{
  "dataset": "nuckcrews/inference-audit",
  "ended_at": "2026-09-01T16:02:37.349306+00:00",
  "errors": 266,
  "models": [
    "openai/gpt-oss-120b"
  ],
  "observations": 3932,
  "providers": [
    "AkashML",
    "Amazon Bedrock",
    "BaseTen",
    "Cerebras",
    "CoreWeave",
    "DeepInfra",
    "DigitalOcean",
    "Google",
    "Groq",
    "Mancer 2",
    "Mara",
    "Nebius",
    "Novita",
    "Parasail",
    "Phala",
    "SambaNova",
    "SiliconFlow",
    "Together"
  ],
  "publication_transform": {
    "removed": [
      "response_id",
      "reasoning",
      "logprobs"
    ],
    "retained": [
      "reasoning_present",
      "reasoning_chars",
      "logprobs_available"
    ]
  },
  "reports": [
    {
      "campaign": "cache-runs",
      "findings": [
        "AkashML: 1 request errors"
      ],
      "observations": 40,
      "targets": [
        {
          "cache_hit_rate": 0.0156824307767704,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": null,
          "cut_rate": 0.0,
          "empty_rate": 0.0,
          "errors": 1,
          "expectation_pass_rate": 1.0,
          "experimental_kl": null,
          "json_valid_rate": null,
          "label": "AkashML",
          "median_latency_ms": 2495.0,
          "median_ttft_ms": null,
          "observations": 8,
          "prompt_token_ratio": 1.0,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": null,
          "truncation_rate": 0.0
        },
        {
          "cache_hit_rate": 0.5018377848566528,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": null,
          "cut_rate": 0.0,
          "empty_rate": 0.0,
          "errors": 0,
          "expectation_pass_rate": 1.0,
          "experimental_kl": null,
          "json_valid_rate": null,
          "label": "DigitalOcean",
          "median_latency_ms": 3546.0,
          "median_ttft_ms": null,
          "observations": 8,
          "prompt_token_ratio": 1.0,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": null,
          "truncation_rate": 0.0
        },
        {
          "cache_hit_rate": 0.0,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": null,
          "cut_rate": 0.0,
          "empty_rate": 0.0,
          "errors": 0,
          "expectation_pass_rate": 1.0,
          "experimental_kl": null,
          "json_valid_rate": null,
          "label": "Groq",
          "median_latency_ms": 664.0,
          "median_ttft_ms": null,
          "observations": 8,
          "prompt_token_ratio": 1.0009801519235482,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": null,
          "truncation_rate": 0.0
        },
        {
          "cache_hit_rate": 0.9958343543249204,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": null,
          "cut_rate": 0.0,
          "empty_rate": 0.0,
          "errors": 0,
          "expectation_pass_rate": 1.0,
          "experimental_kl": null,
          "json_valid_rate": null,
          "label": "Nebius",
          "median_latency_ms": 739.0,
          "median_ttft_ms": null,
          "observations": 8,
          "prompt_token_ratio": 1.0,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": null,
          "truncation_rate": 0.0
        },
        {
          "cache_hit_rate": 0.9958343543249204,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": null,
          "cut_rate": 0.0,
          "empty_rate": 0.0,
          "errors": 0,
          "expectation_pass_rate": 1.0,
          "experimental_kl": null,
          "json_valid_rate": null,
          "label": "Parasail",
          "median_latency_ms": 1273.0,
          "median_ttft_ms": null,
          "observations": 8,
          "prompt_token_ratio": 1.0,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": null,
          "truncation_rate": 0.0
        }
      ]
    },
    {
      "campaign": "logprobs",
      "findings": [
        "AkashML: 4 request errors",
        "AkashML: 6.2% of successful responses reported or exhibited truncation",
        "Cerebras: 11.1% of successful responses reported or exhibited truncation",
        "CoreWeave: 11.1% of successful responses reported or exhibited truncation",
        "DigitalOcean: 1 request errors",
        "DigitalOcean: 5.7% of successful responses reported or exhibited truncation",
        "Mancer 2: 1 request errors",
        "Mancer 2: 11.4% of successful responses reported or exhibited truncation",
        "Mara: 5 request errors",
        "Mara: 12.9% of successful responses reported or exhibited truncation",
        "Novita: 5.6% of successful responses reported or exhibited truncation",
        "Parasail: 1 request errors",
        "Parasail: 5.7% of successful responses reported or exhibited truncation",
        "Phala: 11 request errors",
        "Phala: 4.0% of successful responses reported or exhibited truncation",
        "Phala: median reported prompt tokens were 0.198x the cross-target modal count"
      ],
      "observations": 324,
      "targets": [
        {
          "cache_hit_rate": 0.7950617283950617,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 0.8518518518518519,
          "cut_rate": 0.03125,
          "empty_rate": 0.0,
          "errors": 4,
          "expectation_pass_rate": 1.0,
          "experimental_kl": 0.01280095158069259,
          "json_valid_rate": 1.0,
          "label": "AkashML",
          "median_latency_ms": 2862.0,
          "median_ttft_ms": null,
          "observations": 36,
          "prompt_token_ratio": 1.0,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": null,
          "truncation_rate": 0.0625
        },
        {
          "cache_hit_rate": 0.0,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 0.5666666666666667,
          "cut_rate": 0.027777777777777776,
          "empty_rate": 0.0,
          "errors": 0,
          "expectation_pass_rate": 1.0,
          "experimental_kl": 0.01211303971662347,
          "json_valid_rate": 1.0,
          "label": "Cerebras",
          "median_latency_ms": 409.5,
          "median_ttft_ms": null,
          "observations": 36,
          "prompt_token_ratio": 1.0,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": null,
          "truncation_rate": 0.1111111111111111
        },
        {
          "cache_hit_rate": 0.7806038970697605,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 0.7333333333333333,
          "cut_rate": 0.027777777777777776,
          "empty_rate": 0.0,
          "errors": 0,
          "expectation_pass_rate": 1.0,
          "experimental_kl": 0.0027471800229361873,
          "json_valid_rate": 1.0,
          "label": "CoreWeave",
          "median_latency_ms": 4743.5,
          "median_ttft_ms": null,
          "observations": 36,
          "prompt_token_ratio": 1.0,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": null,
          "truncation_rate": 0.1111111111111111
        },
        {
          "cache_hit_rate": 0.7901234567901234,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 0.6206896551724138,
          "cut_rate": 0.02857142857142857,
          "empty_rate": 0.0,
          "errors": 1,
          "expectation_pass_rate": 1.0,
          "experimental_kl": 0.011534949269359027,
          "json_valid_rate": 1.0,
          "label": "DigitalOcean",
          "median_latency_ms": 5131.0,
          "median_ttft_ms": null,
          "observations": 36,
          "prompt_token_ratio": 1.0,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": null,
          "truncation_rate": 0.05714285714285714
        },
        {
          "cache_hit_rate": 0.0,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 0.6551724137931034,
          "cut_rate": 0.02857142857142857,
          "empty_rate": 0.0,
          "errors": 1,
          "expectation_pass_rate": 1.0,
          "experimental_kl": 0.006272406845590522,
          "json_valid_rate": 1.0,
          "label": "Mancer 2",
          "median_latency_ms": 4691.0,
          "median_ttft_ms": null,
          "observations": 36,
          "prompt_token_ratio": 1.0,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": null,
          "truncation_rate": 0.11428571428571428
        },
        {
          "cache_hit_rate": 0.0,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 0.6666666666666666,
          "cut_rate": 0.03225806451612903,
          "empty_rate": 0.0,
          "errors": 5,
          "expectation_pass_rate": 1.0,
          "experimental_kl": 0.007960558653410369,
          "json_valid_rate": 1.0,
          "label": "Mara",
          "median_latency_ms": 2289.0,
          "median_ttft_ms": null,
          "observations": 36,
          "prompt_token_ratio": 1.0,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": null,
          "truncation_rate": 0.12903225806451613
        },
        {
          "cache_hit_rate": 0.0,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 0.5666666666666667,
          "cut_rate": 0.027777777777777776,
          "empty_rate": 0.0,
          "errors": 0,
          "expectation_pass_rate": 1.0,
          "experimental_kl": 0.06512189027037207,
          "json_valid_rate": 1.0,
          "label": "Novita",
          "median_latency_ms": 3247.5,
          "median_ttft_ms": null,
          "observations": 36,
          "prompt_token_ratio": 1.019419306184012,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": null,
          "truncation_rate": 0.05555555555555555
        },
        {
          "cache_hit_rate": 0.7901234567901234,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 0.6206896551724138,
          "cut_rate": 0.0,
          "empty_rate": 0.0,
          "errors": 1,
          "expectation_pass_rate": 1.0,
          "experimental_kl": 0.009728883869327354,
          "json_valid_rate": 1.0,
          "label": "Parasail",
          "median_latency_ms": 1702.0,
          "median_ttft_ms": null,
          "observations": 36,
          "prompt_token_ratio": 1.0,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": null,
          "truncation_rate": 0.05714285714285714
        },
        {
          "cache_hit_rate": 0.0,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 0.5714285714285714,
          "cut_rate": 0.0,
          "empty_rate": 0.0,
          "errors": 11,
          "expectation_pass_rate": 1.0,
          "experimental_kl": 0.006252781653790993,
          "json_valid_rate": 1.0,
          "label": "Phala",
          "median_latency_ms": 1549.0,
          "median_ttft_ms": null,
          "observations": 36,
          "prompt_token_ratio": 0.19753086419753085,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": null,
          "truncation_rate": 0.04
        }
      ]
    },
    {
      "campaign": "providers-gpt-oss-120b-2026-09-01",
      "findings": [
        "AkashML: 22 request errors",
        "AkashML: 12.5% of successful responses reported or exhibited truncation",
        "Amazon Bedrock: 13.8% of successful responses reported or exhibited truncation",
        "Cerebras: 17.0% of successful responses reported or exhibited truncation",
        "DeepInfra: 5 request errors",
        "DeepInfra: 12.4% of successful responses reported or exhibited truncation",
        "DeepInfra: median reported prompt tokens were 0.845x the cross-target modal count",
        "Google: 12 request errors",
        "Google: 13.4% of successful responses reported or exhibited truncation",
        "Groq: 11.7% of successful responses reported or exhibited truncation",
        "Novita: 60 request errors",
        "Novita: 8.8% of successful responses reported or exhibited truncation",
        "Parasail: 9 request errors",
        "Parasail: 12.9% of successful responses reported or exhibited truncation",
        "SambaNova: 17.0% of successful responses reported or exhibited truncation",
        "Together: 4 request errors",
        "Together: 12.2% of successful responses reported or exhibited truncation"
      ],
      "observations": 940,
      "targets": [
        {
          "cache_hit_rate": 0.7619047619047619,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 0.7142857142857143,
          "cut_rate": 0.05555555555555555,
          "empty_rate": 0.0,
          "errors": 22,
          "expectation_pass_rate": 1.0,
          "experimental_kl": null,
          "json_valid_rate": 1.0,
          "label": "AkashML",
          "median_latency_ms": 2720.5,
          "median_ttft_ms": null,
          "observations": 94,
          "prompt_token_ratio": 1.0,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": 0.7777777777777778,
          "truncation_rate": 0.125
        },
        {
          "cache_hit_rate": 0.0,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 0.6833333333333333,
          "cut_rate": 0.0425531914893617,
          "empty_rate": 0.0,
          "errors": 0,
          "expectation_pass_rate": 1.0,
          "experimental_kl": null,
          "json_valid_rate": 1.0,
          "label": "Amazon Bedrock",
          "median_latency_ms": 591.5,
          "median_ttft_ms": null,
          "observations": 94,
          "prompt_token_ratio": 1.0,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": 0.6382978723404256,
          "truncation_rate": 0.13829787234042554
        },
        {
          "cache_hit_rate": 0.0,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 0.6166666666666667,
          "cut_rate": 0.05319148936170213,
          "empty_rate": 0.0,
          "errors": 0,
          "expectation_pass_rate": 1.0,
          "experimental_kl": null,
          "json_valid_rate": 1.0,
          "label": "Cerebras",
          "median_latency_ms": 371.0,
          "median_ttft_ms": null,
          "observations": 94,
          "prompt_token_ratio": 1.0,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": 1.0,
          "truncation_rate": 0.1702127659574468
        },
        {
          "cache_hit_rate": 0.0,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 0.625,
          "cut_rate": 0.011235955056179775,
          "empty_rate": 0.0,
          "errors": 5,
          "expectation_pass_rate": 1.0,
          "experimental_kl": null,
          "json_valid_rate": 1.0,
          "label": "DeepInfra",
          "median_latency_ms": 5622.0,
          "median_ttft_ms": null,
          "observations": 94,
          "prompt_token_ratio": 0.8452380952380952,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": 0.7906976744186046,
          "truncation_rate": 0.12359550561797752
        },
        {
          "cache_hit_rate": 0.0,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 0.5185185185185185,
          "cut_rate": 0.18292682926829268,
          "empty_rate": 0.04878048780487805,
          "errors": 12,
          "expectation_pass_rate": 0.75,
          "experimental_kl": null,
          "json_valid_rate": 0.9,
          "label": "Google",
          "median_latency_ms": 4137.0,
          "median_ttft_ms": null,
          "observations": 94,
          "prompt_token_ratio": 1.0408163265306123,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": 0.6388888888888888,
          "truncation_rate": 0.13414634146341464
        },
        {
          "cache_hit_rate": 0.0,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 0.8333333333333334,
          "cut_rate": 0.06382978723404255,
          "empty_rate": 0.0,
          "errors": 0,
          "expectation_pass_rate": 1.0,
          "experimental_kl": null,
          "json_valid_rate": 1.0,
          "label": "Groq",
          "median_latency_ms": 515.5,
          "median_ttft_ms": null,
          "observations": 94,
          "prompt_token_ratio": 1.0476190476190477,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": 0.7872340425531915,
          "truncation_rate": 0.11702127659574468
        },
        {
          "cache_hit_rate": 0.0,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 0.8,
          "cut_rate": 0.058823529411764705,
          "empty_rate": 0.0,
          "errors": 60,
          "expectation_pass_rate": 1.0,
          "experimental_kl": null,
          "json_valid_rate": 1.0,
          "label": "Novita",
          "median_latency_ms": 1599.5,
          "median_ttft_ms": null,
          "observations": 94,
          "prompt_token_ratio": 1.0227390180878553,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": 1.0,
          "truncation_rate": 0.08823529411764706
        },
        {
          "cache_hit_rate": 0.7619047619047619,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 0.6415094339622641,
          "cut_rate": 0.047058823529411764,
          "empty_rate": 0.0,
          "errors": 9,
          "expectation_pass_rate": 1.0,
          "experimental_kl": null,
          "json_valid_rate": 1.0,
          "label": "Parasail",
          "median_latency_ms": 1647.0,
          "median_ttft_ms": null,
          "observations": 94,
          "prompt_token_ratio": 1.0,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": 0.6052631578947368,
          "truncation_rate": 0.12941176470588237
        },
        {
          "cache_hit_rate": 0.0,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 0.6833333333333333,
          "cut_rate": 0.05319148936170213,
          "empty_rate": 0.0,
          "errors": 0,
          "expectation_pass_rate": 1.0,
          "experimental_kl": null,
          "json_valid_rate": 1.0,
          "label": "SambaNova",
          "median_latency_ms": 794.5,
          "median_ttft_ms": null,
          "observations": 94,
          "prompt_token_ratio": 1.0,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": 0.9787234042553191,
          "truncation_rate": 0.1702127659574468
        },
        {
          "cache_hit_rate": 0.0,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 0.6166666666666667,
          "cut_rate": 0.05555555555555555,
          "empty_rate": 0.0,
          "errors": 4,
          "expectation_pass_rate": 1.0,
          "experimental_kl": null,
          "json_valid_rate": 1.0,
          "label": "Together",
          "median_latency_ms": 4279.5,
          "median_ttft_ms": null,
          "observations": 94,
          "prompt_token_ratio": 1.0,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": 0.4883720930232558,
          "truncation_rate": 0.12222222222222222
        }
      ]
    },
    {
      "campaign": "robust-standard",
      "findings": [
        "AkashML: 3 request errors",
        "AkashML: 0.7% of successful responses reported or exhibited truncation",
        "Amazon Bedrock: 2.1% of successful responses reported or exhibited truncation",
        "BaseTen: 24 request errors",
        "BaseTen: median reported prompt tokens were 1.131x the cross-target modal count",
        "Cerebras: 2.1% of successful responses reported or exhibited truncation",
        "CoreWeave: 5 request errors",
        "CoreWeave: 0.7% of successful responses reported or exhibited truncation",
        "DeepInfra: 0.7% of successful responses reported or exhibited truncation",
        "DeepInfra: median reported prompt tokens were 0.845x the cross-target modal count",
        "DigitalOcean: 2 request errors",
        "Google: 12 request errors",
        "Mancer 2: 8 request errors",
        "Mancer 2: 3.0% of successful responses reported or exhibited truncation",
        "Mara: 8 request errors",
        "Nebius: 1.4% of successful responses reported or exhibited truncation",
        "Novita: 2 request errors",
        "Novita: 0.7% of successful responses reported or exhibited truncation",
        "Parasail: 0.7% of successful responses reported or exhibited truncation",
        "Phala: 55 request errors",
        "Together: 1.4% of successful responses reported or exhibited truncation"
      ],
      "observations": 2538,
      "targets": [
        {
          "cache_hit_rate": 0.7619047619047619,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 0.7640449438202247,
          "cut_rate": 0.057971014492753624,
          "empty_rate": 0.0,
          "errors": 3,
          "expectation_pass_rate": 1.0,
          "experimental_kl": null,
          "json_valid_rate": 1.0,
          "label": "AkashML",
          "median_latency_ms": 2988.0,
          "median_ttft_ms": null,
          "observations": 141,
          "prompt_token_ratio": 1.0,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": 0.6170212765957447,
          "truncation_rate": 0.007246376811594203
        },
        {
          "cache_hit_rate": 0.0,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 0.6777777777777778,
          "cut_rate": 0.04964539007092199,
          "empty_rate": 0.0,
          "errors": 0,
          "expectation_pass_rate": 1.0,
          "experimental_kl": null,
          "json_valid_rate": 1.0,
          "label": "Amazon Bedrock",
          "median_latency_ms": 579.0,
          "median_ttft_ms": null,
          "observations": 141,
          "prompt_token_ratio": 1.0,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": 0.6595744680851063,
          "truncation_rate": 0.02127659574468085
        },
        {
          "cache_hit_rate": 0.6736842105263158,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 0.6133333333333333,
          "cut_rate": 0.06837606837606838,
          "empty_rate": 0.0,
          "errors": 24,
          "expectation_pass_rate": 0.9545454545454546,
          "experimental_kl": null,
          "json_valid_rate": 1.0,
          "label": "BaseTen",
          "median_latency_ms": 1162.0,
          "median_ttft_ms": null,
          "observations": 141,
          "prompt_token_ratio": 1.130952380952381,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": 0.6590909090909091,
          "truncation_rate": 0.0
        },
        {
          "cache_hit_rate": 0.0,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 0.5888888888888889,
          "cut_rate": 0.04964539007092199,
          "empty_rate": 0.0,
          "errors": 0,
          "expectation_pass_rate": 1.0,
          "experimental_kl": null,
          "json_valid_rate": 1.0,
          "label": "Cerebras",
          "median_latency_ms": 453.0,
          "median_ttft_ms": null,
          "observations": 141,
          "prompt_token_ratio": 1.0,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": 1.0,
          "truncation_rate": 0.02127659574468085
        },
        {
          "cache_hit_rate": 0.7529411764705882,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 0.7441860465116279,
          "cut_rate": 0.07352941176470588,
          "empty_rate": 0.0,
          "errors": 5,
          "expectation_pass_rate": 1.0,
          "experimental_kl": null,
          "json_valid_rate": 1.0,
          "label": "CoreWeave",
          "median_latency_ms": 4943.0,
          "median_ttft_ms": null,
          "observations": 141,
          "prompt_token_ratio": 1.0,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": 0.5957446808510638,
          "truncation_rate": 0.007352941176470588
        },
        {
          "cache_hit_rate": 0.0,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 0.6444444444444445,
          "cut_rate": 0.0070921985815602835,
          "empty_rate": 0.0,
          "errors": 0,
          "expectation_pass_rate": 1.0,
          "experimental_kl": null,
          "json_valid_rate": 1.0,
          "label": "DeepInfra",
          "median_latency_ms": 5271.0,
          "median_ttft_ms": null,
          "observations": 141,
          "prompt_token_ratio": 0.8452380952380952,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": 0.6808510638297872,
          "truncation_rate": 0.0070921985815602835
        },
        {
          "cache_hit_rate": 0.7619047619047619,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 0.5444444444444444,
          "cut_rate": 0.02877697841726619,
          "empty_rate": 0.0,
          "errors": 2,
          "expectation_pass_rate": 1.0,
          "experimental_kl": null,
          "json_valid_rate": 1.0,
          "label": "DigitalOcean",
          "median_latency_ms": 6072.0,
          "median_ttft_ms": null,
          "observations": 141,
          "prompt_token_ratio": 1.0,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": 0.425531914893617,
          "truncation_rate": 0.0
        },
        {
          "cache_hit_rate": 0.0,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 0.46987951807228917,
          "cut_rate": 0.20930232558139536,
          "empty_rate": 0.023255813953488372,
          "errors": 12,
          "expectation_pass_rate": 0.7142857142857143,
          "experimental_kl": null,
          "json_valid_rate": 0.875,
          "label": "Google",
          "median_latency_ms": 3849.0,
          "median_ttft_ms": null,
          "observations": 141,
          "prompt_token_ratio": 1.0404040404040404,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": 0.34782608695652173,
          "truncation_rate": 0.0
        },
        {
          "cache_hit_rate": 0.0,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 0.5666666666666667,
          "cut_rate": 0.06382978723404255,
          "empty_rate": 0.0,
          "errors": 0,
          "expectation_pass_rate": 1.0,
          "experimental_kl": null,
          "json_valid_rate": 1.0,
          "label": "Groq",
          "median_latency_ms": 556.0,
          "median_ttft_ms": null,
          "observations": 141,
          "prompt_token_ratio": 1.0476190476190477,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": 0.8085106382978723,
          "truncation_rate": 0.0
        },
        {
          "cache_hit_rate": 0.0,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 0.5952380952380952,
          "cut_rate": 0.05263157894736842,
          "empty_rate": 0.022556390977443608,
          "errors": 8,
          "expectation_pass_rate": 1.0,
          "experimental_kl": null,
          "json_valid_rate": 1.0,
          "label": "Mancer 2",
          "median_latency_ms": 4973.0,
          "median_ttft_ms": null,
          "observations": 141,
          "prompt_token_ratio": 1.0,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": 0.574468085106383,
          "truncation_rate": 0.03007518796992481
        },
        {
          "cache_hit_rate": 0.0,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 0.8433734939759037,
          "cut_rate": 0.05263157894736842,
          "empty_rate": 0.0,
          "errors": 8,
          "expectation_pass_rate": 1.0,
          "experimental_kl": null,
          "json_valid_rate": 1.0,
          "label": "Mara",
          "median_latency_ms": 1123.0,
          "median_ttft_ms": null,
          "observations": 141,
          "prompt_token_ratio": 1.0,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": 1.0,
          "truncation_rate": 0.0
        },
        {
          "cache_hit_rate": 0.7619047619047619,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 0.7222222222222222,
          "cut_rate": 0.06382978723404255,
          "empty_rate": 0.0,
          "errors": 0,
          "expectation_pass_rate": 1.0,
          "experimental_kl": null,
          "json_valid_rate": 1.0,
          "label": "Nebius",
          "median_latency_ms": 773.0,
          "median_ttft_ms": null,
          "observations": 141,
          "prompt_token_ratio": 1.0,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": 0.5531914893617021,
          "truncation_rate": 0.014184397163120567
        },
        {
          "cache_hit_rate": 0.0,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 0.5842696629213483,
          "cut_rate": 0.050359712230215826,
          "empty_rate": 0.0,
          "errors": 2,
          "expectation_pass_rate": 1.0,
          "experimental_kl": null,
          "json_valid_rate": 1.0,
          "label": "Novita",
          "median_latency_ms": 3253.0,
          "median_ttft_ms": null,
          "observations": 141,
          "prompt_token_ratio": 1.0204081632653061,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": 0.48936170212765956,
          "truncation_rate": 0.007194244604316547
        },
        {
          "cache_hit_rate": 0.7619047619047619,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 0.6333333333333333,
          "cut_rate": 0.028368794326241134,
          "empty_rate": 0.0,
          "errors": 0,
          "expectation_pass_rate": 1.0,
          "experimental_kl": null,
          "json_valid_rate": 1.0,
          "label": "Parasail",
          "median_latency_ms": 1658.0,
          "median_ttft_ms": null,
          "observations": 141,
          "prompt_token_ratio": 1.0,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": 0.5531914893617021,
          "truncation_rate": 0.0070921985815602835
        },
        {
          "cache_hit_rate": 0.7529411764705882,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 0.6481481481481481,
          "cut_rate": 0.046511627906976744,
          "empty_rate": 0.0,
          "errors": 55,
          "expectation_pass_rate": 1.0,
          "experimental_kl": null,
          "json_valid_rate": 1.0,
          "label": "Phala",
          "median_latency_ms": 2418.5,
          "median_ttft_ms": null,
          "observations": 141,
          "prompt_token_ratio": 1.0,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": 0.75,
          "truncation_rate": 0.0
        },
        {
          "cache_hit_rate": 0.0,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 0.8333333333333334,
          "cut_rate": 0.04964539007092199,
          "empty_rate": 0.0,
          "errors": 0,
          "expectation_pass_rate": 1.0,
          "experimental_kl": null,
          "json_valid_rate": 1.0,
          "label": "SambaNova",
          "median_latency_ms": 1897.0,
          "median_ttft_ms": null,
          "observations": 141,
          "prompt_token_ratio": 1.0,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": 0.723404255319149,
          "truncation_rate": 0.0
        },
        {
          "cache_hit_rate": 0.0,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 0.6333333333333333,
          "cut_rate": 0.04964539007092199,
          "empty_rate": 0.0,
          "errors": 0,
          "expectation_pass_rate": 1.0,
          "experimental_kl": null,
          "json_valid_rate": 1.0,
          "label": "SiliconFlow",
          "median_latency_ms": 5524.0,
          "median_ttft_ms": null,
          "observations": 141,
          "prompt_token_ratio": 1.0,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": 0.574468085106383,
          "truncation_rate": 0.0
        },
        {
          "cache_hit_rate": 0.0,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 0.6,
          "cut_rate": 0.04964539007092199,
          "empty_rate": 0.0,
          "errors": 0,
          "expectation_pass_rate": 1.0,
          "experimental_kl": null,
          "json_valid_rate": 1.0,
          "label": "Together",
          "median_latency_ms": 2853.0,
          "median_ttft_ms": null,
          "observations": 141,
          "prompt_token_ratio": 1.0,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": 0.5319148936170213,
          "truncation_rate": 0.014184397163120567
        }
      ]
    },
    {
      "campaign": "stream-smoke",
      "findings": [
        "AkashML: 2 request errors",
        "BaseTen: 2 request errors",
        "BaseTen: median reported prompt tokens were 1.131x the cross-target modal count",
        "DeepInfra: median reported prompt tokens were 0.845x the cross-target modal count",
        "DigitalOcean: 2 request errors",
        "Google: 1 request errors",
        "Novita: 1 request errors",
        "Phala: 3 request errors"
      ],
      "observations": 90,
      "targets": [
        {
          "cache_hit_rate": 0.6153846153846154,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 1.0,
          "cut_rate": 0.3333333333333333,
          "empty_rate": 0.0,
          "errors": 2,
          "expectation_pass_rate": 1.0,
          "experimental_kl": null,
          "json_valid_rate": 1.0,
          "label": "AkashML",
          "median_latency_ms": 3163.0,
          "median_ttft_ms": 2482.0,
          "observations": 5,
          "prompt_token_ratio": 1.0,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": null,
          "truncation_rate": 0.0
        },
        {
          "cache_hit_rate": 0.0,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 1.0,
          "cut_rate": 0.2,
          "empty_rate": 0.0,
          "errors": 0,
          "expectation_pass_rate": 1.0,
          "experimental_kl": null,
          "json_valid_rate": 1.0,
          "label": "Amazon Bedrock",
          "median_latency_ms": 356.0,
          "median_ttft_ms": 356.0,
          "observations": 5,
          "prompt_token_ratio": 1.0,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": null,
          "truncation_rate": 0.0
        },
        {
          "cache_hit_rate": 0.6736842105263158,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 0.5,
          "cut_rate": 0.3333333333333333,
          "empty_rate": 0.0,
          "errors": 2,
          "expectation_pass_rate": 1.0,
          "experimental_kl": null,
          "json_valid_rate": 1.0,
          "label": "BaseTen",
          "median_latency_ms": 965.0,
          "median_ttft_ms": 789.0,
          "observations": 5,
          "prompt_token_ratio": 1.130952380952381,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": null,
          "truncation_rate": 0.0
        },
        {
          "cache_hit_rate": 0.0,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 0.75,
          "cut_rate": 0.2,
          "empty_rate": 0.0,
          "errors": 0,
          "expectation_pass_rate": 1.0,
          "experimental_kl": null,
          "json_valid_rate": 1.0,
          "label": "Cerebras",
          "median_latency_ms": 309.0,
          "median_ttft_ms": 308.0,
          "observations": 5,
          "prompt_token_ratio": 1.0,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": null,
          "truncation_rate": 0.0
        },
        {
          "cache_hit_rate": 0.7619047619047619,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 1.0,
          "cut_rate": 0.2,
          "empty_rate": 0.0,
          "errors": 0,
          "expectation_pass_rate": 1.0,
          "experimental_kl": null,
          "json_valid_rate": 1.0,
          "label": "CoreWeave",
          "median_latency_ms": 3814.0,
          "median_ttft_ms": 2883.0,
          "observations": 5,
          "prompt_token_ratio": 1.0,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": null,
          "truncation_rate": 0.0
        },
        {
          "cache_hit_rate": 0.0,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 1.0,
          "cut_rate": 0.2,
          "empty_rate": 0.0,
          "errors": 0,
          "expectation_pass_rate": 1.0,
          "experimental_kl": null,
          "json_valid_rate": 1.0,
          "label": "DeepInfra",
          "median_latency_ms": 3406.0,
          "median_ttft_ms": 2675.0,
          "observations": 5,
          "prompt_token_ratio": 0.8452380952380952,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": null,
          "truncation_rate": 0.0
        },
        {
          "cache_hit_rate": 0.7619047619047619,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 0.6666666666666666,
          "cut_rate": 0.0,
          "empty_rate": 0.0,
          "errors": 2,
          "expectation_pass_rate": 1.0,
          "experimental_kl": null,
          "json_valid_rate": null,
          "label": "DigitalOcean",
          "median_latency_ms": 3704.0,
          "median_ttft_ms": 2397.0,
          "observations": 5,
          "prompt_token_ratio": 1.0,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": null,
          "truncation_rate": 0.0
        },
        {
          "cache_hit_rate": 0.0,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 0.6666666666666666,
          "cut_rate": 0.25,
          "empty_rate": 0.0,
          "errors": 1,
          "expectation_pass_rate": 1.0,
          "experimental_kl": null,
          "json_valid_rate": 1.0,
          "label": "Google",
          "median_latency_ms": 5450.5,
          "median_ttft_ms": 4932.5,
          "observations": 5,
          "prompt_token_ratio": 1.0382783882783884,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": null,
          "truncation_rate": 0.0
        },
        {
          "cache_hit_rate": 0.0,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 0.75,
          "cut_rate": 0.2,
          "empty_rate": 0.0,
          "errors": 0,
          "expectation_pass_rate": 1.0,
          "experimental_kl": null,
          "json_valid_rate": 1.0,
          "label": "Groq",
          "median_latency_ms": 382.0,
          "median_ttft_ms": 333.0,
          "observations": 5,
          "prompt_token_ratio": 1.0476190476190477,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": null,
          "truncation_rate": 0.0
        },
        {
          "cache_hit_rate": 0.0,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 1.0,
          "cut_rate": 0.2,
          "empty_rate": 0.0,
          "errors": 0,
          "expectation_pass_rate": 1.0,
          "experimental_kl": null,
          "json_valid_rate": 1.0,
          "label": "Mancer 2",
          "median_latency_ms": 4453.0,
          "median_ttft_ms": 4453.0,
          "observations": 5,
          "prompt_token_ratio": 1.0,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": null,
          "truncation_rate": 0.0
        },
        {
          "cache_hit_rate": 0.0,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 1.0,
          "cut_rate": 0.2,
          "empty_rate": 0.0,
          "errors": 0,
          "expectation_pass_rate": 1.0,
          "experimental_kl": null,
          "json_valid_rate": 1.0,
          "label": "Mara",
          "median_latency_ms": 1033.0,
          "median_ttft_ms": 1033.0,
          "observations": 5,
          "prompt_token_ratio": 1.0,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": null,
          "truncation_rate": 0.0
        },
        {
          "cache_hit_rate": 0.8648648648648649,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 1.0,
          "cut_rate": 0.2,
          "empty_rate": 0.0,
          "errors": 0,
          "expectation_pass_rate": 1.0,
          "experimental_kl": null,
          "json_valid_rate": 1.0,
          "label": "Nebius",
          "median_latency_ms": 521.0,
          "median_ttft_ms": 521.0,
          "observations": 5,
          "prompt_token_ratio": 1.0,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": null,
          "truncation_rate": 0.0
        },
        {
          "cache_hit_rate": 0.0,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 1.0,
          "cut_rate": 0.25,
          "empty_rate": 0.0,
          "errors": 1,
          "expectation_pass_rate": 1.0,
          "experimental_kl": null,
          "json_valid_rate": 1.0,
          "label": "Novita",
          "median_latency_ms": 3723.0,
          "median_ttft_ms": 1660.5,
          "observations": 5,
          "prompt_token_ratio": 1.0119047619047619,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": null,
          "truncation_rate": 0.0
        },
        {
          "cache_hit_rate": 0.7619047619047619,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 1.0,
          "cut_rate": 0.2,
          "empty_rate": 0.0,
          "errors": 0,
          "expectation_pass_rate": 1.0,
          "experimental_kl": null,
          "json_valid_rate": 1.0,
          "label": "Parasail",
          "median_latency_ms": 1225.0,
          "median_ttft_ms": 779.0,
          "observations": 5,
          "prompt_token_ratio": 1.0,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": null,
          "truncation_rate": 0.0
        },
        {
          "cache_hit_rate": 0.6857142857142857,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 1.0,
          "cut_rate": 0.5,
          "empty_rate": 0.0,
          "errors": 3,
          "expectation_pass_rate": 1.0,
          "experimental_kl": null,
          "json_valid_rate": 1.0,
          "label": "Phala",
          "median_latency_ms": 1566.0,
          "median_ttft_ms": 1231.5,
          "observations": 5,
          "prompt_token_ratio": 1.0,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": null,
          "truncation_rate": 0.0
        },
        {
          "cache_hit_rate": 0.0,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 1.0,
          "cut_rate": 0.2,
          "empty_rate": 0.0,
          "errors": 0,
          "expectation_pass_rate": 1.0,
          "experimental_kl": null,
          "json_valid_rate": 1.0,
          "label": "SambaNova",
          "median_latency_ms": 1184.0,
          "median_ttft_ms": 1184.0,
          "observations": 5,
          "prompt_token_ratio": 1.0,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": null,
          "truncation_rate": 0.0
        },
        {
          "cache_hit_rate": 0.0,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 1.0,
          "cut_rate": 0.2,
          "empty_rate": 0.0,
          "errors": 0,
          "expectation_pass_rate": 1.0,
          "experimental_kl": null,
          "json_valid_rate": 1.0,
          "label": "SiliconFlow",
          "median_latency_ms": 1715.0,
          "median_ttft_ms": 1511.0,
          "observations": 5,
          "prompt_token_ratio": 1.0,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": null,
          "truncation_rate": 0.0
        },
        {
          "cache_hit_rate": 0.0,
          "cache_report_rate": 1.0,
          "consensus_agreement_rate": 1.0,
          "cut_rate": 0.2,
          "empty_rate": 0.0,
          "errors": 0,
          "expectation_pass_rate": 1.0,
          "experimental_kl": null,
          "json_valid_rate": 1.0,
          "label": "Together",
          "median_latency_ms": 1931.0,
          "median_ttft_ms": 1370.0,
          "observations": 5,
          "prompt_token_ratio": 1.0,
          "reasoning_off_violation_rate": null,
          "self_agreement_rate": null,
          "truncation_rate": 0.0
        }
      ]
    }
  ],
  "schema_version": 1,
  "source_files": [
    {
      "file": "cache-runs.jsonl",
      "observations": 40,
      "sha256": "b15386d53666401eb147290cca1254f1f0a50e8c02eb657f3c3d8dc28c60221f"
    },
    {
      "file": "logprobs.jsonl",
      "observations": 324,
      "sha256": "dff6124672a08d7966b0d4e45d08806945de3cbd7a3112726352b75361d793a7"
    },
    {
      "file": "providers-gpt-oss-120b-2026-09-01.jsonl",
      "observations": 940,
      "sha256": "635eb85e0c1aaa07f606d33135a2415dba47095b096ca5d1aac23c1a5ab1ac98"
    },
    {
      "file": "robust-standard.jsonl",
      "observations": 2538,
      "sha256": "ee1ded65380d140b766926a075553ce03b394875695cd784fabc1bbbf07d2d80"
    },
    {
      "file": "stream-smoke.jsonl",
      "observations": 90,
      "sha256": "e940f687f89dfa3c464421361226b884f35a0f5e74f2957e076afce7a267f852"
    }
  ],
  "started_at": "2026-09-01T14:16:57.236355+00:00",
  "successes": 3666,
  "tool": "inference-audit",
  "tool_revision": "01840b569152e22c6fc86950f9844a2b20db5a4a"
}