55\u53ef\u63a5\u53d7\u5019\u9009\u7684\u6027\u80fd\u5bf9\u7167\u3002",
      "evidence": "analysis/matrix_coverage.json; analysis/integrated_cross_diagnostic_v1_output_comparison.json"
    },
    {
      "id": "expanded_baseline",
      "title": "\u5927\u6837\u672c\u57fa\u7ebf\u5ba1\u6838",
      "status": "partial",
      "recorded": "\u5df2\u901a\u8fc7\u5ba1\u6838\u7684\u72ec\u7acb\u8fd0\u884c\uff1a5\uff1b\u539f\u59cb\u8ba1\u65f6\u4e0esession\u5206\u7ec4bootstrap\u6309\u8fd0\u884c\u4fdd\u5b58\u3002",
      "missing": "\u6b63\u5728\u8fd0\u884c\u548c\u672a\u5f00\u59cb\u7684\u91cd\u590d\u4e0d\u8ba1\u4e3a\u5b8c\u6210\uff1b\u591a\u6b21\u72ec\u7acb\u8fd0\u884c\u7a33\u5b9a\u6027\u4ecd\u9700\u5ba1\u6838\u3002",
      "impact": "1000\u4e2a\u8fdb\u7a0b\u5185\u7a33\u6001\u95f4\u9694\u4e0d\u7b49\u4e8e1000\u6b21\u72ec\u7acb\u542f\u52a8\uff0cCPU-ready\u5c3e\u5ef6\u8fdf\u4e0d\u7b49\u4e8e\u5ba2\u6237\u7aefSLA\u3002",
      "next": "\u6bcf\u6b21\u6210\u529f\u9000\u51fa\u540e\u68c0\u67e5250 session\u548c1000\u4e2a\u7a33\u6001\u95f4\u9694\uff0c\u4fdd\u7559\u5168\u90e8\u505c\u987f\u3002",
      "evidence": "raw/tail_timing_audit_workflow_v1.json; analysis/*_timing/metrics.json"
    }
  ],
  "kernel_counts": {
    "CUPTI_ACTIVITY_KIND_KERNEL": 954123,
    "CUPTI_ACTIVITY_KIND_MEMCPY": 194059,
    "CUPTI_ACTIVITY_KIND_RUNTIME": 1875583,
    "kernel_device_counts": [
      [
        0,
        415515
      ],
      [
        1,
        179592
      ],
      [
        2,
        179471
      ],
      [
        3,
        179545
      ]
    ]
  },
  "measurement_boundary": "CPU-ready uint8; excludes encoding and network",
  "raw_root": "/data/zhoutaichang/feature/Lingbot_world_realtime/runs/analysis_report_20260908",
  "refresh_seconds": 60
}
