{
  "date": "2026-10-28",
  "scope": "Direct in-cluster startup and warm-call confirmation only; excludes parallel throughput",
  "startup": [
    {
      "api": "sandbox",
      "rounds": 6,
      "allocation_s": {
        "median": 0.0011169277131557465,
        "range": [
          0.0009907316416501999,
          0.0012924093753099442
        ],
        "values_by_round": [
          0.0009907316416501999,
          0.0012917201966047287,
          0.001018112525343895,
          0.0011560264974832535,
          0.0010778289288282394,
          0.0012924093753099442
        ]
      },
      "first_call_s": {
        "median": 0.08397808577865362,
        "range": [
          0.07761088944971561,
          0.43080203607678413
        ],
        "values_by_round": [
          0.43080203607678413,
          0.07761088944971561,
          0.09316668286919594,
          0.09023474343121052,
          0.07772142812609673,
          0.07770074531435966
        ]
      },
      "seconds": {
        "median": 0.08519196230918169,
        "range": [
          0.07879925705492496,
          0.43179276771843433
        ],
        "values_by_round": [
          0.43179276771843433,
          0.07890260964632034,
          0.09418479539453983,
          0.09139076992869377,
          0.07879925705492496,
          0.07899315468966961
        ]
      }
    },
    {
      "api": "session",
      "rounds": 6,
      "allocation_s": {
        "median": 0.0010958323255181313,
        "range": [
          0.0010738540440797806,
          0.0012312531471252441
        ],
        "values_by_round": [
          0.0012312531471252441,
          0.0010847579687833786,
          0.0010738540440797806,
          0.0010935720056295395,
          0.001098092645406723,
          0.0011977292597293854
        ]
      },
      "first_call_s": {
        "median": 0.09306954126805067,
        "range": [
          0.07034713216125965,
          0.19867286831140518
        ],
        "values_by_round": [
          0.19867286831140518,
          0.09401835501194,
          0.07959413900971413,
          0.07034713216125965,
          0.09482796303927898,
          0.09212072752416134
        ]
      },
      "seconds": {
        "median": 0.09421078488230705,
        "range": [
          0.07144070416688919,
          0.19990412145853043
        ],
        "values_by_round": [
          0.19990412145853043,
          0.09510311298072338,
          0.08066799305379391,
          0.07144070416688919,
          0.09592605568468571,
          0.09331845678389072
        ]
      }
    }
  ],
  "warm_descriptive": [
    {
      "api": "sandbox",
      "language": "python",
      "rounds": 6,
      "median_of_round_medians_s": 0.025164176244288683,
      "median_of_round_p95_s": 0.029147691931575535
    },
    {
      "api": "sandbox",
      "language": "shell",
      "rounds": 6,
      "median_of_round_medians_s": 0.008600275497883558,
      "median_of_round_p95_s": 0.010104039963334798
    },
    {
      "api": "session",
      "language": "python",
      "rounds": 6,
      "median_of_round_medians_s": 0.02363781351596117,
      "median_of_round_p95_s": 0.02654297649860382
    },
    {
      "api": "session",
      "language": "shell",
      "rounds": 6,
      "median_of_round_medians_s": 0.008674916345626116,
      "median_of_round_p95_s": 0.009744110936298966
    }
  ],
  "warm_per_round": [
    {
      "round": 0,
      "api": "sandbox",
      "language": "python",
      "median_s": 0.022817584685981274,
      "p95_s": 0.02603596551343798
    },
    {
      "round": 1,
      "api": "sandbox",
      "language": "python",
      "median_s": 0.025925577618181705,
      "p95_s": 0.028806668240576982
    },
    {
      "round": 2,
      "api": "sandbox",
      "language": "python",
      "median_s": 0.02440277487039566,
      "p95_s": 0.029636050947010516
    },
    {
      "round": 3,
      "api": "sandbox",
      "language": "python",
      "median_s": 0.02351696416735649,
      "p95_s": 0.026694945152848957
    },
    {
      "round": 4,
      "api": "sandbox",
      "language": "python",
      "median_s": 0.027692271396517754,
      "p95_s": 0.02948871562257409
    },
    {
      "round": 5,
      "api": "sandbox",
      "language": "python",
      "median_s": 0.027911637909710407,
      "p95_s": 0.030402598809450863
    },
    {
      "round": 0,
      "api": "sandbox",
      "language": "shell",
      "median_s": 0.007782183587551117,
      "p95_s": 0.009754778631031512
    },
    {
      "round": 1,
      "api": "sandbox",
      "language": "shell",
      "median_s": 0.008493142202496529,
      "p95_s": 0.010305747296661139
    },
    {
      "round": 2,
      "api": "sandbox",
      "language": "shell",
      "median_s": 0.008539792150259018,
      "p95_s": 0.010215799137949943
    },
    {
      "round": 3,
      "api": "sandbox",
      "language": "shell",
      "median_s": 0.00873959343880415,
      "p95_s": 0.010308524779975414
    },
    {
      "round": 4,
      "api": "sandbox",
      "language": "shell",
      "median_s": 0.008660758845508099,
      "p95_s": 0.009930831752717494
    },
    {
      "round": 5,
      "api": "sandbox",
      "language": "shell",
      "median_s": 0.008918916806578636,
      "p95_s": 0.009992280788719653
    },
    {
      "round": 0,
      "api": "session",
      "language": "python",
      "median_s": 0.02731963060796261,
      "p95_s": 0.029959340393543244
    },
    {
      "round": 1,
      "api": "session",
      "language": "python",
      "median_s": 0.02277345024049282,
      "p95_s": 0.025149164069443942
    },
    {
      "round": 2,
      "api": "session",
      "language": "python",
      "median_s": 0.02534108515828848,
      "p95_s": 0.027469959948211907
    },
    {
      "round": 3,
      "api": "session",
      "language": "python",
      "median_s": 0.02450217679142952,
      "p95_s": 0.02713785655796528
    },
    {
      "round": 4,
      "api": "session",
      "language": "python",
      "median_s": 0.022564788348972797,
      "p95_s": 0.02594809643924236
    },
    {
      "round": 5,
      "api": "session",
      "language": "python",
      "median_s": 0.022758844308555126,
      "p95_s": 0.025289582088589665
    },
    {
      "round": 0,
      "api": "session",
      "language": "shell",
      "median_s": 0.009131183847784996,
      "p95_s": 0.010166460741311311
    },
    {
      "round": 1,
      "api": "session",
      "language": "shell",
      "median_s": 0.007888581603765488,
      "p95_s": 0.0095782533288002
    },
    {
      "round": 2,
      "api": "session",
      "language": "shell",
      "median_s": 0.008715993724763393,
      "p95_s": 0.009829531610012054
    },
    {
      "round": 3,
      "api": "session",
      "language": "shell",
      "median_s": 0.008633838966488838,
      "p95_s": 0.009625609219074249
    },
    {
      "round": 4,
      "api": "session",
      "language": "shell",
      "median_s": 0.008230451494455338,
      "p95_s": 0.009658690262585878
    },
    {
      "round": 5,
      "api": "session",
      "language": "shell",
      "median_s": 0.008799654431641102,
      "p95_s": 0.010193163156509399
    }
  ],
  "protocol": {
    "rounds": 6,
    "startup_observations": 12,
    "warm_calls": 1200,
    "calls_per_api_language_per_round": 50,
    "excluded_warmups_per_group": 5,
    "between_round_copies": "Checkpoints saved inside client pod; all checkpoint collection occurs after measured rounds finish",
    "image_reference": "Exact Linux amd64 platform-manifest digest; image caches retained"
  },
  "runtime": {
    "sdk_version": "0.7.0",
    "client_image": "docker.io/xflops/flame-console@sha256:8d8257449f592dc571a530feb9f7e2591754fb8a446f87d3246d778ebdf7054e",
    "execution_image": "python:3.12-slim@sha256:1148dbb014615300fa7e713b2f94f550c0d1c69780c0805f84f093ceee34dd0b",
    "worker_manager_image": "docker.io/xflops/flame-executor-manager@sha256:11bd2dc67de4b229878197ae5caeb1ab365deca1e8db53eb21f7a977959168b6",
    "packages": {
      "flamepy": "0.7.0",
      "grpcio": "1.85.0",
      "protobuf": "7.36.2"
    },
    "client_python": "3.12.3 (main, Aug 31 2026, 10:18:26) [GCC 13.3.0]",
    "pr663_deployed": false
  },
  "validation": {
    "all_outputs_checked": true,
    "local_summary_matches_cluster": true,
    "existing_sessions_preserved": true,
    "owned_sessions_deleted": true,
    "owned_namespace_deleted": true,
    "production_unchanged_during_measurements": true
  },
  "provenance": {
    "source_manifest": {
      "agent_perf.py": "dfe2dd859e2bd87cc6a383179a8ff15a3383287a92f286179583997aa4e87031",
      "agentperf_stats.py": "16b80364eb27568d5d464165ed045e62329f872c93ac2a84ca7d003231562038",
      "inputs.json": "542d84012733e5c30658428a7e269d4449bb00124fa52c04b5779d318549e8d8",
      "sdk/flamepy/__init__.py": "8cb72375130bf888449b7c68f36bf736ddd28341a0bc9d2855177829aaf02256",
      "sdk/flamepy/__main__.py": "0b1e739b8268c0569d6829d0c32ae9ab4930a3dbf8b4fff4bc8926b29158e7b2",
      "sdk/flamepy/serving/instance.py": "9cd4a7d831ce6c6555af671283db8f2ae101d7ebd645193cbc5179b46b4dbda8",
      "sdk/flamepy/serving/client.py": "48b05dc80f95280d78d88cee8995df19d92ece8b3af86ec3232874e04feefe1d",
      "sdk/flamepy/serving/__init__.py": "5577831fc4a94e69b3148152ba3ae66b0b60f48c705005f32927fb0fdadd7288",
      "sdk/flamepy/core/service.py": "d059256f79789b4dba89e2fff8a367b3d510ebcb234564e4091e6ee1a0739556",
      "sdk/flamepy/core/client.py": "f348b8934cefa71c8f9ff4c52014518d5be987dc47c6a8fcec771400b1e3058a",
      "sdk/flamepy/core/cache.py": "f1f25b30ee3a8b994d13d7ffa41cb24645edc9f056fc2fa311c9ec503a8e36ba",
      "sdk/flamepy/core/__init__.py": "5ee1f051e0b2d70541cb7de3f1f3dde3f6561cedaaa1c77cc0076927ec5fe9cb",
      "sdk/flamepy/core/types.py": "baab3a17a38e5738a16a2926472719c308b713326902abf13d33958b0afdb8aa",
      "sdk/flamepy/core/_bridge.py": "6645cb33973a0730575f960c961769c54ede5fd430618cbfc482700cddc21c1d",
      "sdk/flamepy/app/client.py": "998d47eb6c469ae3ef084677f86a871a722a6ad9074ca3e9bca3ea790de2c4a2",
      "sdk/flamepy/app/__init__.py": "b16908914ffc670b03948d4d89b19eaac6b58746988b032af7ef31593901bd53",
      "sdk/flamepy/app/types.py": "b610eac602ef6391cbf8376b9de183bae0a6b6368b7fa7a38de0d44533534208",
      "sdk/flamepy/app/_context.py": "235d1741aa86d31ed758b5c5a48192480a71fb418196cad9118aa921cf09c65a",
      "sdk/flamepy/app/storage.py": "97d820e82412bda838a3de62f64f212c908b845ff9e4b85e45990c6b43d41510",
      "sdk/flamepy/app/helper.py": "d017110a2ac8366fc8fea228fb9cf6cb7ec8d0d09a8bb8cceadfa9a40f8adf7b",
      "sdk/flamepy/app/runpy.py": "1b23e7fd1bc4c4f3d10b8be1d3dd6e061451fd94114d8c58aff925b4f1a3610f",
      "sdk/flamepy/cache/__init__.py": "bfba8dbfe3f41d04a479f2669880f2695e85ac855a71c2e9f6ed50c7c73c84cb",
      "sdk/flamepy/proto/types_pb2_grpc.py": "8b29cbbf196883870eb2e1fb3c9670104c033135e477d326b3150cb2770b9aa5",
      "sdk/flamepy/proto/__init__.py": "0a2aa9557b6346e29f5ebde0209c1a8e834e5e12c326c406a206162c0b78bb1f",
      "sdk/flamepy/proto/cache_pb2_grpc.py": "f77a6ced1a815c4685c6d73005d969bd7e418ddb2670b86e9cf2a73f6c076110",
      "sdk/flamepy/proto/shim_pb2_grpc.py": "5e712a8cf7a5f375249e9d2b6e52b59bc535c606a1a5e4655bceacd45e7d5ee8",
      "sdk/flamepy/proto/frontend_pb2.py": "9d072af93c3e369a933fbd1ea06b7373d590706aa25d7309c29c9ca029bdf753",
      "sdk/flamepy/proto/shim_pb2.py": "a57b7d9b2d141234cd23063b24d6086120fa8314ce825912b4e97f8a5084bc81",
      "sdk/flamepy/proto/frontend_pb2_grpc.py": "c54a6d6e745e29ed2037522f1c996f869d54ddbe31011af66ca38d872e5afc58",
      "sdk/flamepy/proto/cache_pb2.py": "045ad70798185c0495db91774d932f874ebb53d29ea4ae7fc21ad0ec530200f5",
      "sdk/flamepy/proto/types_pb2.py": "6a355cf62e46205c35ea4bfe6ddf40c0a7df73b29abe08215a0893d50c7cddbb",
      "sdk/flamepy/util/__init__.py": "20fe8c14a053abfc8ce1e597ef8c1a36e21888c004ba8dad4dfc62a42eba13d4",
      "sdk/flamepy/agent/session.py": "91694a9bbe2aff74ca7f9dcf468a655d500f9d4dbd285de30b07a477d378fb40",
      "sdk/flamepy/agent/__init__.py": "f02869c06e3cbf1bf147dd2703a06d8e0ef6c5234d8e276d748fb43089892a3a",
      "sdk/flamepy/agent/sandbox.py": "a3c039c4ea3087047b149baf05c3608d046563c6b837569f7865e8ff2f6e7a80",
      "sdk/flamepy/core/aio/service.py": "5c7b1b75b922a97651df799e4184cfa21b561ee8df56121bf179e7c62cb51aa2",
      "sdk/flamepy/core/aio/client.py": "430556d1bed36134f84b960cd0ba86795bfffb9953f5af83451ecccc2016cb8c",
      "sdk/flamepy/core/aio/cache.py": "af90bb1e7021285ee016e64859382bf1f89a33d048cfcf9cd11e023cd22fbc3f",
      "sdk/flamepy/core/aio/__init__.py": "05cf81c723ab2a9cfbe4443215c66f6592abc3e9cd9b7d53e14fe3d9cb0bb825",
      "sdk/flamepy-0.7.0.dist-info/METADATA": "4f6c9c7d2711f700d873f6235fdd827aef292add152e16d81dff37ebd95477e8",
      "sdk/flamepy-0.7.0.dist-info/WHEEL": "98e75747d0b7c0c568bfb71263c3d8f970b3a374a6996087bc642be5cb09a1aa"
    },
    "frozen_input_sha256": {
      "agent_perf.py": "dfe2dd859e2bd87cc6a383179a8ff15a3383287a92f286179583997aa4e87031",
      "agentperf_stats.py": "16b80364eb27568d5d464165ed045e62329f872c93ac2a84ca7d003231562038",
      "inputs.json": "542d84012733e5c30658428a7e269d4449bb00124fa52c04b5779d318549e8d8",
      "sdk-0.7.0.tgz": "b4a3241e7d61f6310eaf6bc174a8e95297d07ccbd67eb86812bc9edd1cee8782"
    },
    "raw_evidence_sha256": "1715ebebb5b3d46d86fbddb2b8b683e745fdb5bdb099849799a74fa65497cb07",
    "raw_evidence_bytes": 295760
  },
  "worker_revision": "905a36049650af25cd86f07475bcdae091c3c98f",
  "measurement_definitions": {
    "startup": "Creation API call through first completed Python command; SDK already connected; image pinned to its Linux amd64 platform manifest with worker image caches retained.",
    "warm_median": "Median of six within-round medians, each with50calls after5excludedwarmups.",
    "typical_round_p95": "Median of six within-round p95 estimates, each based on50calls; not a pooled tail estimate.",
    "network": "Direct internal Kubernetes Services with proxy variables removed. In-cluster RPC remains included; Mac-to-cluster transport is outside measured intervals.",
    "executor_lifecycle": "New logical environments reuse the same two executors across rounds. First-use observations are432ms/200ms; later allocations on retained executors are71\u201395ms. Aggregate medians mix both conditions and are not fully cold executor-startup results."
  },
  "resources": {
    "worker_cpu_request": 1,
    "worker_memory_request_bytes": 536870912,
    "worker_gpu_request": 0,
    "client_cpu_request": 1,
    "client_memory_request_bytes": 1073741824,
    "client_cpu_limit": 2,
    "client_memory_limit_bytes": 2147483648
  },
  "limitations": [
    "Measures direct agent.Sandbox and agent.Session SDK calls, excluding Hermes orchestration and LLM inference.",
    "Startup uses retained image caches and an established SDK connection; uncached image downloads and initial client connection setup are not measured.",
    "Startup phases have not been isolated, so the result does not attribute time to scheduler, storage, RPC or sandbox runtime individually.",
    "Parallel throughput is not validated by this report. A separate eight-worker Session pool allocation timed out and its partial results are excluded.",
    "Six rounds on one deployment; observed ranges and typical round p95 estimates are descriptive, not production latency guarantees.",
    "The configured SDK benchmark pins the platform digest explicitly. It does not measure PR663 or establish the same startup latency for mutable image tags or the plugin default image reference.",
    "The85/94ms medians include retained-executor reuse; a fresh logical session does not imply a newly launched worker service."
  ]
}