{
  "schema_version": 2,
  "model": "poolside/Laguna-S-2.1",
  "quantization": "INT4",
  "gpu_count": 4,
  "hardware": "Intel Arc Pro B70",
  "mode": "DFlash depth 7, single active generation",
  "lane": "exact M8 post-W2 gather-finalize",
  "status": "stage0_pass_source_and_cpu_binary_frozen",
  "selector": "VLLM_XPU_LAGUNA_M8_GATHER_FINALIZE",
  "source": {
    "vllm_start": "503f7784cf9d1704109b1e4650427fb4f417d604",
    "vllm_frozen": "5519c08c168838b7e0a418499603b907f127cbf9",
    "xpu_kernels_start": "c59aaadbbfd350c2b5f4ad663e247c2811ae3181",
    "xpu_kernels_production": "2020d1921de1af35356fce85a8a2f7703215612c",
    "xpu_kernels_frozen": "4772f727590c51b72add79350b913d098cf67872"
  },
  "approved_record": {
    "localmaxxing_id": "cmrx6p5dv001bo4017hb7sixz",
    "tokens_per_second": 33.89498511171744,
    "vllm": "8936aac144929190c1e53f8b8624ca397ce16f5b",
    "xpu_kernels": "b6076ce1249ffee0e30bee528f4cd15c3bffb234"
  },
  "structural_target": {
    "control_launches_per_47_layer_cycle": 94,
    "candidate_launches_per_47_layer_cycle": 47,
    "route_parallel_w2_changed": false,
    "minimum_per_card_median_saving_ms_per_cycle": 0.15
  },
  "exactness_contract": {
    "tokens": 8,
    "topk": 10,
    "hidden_size": 3072,
    "global_experts": 256,
    "local_experts": 64,
    "canonical_row_expression": "token * 10 + slot",
    "public_route_map": false,
    "slot_order": "0..9",
    "boundaries": [
      "BF16(fixed-order FP32 gather)",
      "BF16(float(routed_bf16) * 2.5)",
      "BF16(float(shared_bf16) + float(scaled_bf16))"
    ],
    "generic_m1_through_m8_incumbent": true,
    "explicit_combined_output_signal": true,
    "shared_output_consumptions": 1
  },
  "diagnostic_contract": {
    "schema": "_moe_C::laguna_m8_moe_gather_finalize_diagnostic",
    "shared_arithmetic_helper": true,
    "production_schema_unchanged": true,
    "production_debug_stores_compiled_out": true,
    "outputs": [
      "routed_bf16",
      "scaled_bf16",
      "final_bf16"
    ]
  },
  "cpu_build": {
    "compiler": "/opt/intel/oneapi/compiler/2025.3/bin/icpx",
    "artifact_root": "/mnt/fast-ai/llm-optimization-artifacts/laguna-s-2.1/binaries/gather-finalize-4772f72-20260724T075721Z",
    "candidate_moe_c_sha256": "6a6794249421aceb51f14980a3e2c0b0a9d7b492abf2f8d25b129b86f099bc5b",
    "incumbent_moe_c_sha256": "0057b266d567731a9f9f592cefd9103bbf027ebb83c876d26c17ffb09994a3a0",
    "build_log_sha256": "6428c5afca768d117fca382d890c792ae4c51d111e567307b7ce4473951a2130",
    "installed_archive_bytes_identical": true,
    "candidate_imported": false,
    "xpu_enumerated_or_allocated": false,
    "candidate_kernel_spill_warnings": 0,
    "diagnostic_kernel_spill_warnings": 0
  },
  "validation": {
    "xpu_kernel_static_and_host_oracle": {
      "passed": 16,
      "failed": 0
    },
    "vllm_focused": {
      "passed": 22,
      "failed": 0
    },
    "vllm_expanded": {
      "passed": 52,
      "failed": 0,
      "device_tests_skipped": 1
    },
    "ruff": "pass",
    "python_ast": "pass",
    "clang_format_oneapi_2025_3": "pass",
    "git_diff_check": "pass",
    "independent_read_only_audits": 5
  },
  "actions_not_performed": [
    "candidate native import",
    "XPU device discovery or allocation",
    "XPU primitive",
    "model load",
    "service start",
    "generation",
    "benchmark",
    "network access",
    "payload creation",
    "LocalMaxxing submission",
    "reboot"
  ],
  "next_authority": "separate frozen four-card component packet"
}