{
  "title": "Mirai voice infrastructure: archived observations",
  "published": "2026-09-17",
  "deployment_checkpoint": "nvidia/Gemma-4-26B-A4B-NVFP4",
  "identity_limit": "Deployment documentation maps the voice-agent alias to this checkpoint. Individual July benchmark files do not record checkpoint revision, server version, driver, or full launch arguments. Current configuration is not an archived run manifest.",
  "method": {
    "ttft": "Client request start, including a fresh TCP connection, to first nonempty SSE content delta. Excludes ASR, TTS, audio playback, and tool completion.",
    "workload": "Short Hindi/English voice support instruction prompt; shared system prefix, varying user reference. Original domain-specific prompt is not distributed.",
    "max_tokens": 80,
    "temperature": 0.4,
    "waves_per_run": 3,
    "warmup_waves_excluded": 1,
    "measured_waves_per_run": 2,
    "percentiles": "Linear interpolation at (n-1)*p over pooled saved per-request values, already rounded to 0.1 ms.",
    "throughput": "Sum of recorded completion counts divided by sum of measured wave durations. Durations were rounded to 0.001 s before archiving. Pauses and warmup excluded; not sustained traffic.",
    "instrumentation_limits": [
      "Legacy harness prefers completion_tokens from server usage, but falls back to nonempty SSE content-chunk count when unavailable. Per-request provenance of token counts was not saved. Throughput is therefore a recorded estimate, not a token-audited benchmark.",
      "Legacy parser did not retain incomplete SSE lines between reads. Malformed or fragmented lines could be skipped. Raw streams and output-quality scores were not archived.",
      "Closed-loop burst waves, short generated responses, and warmed shared prefix do not model production arrival patterns or long-context/tool-heavy calls."
    ]
  },
  "studies": [
    {
      "id": "concurrency",
      "hardware": "RTX PRO 6000 Blackwell, 96 GB",
      "started_utc": "2026-07-28T03:21:16Z",
      "finished_utc": "2026-07-28T03:25:31Z",
      "served_alias": "voice-agent",
      "system_prompt_chars": 3403,
      "unique_prefix": false,
      "runs": [
        {
          "N": 12,
          "rep": 0,
          "requests_measured": 24,
          "ok": 24,
          "errors": 0,
          "completion_tokens": 565,
          "prompt_tokens_sample": 758,
          "measured_wall_s": 0.735,
          "ttft_all_ms": [
            98.3,
            122.1,
            130.8,
            131.7,
            131.7,
            132.3,
            132.7,
            135.3,
            136.1,
            136.5,
            136.6,
            136.9,
            137.2,
            137.3,
            137.3,
            137.4,
            137.4,
            138.1,
            138.1,
            138.2,
            139.4,
            139.8,
            140.0,
            140.0
          ]
        },
        {
          "N": 16,
          "rep": 0,
          "requests_measured": 32,
          "ok": 32,
          "errors": 0,
          "completion_tokens": 766,
          "prompt_tokens_sample": 758,
          "measured_wall_s": 0.763,
          "ttft_all_ms": [
            125.5,
            127.6,
            157.8,
            158.0,
            158.1,
            158.5,
            158.9,
            158.9,
            159.2,
            159.9,
            160.3,
            160.8,
            160.9,
            161.1,
            161.4,
            162.0,
            162.4,
            162.7,
            162.8,
            163.0,
            163.8,
            163.8,
            163.9,
            164.1,
            165.1,
            172.8,
            173.0,
            173.1,
            173.4,
            174.1,
            174.2,
            175.1
          ]
        },
        {
          "N": 20,
          "rep": 0,
          "requests_measured": 40,
          "ok": 40,
          "errors": 0,
          "completion_tokens": 940,
          "prompt_tokens_sample": 758,
          "measured_wall_s": 0.873,
          "ttft_all_ms": [
            141.2,
            150.8,
            176.3,
            176.4,
            176.7,
            177.6,
            177.7,
            177.7,
            178.4,
            179.3,
            179.3,
            179.5,
            180.4,
            181.0,
            181.0,
            181.2,
            181.5,
            181.6,
            182.2,
            182.4,
            183.0,
            190.4,
            190.5,
            190.5,
            190.9,
            191.5,
            191.9,
            192.3,
            192.7,
            192.7,
            193.3,
            193.5,
            193.5,
            193.7,
            200.4,
            200.9,
            201.1,
            201.4,
            202.2,
            203.0
          ]
        },
        {
          "N": 24,
          "rep": 0,
          "requests_measured": 48,
          "ok": 48,
          "errors": 0,
          "completion_tokens": 1125,
          "prompt_tokens_sample": 758,
          "measured_wall_s": 0.875,
          "ttft_all_ms": [
            146.8,
            149.6,
            190.0,
            190.9,
            191.0,
            191.1,
            191.8,
            192.1,
            192.5,
            193.1,
            193.8,
            194.0,
            194.0,
            194.4,
            194.9,
            195.3,
            195.4,
            195.8,
            196.0,
            196.2,
            196.7,
            196.9,
            197.1,
            197.9,
            198.3,
            198.3,
            198.4,
            198.9,
            199.5,
            199.7,
            199.7,
            199.7,
            202.3,
            203.0,
            203.2,
            204.1,
            205.5,
            205.7,
            206.0,
            206.2,
            206.7,
            206.7,
            207.2,
            207.4,
            207.8,
            208.1,
            208.8,
            209.2
          ]
        },
        {
          "N": 32,
          "rep": 0,
          "requests_measured": 64,
          "ok": 64,
          "errors": 0,
          "completion_tokens": 1528,
          "prompt_tokens_sample": 758,
          "measured_wall_s": 0.957,
          "ttft_all_ms": [
            124.4,
            154.4,
            161.2,
            161.6,
            161.7,
            162.5,
            162.6,
            163.1,
            192.1,
            192.5,
            192.7,
            193.2,
            193.3,
            194.0,
            194.1,
            195.0,
            207.3,
            207.6,
            208.3,
            208.4,
            208.4,
            209.1,
            209.1,
            209.3,
            209.3,
            211.0,
            211.2,
            211.2,
            211.5,
            211.5,
            211.7,
            211.8,
            211.8,
            212.0,
            212.0,
            212.5,
            212.8,
            213.5,
            213.9,
            213.9,
            214.3,
            215.7,
            215.8,
            215.9,
            216.4,
            217.4,
            217.5,
            217.6,
            218.2,
            218.2,
            219.5,
            219.9,
            220.5,
            220.6,
            221.2,
            221.4,
            221.6,
            221.7,
            223.0,
            223.3,
            223.6,
            224.1,
            224.2,
            225.5
          ]
        },
        {
          "N": 48,
          "rep": 0,
          "requests_measured": 96,
          "ok": 96,
          "errors": 0,
          "completion_tokens": 2271,
          "prompt_tokens_sample": 758,
          "measured_wall_s": 1.098,
          "ttft_all_ms": [
            199.6,
            202.3,
            247.0,
            247.4,
            247.5,
            247.6,
            247.7,
            248.8,
            249.2,
            249.3,
            249.4,
            249.6,
            250.0,
            250.0,
            250.1,
            250.7,
            250.8,
            250.9,
            251.1,
            251.2,
            251.3,
            251.4,
            251.8,
            252.4,
            252.9,
            253.3,
            253.4,
            253.7,
            254.3,
            277.2,
            279.3,
            279.8,
            280.1,
            280.8,
            281.0,
            281.1,
            281.9,
            282.3,
            282.4,
            282.6,
            282.8,
            282.8,
            283.1,
            283.1,
            283.2,
            283.4,
            283.5,
            283.6,
            283.6,
            283.9,
            284.0,
            284.1,
            284.3,
            284.3,
            284.6,
            284.9,
            285.2,
            285.4,
            285.5,
            285.5,
            285.6,
            285.9,
            286.2,
            286.4,
            286.6,
            286.6,
            286.7,
            286.7,
            287.4,
            287.5,
            287.8,
            287.8,
            287.9,
            288.1,
            288.2,
            288.2,
            288.6,
            288.9,
            289.0,
            289.0,
            289.6,
            289.6,
            289.7,
            290.6,
            290.6,
            290.6,
            290.7,
            291.2,
            291.2,
            291.3,
            291.7,
            292.2,
            293.8,
            293.8,
            295.8,
            296.7
          ]
        },
        {
          "N": 12,
          "rep": 1,
          "requests_measured": 24,
          "ok": 24,
          "errors": 0,
          "completion_tokens": 573,
          "prompt_tokens_sample": 758,
          "measured_wall_s": 0.696,
          "ttft_all_ms": [
            94.8,
            107.7,
            107.7,
            108.4,
            108.7,
            109.1,
            109.4,
            109.8,
            109.9,
            110.2,
            110.7,
            111.8,
            117.0,
            129.1,
            129.8,
            129.9,
            130.4,
            130.5,
            131.2,
            131.2,
            131.2,
            131.3,
            131.5,
            132.8
          ]
        },
        {
          "N": 16,
          "rep": 1,
          "requests_measured": 32,
          "ok": 32,
          "errors": 0,
          "completion_tokens": 754,
          "prompt_tokens_sample": 758,
          "measured_wall_s": 0.785,
          "ttft_all_ms": [
            129.3,
            131.9,
            163.8,
            164.3,
            164.3,
            164.4,
            164.9,
            165.0,
            165.5,
            166.2,
            167.0,
            167.1,
            167.8,
            167.9,
            168.0,
            168.0,
            168.4,
            168.7,
            168.9,
            168.9,
            169.0,
            169.2,
            169.2,
            169.7,
            169.8,
            175.8,
            176.8,
            177.1,
            178.0,
            179.0,
            179.6,
            180.1
          ]
        },
        {
          "N": 20,
          "rep": 1,
          "requests_measured": 40,
          "ok": 40,
          "errors": 0,
          "completion_tokens": 956,
          "prompt_tokens_sample": 758,
          "measured_wall_s": 0.962,
          "ttft_all_ms": [
            133.1,
            142.0,
            173.1,
            173.2,
            173.3,
            173.5,
            173.5,
            174.5,
            174.8,
            175.2,
            175.3,
            175.6,
            178.9,
            179.4,
            180.1,
            180.2,
            180.3,
            181.0,
            181.4,
            182.0,
            182.4,
            184.1,
            185.0,
            185.8,
            186.3,
            186.3,
            186.9,
            186.9,
            186.9,
            191.5,
            191.9,
            192.9,
            193.2,
            193.6,
            193.9,
            194.5,
            194.6,
            194.6,
            194.7,
            195.6
          ]
        },
        {
          "N": 24,
          "rep": 1,
          "requests_measured": 48,
          "ok": 48,
          "errors": 0,
          "completion_tokens": 1137,
          "prompt_tokens_sample": 758,
          "measured_wall_s": 0.901,
          "ttft_all_ms": [
            160.8,
            161.6,
            196.0,
            196.3,
            196.7,
            197.0,
            197.1,
            197.2,
            197.3,
            197.5,
            197.6,
            198.0,
            198.1,
            198.4,
            198.5,
            198.6,
            211.4,
            213.0,
            213.2,
            213.7,
            213.8,
            215.0,
            215.3,
            216.5,
            216.5,
            216.9,
            217.0,
            217.4,
            217.5,
            217.8,
            218.1,
            221.9,
            222.0,
            222.3,
            222.3,
            222.9,
            223.1,
            223.5,
            223.9,
            224.8,
            225.0,
            225.3,
            225.3,
            225.6,
            225.8,
            226.4,
            226.7,
            227.0
          ]
        },
        {
          "N": 32,
          "rep": 1,
          "requests_measured": 64,
          "ok": 64,
          "errors": 0,
          "completion_tokens": 1534,
          "prompt_tokens_sample": 758,
          "measured_wall_s": 1.032,
          "ttft_all_ms": [
            147.8,
            165.3,
            187.4,
            188.0,
            188.2,
            188.3,
            188.5,
            189.1,
            189.2,
            189.6,
            190.3,
            201.0,
            201.1,
            201.4,
            201.9,
            202.2,
            202.6,
            203.9,
            204.0,
            204.4,
            204.5,
            204.6,
            205.1,
            205.5,
            205.5,
            205.8,
            206.1,
            206.6,
            207.0,
            207.2,
            207.5,
            207.7,
            207.7,
            208.4,
            208.5,
            208.8,
            209.0,
            209.1,
            209.3,
            209.3,
            209.3,
            209.5,
            209.9,
            210.0,
            210.5,
            210.8,
            210.9,
            210.9,
            210.9,
            211.1,
            211.4,
            211.4,
            212.1,
            212.3,
            212.4,
            212.6,
            212.8,
            213.0,
            213.0,
            213.4,
            213.6,
            213.7,
            214.2,
            216.7
          ]
        },
        {
          "N": 48,
          "rep": 1,
          "requests_measured": 96,
          "ok": 96,
          "errors": 0,
          "completion_tokens": 2278,
          "prompt_tokens_sample": 758,
          "measured_wall_s": 1.158,
          "ttft_all_ms": [
            193.1,
            206.0,
            237.1,
            237.9,
            238.5,
            238.6,
            239.2,
            239.5,
            255.1,
            255.2,
            255.7,
            255.8,
            255.9,
            256.3,
            256.4,
            256.9,
            257.3,
            258.0,
            258.2,
            258.9,
            259.4,
            259.4,
            265.9,
            266.4,
            267.0,
            267.5,
            267.9,
            268.4,
            269.7,
            269.7,
            270.3,
            270.5,
            271.1,
            271.3,
            271.4,
            271.7,
            271.7,
            271.9,
            272.3,
            272.5,
            273.0,
            273.1,
            273.2,
            273.9,
            274.3,
            274.7,
            275.0,
            275.2,
            275.3,
            276.1,
            277.0,
            277.1,
            278.2,
            278.2,
            278.2,
            279.2,
            279.4,
            279.6,
            279.7,
            281.3,
            282.5,
            282.7,
            283.7,
            284.7,
            285.2,
            287.1,
            287.2,
            288.1,
            288.3,
            288.3,
            288.6,
            288.6,
            289.3,
            289.7,
            290.8,
            290.8,
            290.8,
            290.9,
            291.1,
            291.6,
            291.7,
            291.8,
            291.8,
            291.9,
            294.1,
            294.2,
            294.2,
            295.0,
            295.2,
            295.8,
            296.0,
            296.0,
            297.7,
            298.3,
            298.7,
            299.6
          ]
        },
        {
          "N": 12,
          "rep": 2,
          "requests_measured": 24,
          "ok": 24,
          "errors": 0,
          "completion_tokens": 570,
          "prompt_tokens_sample": 758,
          "measured_wall_s": 0.707,
          "ttft_all_ms": [
            110.4,
            113.5,
            124.5,
            125.6,
            125.6,
            125.7,
            125.8,
            126.1,
            126.1,
            126.9,
            127.0,
            127.1,
            127.2,
            132.3,
            132.5,
            133.4,
            133.5,
            133.5,
            133.6,
            134.0,
            134.1,
            134.1,
            134.9,
            135.7
          ]
        },
        {
          "N": 16,
          "rep": 2,
          "requests_measured": 32,
          "ok": 32,
          "errors": 0,
          "completion_tokens": 754,
          "prompt_tokens_sample": 758,
          "measured_wall_s": 0.727,
          "ttft_all_ms": [
            96.2,
            101.6,
            131.1,
            131.6,
            131.8,
            132.0,
            132.2,
            132.4,
            133.0,
            133.6,
            133.9,
            135.8,
            135.9,
            136.7,
            137.0,
            137.0,
            137.2,
            137.5,
            138.0,
            138.5,
            138.8,
            138.9,
            139.0,
            139.3,
            139.7,
            139.9,
            140.0,
            141.6,
            141.7,
            141.7,
            142.5,
            142.7
          ]
        },
        {
          "N": 20,
          "rep": 2,
          "requests_measured": 40,
          "ok": 40,
          "errors": 0,
          "completion_tokens": 986,
          "prompt_tokens_sample": 758,
          "measured_wall_s": 1.068,
          "ttft_all_ms": [
            100.2,
            134.4,
            136.8,
            136.8,
            137.6,
            137.7,
            138.0,
            138.1,
            157.0,
            157.2,
            157.2,
            157.3,
            157.5,
            158.5,
            159.2,
            159.3,
            159.4,
            159.8,
            160.0,
            160.3,
            160.4,
            172.8,
            172.9,
            173.4,
            173.6,
            173.9,
            174.7,
            174.7,
            175.0,
            175.2,
            175.9,
            175.9,
            176.0,
            176.1,
            177.4,
            177.4,
            177.7,
            177.7,
            177.9,
            178.3
          ]
        },
        {
          "N": 24,
          "rep": 2,
          "requests_measured": 48,
          "ok": 48,
          "errors": 0,
          "completion_tokens": 1127,
          "prompt_tokens_sample": 758,
          "measured_wall_s": 0.858,
          "ttft_all_ms": [
            104.4,
            137.0,
            137.8,
            137.9,
            138.4,
            138.5,
            139.4,
            149.5,
            149.7,
            150.3,
            150.4,
            150.5,
            151.1,
            151.7,
            151.9,
            152.0,
            152.4,
            153.4,
            153.6,
            153.6,
            154.0,
            154.2,
            154.3,
            154.4,
            182.2,
            219.9,
            220.5,
            220.6,
            221.0,
            221.5,
            221.8,
            221.8,
            242.6,
            243.1,
            243.7,
            243.9,
            244.2,
            244.6,
            244.7,
            245.5,
            245.8,
            246.1,
            246.3,
            246.4,
            246.6,
            247.5,
            247.7,
            248.8
          ]
        },
        {
          "N": 32,
          "rep": 2,
          "requests_measured": 64,
          "ok": 64,
          "errors": 0,
          "completion_tokens": 1557,
          "prompt_tokens_sample": 758,
          "measured_wall_s": 1.144,
          "ttft_all_ms": [
            153.6,
            160.6,
            191.9,
            192.0,
            192.1,
            192.3,
            193.3,
            193.6,
            193.7,
            193.9,
            198.0,
            198.1,
            198.9,
            200.0,
            200.0,
            200.7,
            200.8,
            200.9,
            201.0,
            201.4,
            201.4,
            201.6,
            201.8,
            208.6,
            209.1,
            211.2,
            211.5,
            211.7,
            211.9,
            211.9,
            211.9,
            211.9,
            212.0,
            212.7,
            213.6,
            213.6,
            214.6,
            214.7,
            215.3,
            215.4,
            215.5,
            216.2,
            216.9,
            216.9,
            217.6,
            218.0,
            218.5,
            218.6,
            218.7,
            218.7,
            218.8,
            219.3,
            220.4,
            220.6,
            220.6,
            220.7,
            220.7,
            221.4,
            221.9,
            222.1,
            222.3,
            222.9,
            223.5,
            224.1
          ]
        },
        {
          "N": 48,
          "rep": 2,
          "requests_measured": 96,
          "ok": 96,
          "errors": 0,
          "completion_tokens": 2267,
          "prompt_tokens_sample": 758,
          "measured_wall_s": 1.074,
          "ttft_all_ms": [
            179.5,
            216.6,
            225.0,
            225.5,
            225.8,
            226.1,
            226.2,
            226.3,
            226.8,
            227.1,
            227.2,
            267.3,
            267.5,
            267.6,
            267.8,
            267.8,
            268.1,
            268.2,
            268.5,
            268.5,
            268.8,
            269.1,
            269.7,
            269.9,
            270.0,
            270.2,
            270.2,
            270.3,
            270.5,
            270.6,
            270.7,
            270.8,
            270.9,
            270.9,
            271.1,
            271.1,
            271.6,
            272.1,
            272.3,
            272.7,
            272.8,
            273.1,
            273.5,
            273.7,
            273.8,
            273.9,
            274.0,
            274.2,
            274.3,
            274.8,
            277.6,
            280.5,
            280.5,
            281.1,
            281.9,
            281.9,
            282.0,
            282.3,
            282.8,
            282.8,
            283.2,
            283.7,
            284.5,
            286.1,
            294.9,
            297.4,
            298.3,
            298.8,
            301.4,
            301.7,
            301.8,
            302.4,
            302.9,
            303.6,
            303.7,
            303.7,
            303.9,
            304.0,
            306.2,
            306.2,
            307.3,
            307.4,
            308.1,
            308.6,
            308.9,
            308.9,
            309.6,
            309.6,
            310.4,
            311.4,
            311.5,
            311.9,
            312.6,
            313.7,
            314.0,
            314.4
          ]
        },
        {
          "N": 12,
          "rep": 3,
          "requests_measured": 24,
          "ok": 24,
          "errors": 0,
          "completion_tokens": 563,
          "prompt_tokens_sample": 758,
          "measured_wall_s": 0.711,
          "ttft_all_ms": [
            116.8,
            121.4,
            123.8,
            123.8,
            124.2,
            124.3,
            125.2,
            125.3,
            125.5,
            125.6,
            125.9,
            125.9,
            126.4,
            126.7,
            126.7,
            126.9,
            127.2,
            127.2,
            127.3,
            127.4,
            127.6,
            127.6,
            127.8,
            128.7
          ]
        },
        {
          "N": 16,
          "rep": 3,
          "requests_measured": 32,
          "ok": 32,
          "errors": 0,
          "completion_tokens": 770,
          "prompt_tokens_sample": 758,
          "measured_wall_s": 0.823,
          "ttft_all_ms": [
            120.3,
            121.9,
            133.7,
            134.0,
            135.0,
            135.1,
            135.3,
            135.8,
            135.9,
            136.1,
            136.4,
            136.6,
            136.6,
            136.8,
            136.9,
            137.1,
            137.1,
            137.1,
            137.1,
            137.3,
            137.3,
            137.6,
            138.1,
            138.7,
            138.7,
            138.7,
            138.9,
            139.5,
            139.6,
            139.7,
            140.9,
            140.9
          ]
        },
        {
          "N": 20,
          "rep": 3,
          "requests_measured": 40,
          "ok": 40,
          "errors": 0,
          "completion_tokens": 938,
          "prompt_tokens_sample": 758,
          "measured_wall_s": 0.876,
          "ttft_all_ms": [
            136.4,
            136.5,
            177.9,
            178.1,
            178.8,
            179.3,
            179.3,
            179.5,
            179.5,
            180.6,
            180.8,
            181.0,
            181.0,
            181.5,
            181.8,
            181.9,
            182.6,
            183.2,
            183.9,
            184.1,
            184.8,
            185.3,
            185.7,
            186.2,
            186.7,
            186.8,
            187.2,
            187.8,
            188.2,
            189.3,
            190.5,
            191.3,
            191.6,
            191.9,
            195.0,
            195.0,
            195.1,
            195.7,
            196.2,
            197.5
          ]
        },
        {
          "N": 24,
          "rep": 3,
          "requests_measured": 48,
          "ok": 48,
          "errors": 0,
          "completion_tokens": 1130,
          "prompt_tokens_sample": 758,
          "measured_wall_s": 0.86,
          "ttft_all_ms": [
            142.6,
            148.5,
            187.0,
            187.9,
            188.5,
            188.7,
            189.3,
            189.5,
            189.9,
            190.0,
            190.1,
            190.1,
            190.7,
            190.8,
            191.8,
            192.4,
            194.7,
            194.8,
            195.0,
            195.7,
            196.0,
            196.4,
            196.6,
            196.8,
            196.9,
            197.8,
            198.1,
            198.4,
            198.5,
            198.8,
            198.9,
            199.2,
            200.0,
            200.1,
            201.9,
            202.4,
            203.0,
            204.3,
            204.7,
            208.9,
            209.3,
            210.2,
            210.5,
            210.9,
            210.9,
            211.7,
            212.4,
            213.6
          ]
        },
        {
          "N": 32,
          "rep": 3,
          "requests_measured": 64,
          "ok": 64,
          "errors": 0,
          "completion_tokens": 1520,
          "prompt_tokens_sample": 758,
          "measured_wall_s": 0.974,
          "ttft_all_ms": [
            135.9,
            147.0,
            174.1,
            174.8,
            174.9,
            174.9,
            175.8,
            175.8,
            187.9,
            188.0,
            188.3,
            188.4,
            188.5,
            188.7,
            189.4,
            190.1,
            190.2,
            210.4,
            211.9,
            212.7,
            213.5,
            213.5,
            214.1,
            214.5,
            215.2,
            215.5,
            215.5,
            215.6,
            215.8,
            216.3,
            216.3,
            216.4,
            217.0,
            217.3,
            217.4,
            217.7,
            218.1,
            218.8,
            219.4,
            220.8,
            221.3,
            222.6,
            223.3,
            224.0,
            224.1,
            224.1,
            224.3,
            224.6,
            224.7,
            225.6,
            225.8,
            226.0,
            226.5,
            226.5,
            226.6,
            226.9,
            227.3,
            228.1,
            228.4,
            228.5,
            228.8,
            228.9,
            231.5,
            231.7
          ]
        },
        {
          "N": 48,
          "rep": 3,
          "requests_measured": 96,
          "ok": 96,
          "errors": 0,
          "completion_tokens": 2272,
          "prompt_tokens_sample": 758,
          "measured_wall_s": 1.02,
          "ttft_all_ms": [
            135.4,
            176.9,
            179.4,
            179.9,
            180.4,
            181.0,
            181.1,
            181.3,
            181.5,
            181.6,
            181.8,
            181.9,
            182.9,
            183.1,
            183.2,
            183.5,
            183.5,
            200.8,
            200.9,
            201.6,
            202.7,
            204.2,
            205.3,
            205.6,
            205.7,
            206.7,
            207.0,
            207.0,
            207.2,
            208.2,
            208.6,
            208.6,
            208.9,
            209.1,
            209.6,
            209.9,
            210.0,
            210.0,
            210.2,
            210.2,
            211.8,
            212.3,
            212.3,
            212.3,
            212.4,
            212.4,
            212.7,
            213.4,
            216.0,
            223.6,
            223.6,
            224.2,
            224.2,
            224.3,
            224.5,
            224.9,
            225.0,
            226.1,
            226.5,
            253.1,
            256.7,
            257.4,
            257.9,
            258.1,
            258.4,
            258.9,
            259.1,
            260.1,
            260.5,
            260.8,
            260.8,
            261.0,
            261.1,
            261.2,
            261.3,
            261.4,
            262.2,
            262.4,
            262.5,
            262.7,
            262.8,
            263.0,
            263.4,
            263.7,
            264.4,
            264.4,
            264.5,
            264.6,
            265.9,
            266.1,
            266.9,
            267.3,
            267.6,
            268.7,
            269.1,
            269.4
          ]
        },
        {
          "N": 12,
          "rep": 4,
          "requests_measured": 24,
          "ok": 24,
          "errors": 0,
          "completion_tokens": 570,
          "prompt_tokens_sample": 758,
          "measured_wall_s": 0.688,
          "ttft_all_ms": [
            95.2,
            106.9,
            107.0,
            107.0,
            108.0,
            108.1,
            108.2,
            108.3,
            108.5,
            109.1,
            109.1,
            109.2,
            109.6,
            126.5,
            127.4,
            127.5,
            127.7,
            128.2,
            128.2,
            128.4,
            129.1,
            133.0,
            133.2,
            133.4
          ]
        },
        {
          "N": 16,
          "rep": 4,
          "requests_measured": 32,
          "ok": 32,
          "errors": 0,
          "completion_tokens": 751,
          "prompt_tokens_sample": 758,
          "measured_wall_s": 0.774,
          "ttft_all_ms": [
            111.7,
            116.5,
            150.4,
            151.1,
            151.1,
            151.2,
            151.7,
            151.8,
            152.2,
            152.2,
            152.9,
            154.3,
            154.5,
            154.5,
            155.5,
            155.6,
            156.0,
            156.1,
            156.4,
            158.5,
            158.9,
            159.1,
            159.1,
            160.5,
            161.2,
            164.2,
            164.4,
            164.4,
            165.7,
            166.6,
            166.8,
            167.2
          ]
        },
        {
          "N": 20,
          "rep": 4,
          "requests_measured": 40,
          "ok": 40,
          "errors": 0,
          "completion_tokens": 947,
          "prompt_tokens_sample": 758,
          "measured_wall_s": 0.954,
          "ttft_all_ms": [
            127.0,
            151.5,
            165.0,
            165.7,
            165.8,
            166.2,
            166.2,
            166.2,
            166.6,
            166.6,
            166.6,
            167.1,
            167.3,
            167.4,
            167.4,
            167.6,
            168.0,
            168.6,
            168.6,
            169.0,
            169.0,
            169.0,
            169.1,
            169.5,
            169.6,
            169.6,
            169.8,
            169.9,
            170.5,
            170.7,
            172.7,
            173.6,
            173.6,
            174.2,
            174.3,
            174.5,
            175.1,
            175.5,
            175.8,
            176.3
          ]
        },
        {
          "N": 24,
          "rep": 4,
          "requests_measured": 48,
          "ok": 48,
          "errors": 0,
          "completion_tokens": 1140,
          "prompt_tokens_sample": 758,
          "measured_wall_s": 0.887,
          "ttft_all_ms": [
            142.2,
            164.4,
            179.2,
            179.2,
            179.3,
            180.0,
            180.8,
            181.0,
            181.2,
            181.3,
            181.7,
            181.9,
            182.0,
            182.5,
            185.1,
            187.9,
            188.2,
            188.7,
            188.8,
            188.9,
            188.9,
            189.4,
            190.0,
            190.3,
            191.2,
            208.8,
            208.9,
            209.1,
            209.2,
            209.8,
            210.2,
            210.4,
            210.6,
            210.9,
            211.2,
            211.4,
            211.9,
            212.6,
            212.6,
            220.5,
            220.8,
            221.3,
            222.4,
            222.6,
            222.9,
            223.5,
            223.8,
            224.4
          ]
        },
        {
          "N": 32,
          "rep": 4,
          "requests_measured": 64,
          "ok": 64,
          "errors": 0,
          "completion_tokens": 1510,
          "prompt_tokens_sample": 758,
          "measured_wall_s": 0.94,
          "ttft_all_ms": [
            172.5,
            178.1,
            218.8,
            219.2,
            219.6,
            219.6,
            220.4,
            220.4,
            220.9,
            221.2,
            221.8,
            222.0,
            222.0,
            222.2,
            222.3,
            222.3,
            222.4,
            222.5,
            222.5,
            223.3,
            223.3,
            223.4,
            223.6,
            223.8,
            224.1,
            225.1,
            225.3,
            225.4,
            241.1,
            241.8,
            242.9,
            243.1,
            243.9,
            244.1,
            244.4,
            245.0,
            245.8,
            245.9,
            246.0,
            246.3,
            247.1,
            247.1,
            247.3,
            247.4,
            247.5,
            247.5,
            247.5,
            248.0,
            248.2,
            248.2,
            248.3,
            248.3,
            248.4,
            248.5,
            248.5,
            249.0,
            249.5,
            249.5,
            249.7,
            251.9,
            252.1,
            252.2,
            252.3,
            252.3
          ]
        },
        {
          "N": 48,
          "rep": 4,
          "requests_measured": 96,
          "ok": 96,
          "errors": 0,
          "completion_tokens": 2285,
          "prompt_tokens_sample": 758,
          "measured_wall_s": 1.058,
          "ttft_all_ms": [
            170.3,
            209.4,
            209.9,
            210.2,
            210.4,
            211.0,
            211.1,
            211.2,
            211.7,
            230.3,
            239.4,
            239.8,
            240.3,
            241.1,
            241.7,
            241.8,
            242.8,
            243.0,
            243.4,
            243.5,
            244.2,
            244.4,
            244.7,
            244.9,
            245.1,
            245.4,
            246.2,
            246.3,
            246.7,
            247.3,
            247.7,
            248.2,
            248.2,
            249.0,
            249.4,
            250.1,
            250.4,
            250.8,
            250.8,
            251.1,
            252.3,
            253.6,
            255.0,
            255.2,
            255.8,
            257.0,
            259.0,
            259.3,
            259.4,
            283.5,
            284.4,
            284.5,
            284.9,
            285.4,
            285.7,
            285.7,
            286.1,
            286.3,
            286.3,
            286.4,
            287.1,
            287.4,
            288.2,
            288.6,
            289.4,
            312.2,
            312.2,
            313.4,
            315.1,
            315.9,
            315.9,
            316.5,
            316.8,
            317.4,
            317.4,
            318.2,
            318.3,
            318.4,
            319.5,
            319.9,
            320.0,
            320.3,
            320.4,
            320.7,
            320.9,
            321.8,
            321.8,
            322.0,
            323.0,
            323.2,
            324.5,
            324.5,
            325.4,
            325.4,
            325.7,
            325.8
          ]
        }
      ]
    },
    {
      "id": "shared-prefix",
      "hardware": "RTX 5090, 32 GB",
      "started_utc": "2026-07-27T14:55:19Z",
      "finished_utc": "2026-07-27T14:55:45Z",
      "served_alias": "voice-agent",
      "system_prompt_chars": 3403,
      "unique_prefix": false,
      "runs": [
        {
          "N": 16,
          "rep": 0,
          "requests_measured": 32,
          "ok": 32,
          "errors": 0,
          "completion_tokens": 872,
          "prompt_tokens_sample": 758,
          "measured_wall_s": 1.304,
          "ttft_all_ms": [
            128.5,
            131.9,
            193.8,
            193.9,
            194.2,
            194.6,
            194.8,
            194.9,
            195.0,
            195.1,
            195.1,
            195.6,
            195.8,
            195.9,
            196.0,
            196.4,
            196.6,
            196.8,
            199.4,
            200.0,
            200.5,
            200.6,
            200.9,
            202.1,
            202.1,
            202.1,
            202.1,
            202.5,
            202.8,
            202.9,
            203.2,
            204.2
          ]
        },
        {
          "N": 16,
          "rep": 1,
          "requests_measured": 32,
          "ok": 32,
          "errors": 0,
          "completion_tokens": 823,
          "prompt_tokens_sample": 758,
          "measured_wall_s": 1.072,
          "ttft_all_ms": [
            131.7,
            132.2,
            193.6,
            194.0,
            194.4,
            194.6,
            195.2,
            195.3,
            195.5,
            196.1,
            196.5,
            196.6,
            197.1,
            197.5,
            198.2,
            198.3,
            198.4,
            200.6,
            200.7,
            201.5,
            201.6,
            201.9,
            202.0,
            202.3,
            202.6,
            202.8,
            202.9,
            203.0,
            203.8,
            204.0,
            204.1,
            204.3
          ]
        },
        {
          "N": 16,
          "rep": 2,
          "requests_measured": 32,
          "ok": 32,
          "errors": 0,
          "completion_tokens": 862,
          "prompt_tokens_sample": 758,
          "measured_wall_s": 1.227,
          "ttft_all_ms": [
            129.2,
            133.3,
            191.9,
            192.7,
            192.9,
            193.4,
            193.6,
            193.8,
            194.3,
            200.0,
            200.0,
            200.6,
            200.9,
            201.3,
            201.7,
            203.5,
            203.8,
            218.1,
            218.4,
            219.1,
            219.2,
            219.5,
            219.9,
            220.2,
            224.7,
            225.5,
            225.6,
            225.9,
            226.4,
            226.6,
            226.7,
            226.8
          ]
        }
      ]
    },
    {
      "id": "unique-prefix",
      "hardware": "RTX 5090, 32 GB",
      "started_utc": "2026-07-27T14:55:45Z",
      "finished_utc": "2026-07-27T14:56:13Z",
      "served_alias": "voice-agent",
      "system_prompt_chars": 3403,
      "unique_prefix": true,
      "runs": [
        {
          "N": 16,
          "rep": 0,
          "requests_measured": 32,
          "ok": 32,
          "errors": 0,
          "completion_tokens": 729,
          "prompt_tokens_sample": 782,
          "measured_wall_s": 1.463,
          "ttft_all_ms": [
            170.5,
            172.3,
            231.1,
            231.2,
            231.7,
            232.2,
            233.4,
            233.6,
            290.9,
            291.2,
            291.7,
            292.6,
            293.1,
            293.1,
            357.2,
            357.4,
            357.9,
            358.2,
            358.5,
            358.7,
            438.4,
            439.2,
            439.4,
            440.2,
            441.1,
            442.0,
            497.5,
            498.1,
            498.3,
            500.9,
            501.1,
            501.4
          ]
        },
        {
          "N": 16,
          "rep": 1,
          "requests_measured": 32,
          "ok": 32,
          "errors": 0,
          "completion_tokens": 765,
          "prompt_tokens_sample": 782,
          "measured_wall_s": 1.464,
          "ttft_all_ms": [
            171.8,
            175.5,
            231.7,
            232.0,
            232.7,
            234.7,
            235.9,
            236.1,
            290.1,
            290.1,
            290.5,
            293.7,
            295.1,
            295.8,
            361.6,
            362.8,
            362.8,
            362.9,
            363.5,
            363.9,
            440.5,
            440.9,
            441.3,
            441.7,
            442.6,
            443.1,
            498.6,
            499.3,
            499.3,
            499.9,
            500.6,
            502.3
          ]
        },
        {
          "N": 16,
          "rep": 2,
          "requests_measured": 32,
          "ok": 32,
          "errors": 0,
          "completion_tokens": 734,
          "prompt_tokens_sample": 782,
          "measured_wall_s": 1.43,
          "ttft_all_ms": [
            173.9,
            180.9,
            233.0,
            233.5,
            233.6,
            243.0,
            243.9,
            244.2,
            292.7,
            293.3,
            293.6,
            301.0,
            302.1,
            302.1,
            360.7,
            361.1,
            362.0,
            368.1,
            368.7,
            368.9,
            440.1,
            440.8,
            441.9,
            449.5,
            449.7,
            450.4,
            499.4,
            499.5,
            500.7,
            508.5,
            509.8,
            511.1
          ]
        }
      ]
    }
  ],
  "kv_observation": {
    "date": "2026-07-28",
    "hardware": "RTX PRO 6000 Blackwell, 96 GB",
    "available_kv_gib": 67.8,
    "reported_kv_tokens": 645178,
    "max_model_len_at_observation": 8192,
    "reported_max_concurrency": 78.76,
    "note": "Startup log capacity estimate; not measured full-context serving capacity. Do not linearly extrapolate hybrid attention allocation into a guarantee."
  },
  "connection_reuse": {
    "date": "2026-09-08",
    "idle_between_requests_s": 7,
    "keep_alive_timeout_s": 120,
    "timing": "Client generation start to first useful output; separate application probe, not July content-delta harness.",
    "identity_limit": "Stable voice-agent alias; no immutable checkpoint/hardware manifest archived with this probe.",
    "requests": [
      {
        "request": 1,
        "new_connection": true,
        "dns_ms": 1.1,
        "connect_ms": 61.4,
        "status": 200,
        "first_output_ms": 154.7,
        "total_ms": 261.0
      },
      {
        "request": 2,
        "new_connection": false,
        "dns_ms": 0,
        "status": 200,
        "first_output_ms": 92.2,
        "total_ms": 256.4
      },
      {
        "request": 3,
        "new_connection": false,
        "dns_ms": 0,
        "status": 200,
        "first_output_ms": 81.8,
        "total_ms": 269.4
      },
      {
        "request": 4,
        "new_connection": false,
        "dns_ms": 0,
        "status": 200,
        "first_output_ms": 96.9,
        "total_ms": 240.7
      },
      {
        "request": 5,
        "new_connection": false,
        "dns_ms": 0,
        "status": 200,
        "first_output_ms": 93.6,
        "total_ms": 227.1
      },
      {
        "request": 6,
        "new_connection": false,
        "dns_ms": 0,
        "status": 200,
        "first_output_ms": 94.1,
        "total_ms": 197.6
      },
      {
        "request": 7,
        "new_connection": false,
        "dns_ms": 0,
        "status": 200,
        "first_output_ms": 125.9,
        "total_ms": 326.1
      },
      {
        "request": 8,
        "new_connection": false,
        "dns_ms": 0,
        "status": 200,
        "first_output_ms": 93.2,
        "total_ms": 271.6
      },
      {
        "request": 9,
        "new_connection": false,
        "dns_ms": 0,
        "status": 200,
        "first_output_ms": 89.7,
        "total_ms": 273.7
      },
      {
        "request": 10,
        "new_connection": false,
        "dns_ms": 0,
        "status": 200,
        "first_output_ms": 155.9,
        "total_ms": 209.9
      },
      {
        "request": 11,
        "new_connection": false,
        "dns_ms": 0,
        "status": 200,
        "first_output_ms": 96.6,
        "total_ms": 208.9
      },
      {
        "request": 12,
        "new_connection": false,
        "dns_ms": 0,
        "status": 200,
        "first_output_ms": 103.6,
        "total_ms": 326.2
      }
    ],
    "note": "Serial diagnostic sample after transport fixes. No matched before/after or concurrent-load comparison."
  }
}
