{
  "scope": "Phase 4 native-auto and allocated-draft actual ASGI service",
  "accepted": true,
  "checks": {
    "health_explicit": true,
    "sampling_422": true,
    "profile_422": true,
    "compare_422": true,
    "api_0_200": true,
    "api_0_direct_equal": true,
    "api_0_native_equal": true,
    "api_0_bound": true,
    "api_0_fullbudget_label": true,
    "api_1_200": true,
    "api_1_direct_equal": true,
    "api_1_native_equal": true,
    "api_1_bound": true,
    "api_1_fullbudget_label": true,
    "api_2_200": true,
    "api_2_direct_equal": true,
    "api_2_native_equal": true,
    "api_2_bound": true,
    "api_2_fullbudget_label": true,
    "sse_text_done": true,
    "concurrent_0_direct_equal": true,
    "concurrent_0_privacy": true,
    "concurrent_1_direct_equal": true,
    "concurrent_1_privacy": true,
    "cancel_prefill_isolated": true,
    "cancel_between_tokens": true,
    "recovery": true,
    "decode_new_arrival": true,
    "decode_new_arrival_native_0": true,
    "decode_new_arrival_native_1": true,
    "actual_one_token_limit": true,
    "actual_controlled_eos": true,
    "closed": true,
    "auto_health": true,
    "auto_http_0": true,
    "auto_native_strategy_0": true,
    "auto_native_output_0": true,
    "auto_http_1": true,
    "auto_native_strategy_1": true,
    "auto_native_output_1": true,
    "auto_http_2": true,
    "auto_native_strategy_2": true,
    "auto_native_output_2": true,
    "auto_closed": true,
    "source_unchanged": true,
    "no_swap_growth": true
  },
  "passed": 46,
  "total": 46,
  "failures": [],
  "source_sha256": {
    "benchmarks/experiments/phase4_verify_service.py": "507ea44f508a7d36030157f970c2f118afad87a8c90737fa8ce58e1a063c42ad",
    "examples/models/research-smollm2-pair.local.json": "e1757d5395dc5944b9ed32a76ccc48d47ac4d36c87a81901a97153d2715a5f04",
    "src/deepinfer_mac/__init__.py": "c9c2b1560a4d015c1531ad2ab1baf5c4ea633e1960b0c157733711baf138cb50",
    "src/deepinfer_mac/adapters.py": "67f056a36c7efe2282b1872cd3842ac5f4b8bb440d880c7752a8efc4116d4083",
    "src/deepinfer_mac/benchmark/__init__.py": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
    "src/deepinfer_mac/benchmark/export.py": "03f8e75757e3cf3a8a320f987b834c2828a4b8a948035911d0431f97a54e0550",
    "src/deepinfer_mac/benchmark/metrics.py": "348859cfd7834818daf754f28c86fdaef96f2896a866bba14a412aa967af36e0",
    "src/deepinfer_mac/benchmark/runner.py": "ccd76d171d63dba9f6bcc027a830fe047dfba4bc279d3619ace5f2ada8e0aaf4",
    "src/deepinfer_mac/cache.py": "3088bf3c757f400f1706dda66df436f2d36c4a3deac0f3483fa38933d6a72068",
    "src/deepinfer_mac/cli.py": "9977d64fe698997a128390f649b9bbad281b9bc10670401b8352f630329672f3",
    "src/deepinfer_mac/config.py": "6ad85acf30684c928b43505ca7f207e309c2515bd832df8fa18fa81d1f3d2786",
    "src/deepinfer_mac/continuous.py": "aba78090bd7f8860c0b0370d94f64efaba1330bb1969ccfabca45db509db5d49",
    "src/deepinfer_mac/cycle_cost.py": "f87ee249ac1c9ed359261a6f61601ac2a6784359ddf8047377d66a44d46cb9cc",
    "src/deepinfer_mac/doctor.py": "ede6b069f4fb43e50cd4a99038bafd8166f5a35bb8e6f0dfd8f757974ae284ec",
    "src/deepinfer_mac/ds/__init__.py": "c77f66fbd9fafbc80fb0958cf53c534bf205fdb2b70f8f7f582b92d3e488cba2",
    "src/deepinfer_mac/ds/joint_budget.py": "f3ed636aa87745ab2b99a5c00b57bb9d8f7804b78b612cc49a2dd06dc7e64613",
    "src/deepinfer_mac/ds/mtp_varlen.py": "de78b73a6e26d89cf6cfdc7eb472ba061e2ffd8a2842c6a9a060feb382448eae",
    "src/deepinfer_mac/ds/recovery_cost.py": "d62e4c6cb8d0a2afc3bff97f3a789e6c2c3bfdd4b273ff0db2d9ac52cfa31028",
    "src/deepinfer_mac/engine.py": "ed3c9a8b17165935243010b47f150e9ed5523af8e53e345daad7bc635bb74c11",
    "src/deepinfer_mac/hybrid_speculative.py": "6b51b39fc440739248e3fe7a623c2eb96a3b2ec194562dee3c8fb992c12c2e54",
    "src/deepinfer_mac/joint_service.py": "0e1dd1d11324bd036ab3ddb4287e20b3062082073ab113c3849049a3fa3c942f",
    "src/deepinfer_mac/models.py": "c80ec9da3d8350b4a17b6c14b97db6984c47f457c1a792f1b3db223d0b38a70e",
    "src/deepinfer_mac/native_handoff.py": "1ff456a016f063ea6ca4e62d24232deee14128ecdbc9a72479ad5039d98629c9",
    "src/deepinfer_mac/policy.py": "cee3abdd1d5b9f29ce90b30106c4bd7b0ba6c62c3831482edef7ebb9178a5268",
    "src/deepinfer_mac/provenance.py": "57eb6d5de88fb84e21e98cc15dbda6b40ce06f42dd3503a7232aa513ebce9d2f",
    "src/deepinfer_mac/sampling.py": "e8c86b6d3d72f6b5121e23eee54f975044e23d10cc4d88648755cc9867b5d8d1",
    "src/deepinfer_mac/server/__init__.py": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
    "src/deepinfer_mac/server/app.py": "94721393389c515072b54c6ea5c26e120b50926c2e07894abedc4cbb358f0f3c",
    "src/deepinfer_mac/server/continuous_worker.py": "a9da58d4811f6aa78b1de36b618cca5a3e09875d5c088bf7aaeacba43f69376a",
    "src/deepinfer_mac/server/joint_worker.py": "a1a56e2fb5d3232336840b358d33e32f0f4b307f535a5781f991a7a7b3c1d7f0",
    "src/deepinfer_mac/server/runtime_worker.py": "d3e9770439fdea79c6d288b0ad0cd4bffec40d0f5deb5b5bf7ad2745b00282ba",
    "src/deepinfer_mac/server/schemas.py": "87080294b40b3733f5e43493a1bdd25bb9dc8215b1521c25003fc268535ee256",
    "src/deepinfer_mac/server/worker.py": "7ce9ba42ae69ba45f298c9f1bb6aef2ddaf4138b5d3bb0397a6575c1c5505d90",
    "src/deepinfer_mac/session_boundaries.py": "ae8853b7f69e093b92ce2deda71039bcb3e73e1df3cb6e575dd01fde8963b98c",
    "src/deepinfer_mac/speculative.py": "d9b24f6e3215bfb6e414ce7378e7dd0518ac9a1ad4ca00897e69998fb8a5e41e",
    "src/deepinfer_mac/state.py": "d77e5d97484bd3dcdae324d6ee31830480bb7baafb1907bc10da7200fd17b096",
    "src/deepinfer_mac/varlen_verify.py": "b60467a8864807bc50a343ae62237c41c937ec9e2180daafa0d0d5b9f456414e"
  },
  "raw_report_sha256": "f013f1bc30fd4f5c3dc7bc4677c790c4fdd96a89c0120e82e7b6d0e280de11b3",
  "versions": {
    "mlx": "0.32.4",
    "mlx-lm": "0.32.2"
  },
  "metal_peak_bytes": 674364328,
  "no_swap_growth": true,
  "earlier_harness_failures": [
    "First probe assumed token_ids in public aggregate metadata; corrected to worker result.",
    "Second probe assumed SSE X-Run-ID header on nonstream responses; corrected to JSON deepinfer.run_id."
  ],
  "quality_scope": "Three short functional prompts; no broad pretrained or numerical equivalence claim."
}