{
  "reviewedOn": "2026-10-20",
  "scope": "Bounded three-worker converged functional QA; source-built proxy-only Gateway; not release support or performance qualification",
  "gatewaySource": "1d51393773e4974b5e356377496c9d677dba9f9c",
  "topologies": {
    "vllm": "3 x L4 24GB",
    "sglang": "3 x L4 24GB",
    "trtllm": "1 x L4 24GB + 2 x L40S 48GB"
  },
  "instrumentationCode": {
    "path": "verification/receipt-collector.py",
    "usedForEngines": [
      "sglang",
      "trtllm"
    ],
    "vllmEarlierVariant": "same forwarding and byte hashes; did not retain request-body files",
    "sha256": "f9558dab44a24b3230a6f804dd549178f0533e52b87f6e4004fb524d58880326"
  },
  "instrumentation": "Independent transparent receipt collector forwards data to native engine; native health/event paths retained. Timing is not a native-direct performance comparison.",
  "counts": {
    "PASS_BOUNDED_FUNCTIONAL": 26,
    "BLOCKED_NATIVE_ARCHITECTURE": 4
  },
  "historicalGemmaLoadFailure": "open; sequential functional completion does not close prior load RED",
  "catalogPromotion": false,
  "cases": [
    {
      "caseId": "gemma2-vllm-rr",
      "modelKey": "gemma2",
      "model": "google/gemma-2-9b-it",
      "modelRevision": "11c9b309abf73637e4b6f9a3fa1e92e615547819",
      "engine": "vllm",
      "engineVersion": "0.31.0",
      "engineImageDigest": "sha256:a4a4c0437bf7240089da5f08aa370c4aee17ae5290f7a3b468825ee26c4c3a6b",
      "mode": "rr",
      "evaluationProfileDefinitionSha256": "5bd6a2391aca7a1d368282b6010acfbfcc08c9e7a3038edfbef2066eca51946a",
      "result": "PASS_BOUNDED_FUNCTIONAL",
      "rawResultSha256": "4eed8b78668fb5429e81cfa8cc1261d8ee3fe1ddc1c8c5f66f04ac8166d847dd",
      "configuration": "matrix-configurations/gemma2-vllm-rr.json",
      "remainingGates": [
        "release-image-and-host-dataplane",
        "capacity-and-application-SLO",
        "worker-loss-recovery-and-cancellation",
        "clean-replay-and-long-running-resource-return"
      ],
      "nativeNormal": 6,
      "offered": 6,
      "normal": 6,
      "workerReceipts": {
        "workerA": 2,
        "workerB": 2,
        "workerC": 2
      },
      "receiptHashAudit": "pass",
      "cacheHitDelta": 0,
      "invalidModeHttp": 400,
      "invalidMutationNoChange": true,
      "scopedRuleDeletion": "absent",
      "configuredServiceAndEndpointsUnchanged": true,
      "unknownModelRejectedWithoutDispatch": true
    },
    {
      "caseId": "gemma2-vllm-exact",
      "modelKey": "gemma2",
      "model": "google/gemma-2-9b-it",
      "modelRevision": "11c9b309abf73637e4b6f9a3fa1e92e615547819",
      "engine": "vllm",
      "engineVersion": "0.31.0",
      "engineImageDigest": "sha256:a4a4c0437bf7240089da5f08aa370c4aee17ae5290f7a3b468825ee26c4c3a6b",
      "mode": "exact",
      "evaluationProfileDefinitionSha256": "5bd6a2391aca7a1d368282b6010acfbfcc08c9e7a3038edfbef2066eca51946a",
      "result": "PASS_BOUNDED_FUNCTIONAL",
      "rawResultSha256": "4eed8b78668fb5429e81cfa8cc1261d8ee3fe1ddc1c8c5f66f04ac8166d847dd",
      "configuration": "matrix-configurations/gemma2-vllm-exact.json",
      "remainingGates": [
        "release-image-and-host-dataplane",
        "capacity-and-application-SLO",
        "worker-loss-recovery-and-cancellation",
        "clean-replay-and-long-running-resource-return"
      ],
      "nativeNormal": 6,
      "offered": 6,
      "normal": 6,
      "workerReceipts": {
        "prewarmedWorkerA": 6
      },
      "receiptHashAudit": "pass",
      "cacheHitDelta": 6.0,
      "invalidModeHttp": 400,
      "invalidMutationNoChange": true,
      "scopedRuleDeletion": "absent",
      "configuredServiceAndEndpointsUnchanged": true,
      "enforcedState": "READY",
      "reasonCodes": [],
      "goFenced": false,
      "coldNormal": 3,
      "coldReceiptHashAudit": "pass",
      "qaVerdictReview": {
        "initialHarnessResult": "FAIL",
        "originalFailurePreserved": true,
        "normalizedReadOnlyField": "fc_effective.reservation_epoch",
        "configuredMutationNegativeControl": "detected"
      }
    },
    {
      "caseId": "gemma2-sglang-rr",
      "modelKey": "gemma2",
      "model": "google/gemma-2-9b-it",
      "modelRevision": "11c9b309abf73637e4b6f9a3fa1e92e615547819",
      "engine": "sglang",
      "engineVersion": "0.5.19",
      "engineImageDigest": "sha256:37bbbd3444732a464bbc68dee4fb0164e0ce9e18e2f027f3fc967f1152d3c262",
      "mode": "rr",
      "evaluationProfileDefinitionSha256": "5bd6a2391aca7a1d368282b6010acfbfcc08c9e7a3038edfbef2066eca51946a",
      "result": "PASS_BOUNDED_FUNCTIONAL",
      "rawResultSha256": "dd4f9ed083583dc2bdaed3e746c5ef122ffda9bacfebf659cd68041e653398ce",
      "configuration": "matrix-configurations/gemma2-sglang-rr.json",
      "remainingGates": [
        "release-image-and-host-dataplane",
        "capacity-and-application-SLO",
        "worker-loss-recovery-and-cancellation",
        "clean-replay-and-long-running-resource-return"
      ],
      "nativeNormal": 6,
      "offered": 6,
      "normal": 6,
      "workerReceipts": {
        "workerA": 2,
        "workerB": 2,
        "workerC": 2
      },
      "receiptHashAudit": "pass",
      "cacheHitDelta": 0,
      "invalidModeHttp": 400,
      "invalidMutationNoChange": true,
      "scopedRuleDeletion": "absent",
      "configuredServiceAndEndpointsUnchanged": true,
      "unknownModelRejectedWithoutDispatch": true
    },
    {
      "caseId": "gemma2-sglang-exact",
      "modelKey": "gemma2",
      "model": "google/gemma-2-9b-it",
      "modelRevision": "11c9b309abf73637e4b6f9a3fa1e92e615547819",
      "engine": "sglang",
      "engineVersion": "0.5.19",
      "engineImageDigest": "sha256:37bbbd3444732a464bbc68dee4fb0164e0ce9e18e2f027f3fc967f1152d3c262",
      "mode": "exact",
      "evaluationProfileDefinitionSha256": "5bd6a2391aca7a1d368282b6010acfbfcc08c9e7a3038edfbef2066eca51946a",
      "result": "PASS_BOUNDED_FUNCTIONAL",
      "rawResultSha256": "dd4f9ed083583dc2bdaed3e746c5ef122ffda9bacfebf659cd68041e653398ce",
      "configuration": "matrix-configurations/gemma2-sglang-exact.json",
      "remainingGates": [
        "release-image-and-host-dataplane",
        "capacity-and-application-SLO",
        "worker-loss-recovery-and-cancellation",
        "clean-replay-and-long-running-resource-return"
      ],
      "nativeNormal": 6,
      "offered": 6,
      "normal": 6,
      "workerReceipts": {
        "prewarmedWorkerA": 6
      },
      "receiptHashAudit": "pass",
      "cacheHitDelta": 6.0,
      "invalidModeHttp": 400,
      "invalidMutationNoChange": true,
      "scopedRuleDeletion": "absent",
      "configuredServiceAndEndpointsUnchanged": true,
      "enforcedState": "READY",
      "reasonCodes": [],
      "goFenced": false,
      "coldNormal": 3,
      "coldReceiptHashAudit": "pass"
    },
    {
      "caseId": "gemma2-trtllm-rr",
      "modelKey": "gemma2",
      "model": "google/gemma-2-9b-it",
      "modelRevision": "11c9b309abf73637e4b6f9a3fa1e92e615547819",
      "engine": "trtllm",
      "engineVersion": "1.3.0rc29",
      "engineImageDigest": "sha256:e9dab9fccc550dcc04356962ff152908d692e0e47ba143f57402a24a4ad64b77",
      "mode": "rr",
      "evaluationProfileDefinitionSha256": "5bd6a2391aca7a1d368282b6010acfbfcc08c9e7a3038edfbef2066eca51946a",
      "result": "BLOCKED_NATIVE_ARCHITECTURE",
      "rawResultSha256": "9885dcf4fbff2247f775ac6ffceb8d9e80689e29e4942681a6125c1c475d64af",
      "configuration": "matrix-configurations/gemma2-trtllm-rr.json",
      "remainingGates": [
        "release-image-and-host-dataplane",
        "capacity-and-application-SLO",
        "worker-loss-recovery-and-cancellation",
        "clean-replay-and-long-running-resource-return"
      ],
      "architecture": "Gemma2ForCausalLM",
      "inferenceAttempted": false,
      "reason": "Stock rc29 PyTorch native executor rejects architecture before checkpoint loading",
      "retestRequirement": "Obtain a separately supported backend/image tuple, then repeat both native and Gateway modes"
    },
    {
      "caseId": "gemma2-trtllm-exact",
      "modelKey": "gemma2",
      "model": "google/gemma-2-9b-it",
      "modelRevision": "11c9b309abf73637e4b6f9a3fa1e92e615547819",
      "engine": "trtllm",
      "engineVersion": "1.3.0rc29",
      "engineImageDigest": "sha256:e9dab9fccc550dcc04356962ff152908d692e0e47ba143f57402a24a4ad64b77",
      "mode": "exact",
      "evaluationProfileDefinitionSha256": "5bd6a2391aca7a1d368282b6010acfbfcc08c9e7a3038edfbef2066eca51946a",
      "result": "BLOCKED_NATIVE_ARCHITECTURE",
      "rawResultSha256": "9885dcf4fbff2247f775ac6ffceb8d9e80689e29e4942681a6125c1c475d64af",
      "configuration": "matrix-configurations/gemma2-trtllm-exact.json",
      "remainingGates": [
        "release-image-and-host-dataplane",
        "capacity-and-application-SLO",
        "worker-loss-recovery-and-cancellation",
        "clean-replay-and-long-running-resource-return"
      ],
      "architecture": "Gemma2ForCausalLM",
      "inferenceAttempted": false,
      "reason": "Stock rc29 PyTorch native executor rejects architecture before checkpoint loading",
      "retestRequirement": "Obtain a separately supported backend/image tuple, then repeat both native and Gateway modes"
    },
    {
      "caseId": "qwen25-vllm-rr",
      "modelKey": "qwen25",
      "model": "Qwen/Qwen2.5-7B-Instruct",
      "modelRevision": "a09a35458c702b33eeacc393d103063234e8bc28",
      "engine": "vllm",
      "engineVersion": "0.31.0",
      "engineImageDigest": "sha256:a4a4c0437bf7240089da5f08aa370c4aee17ae5290f7a3b468825ee26c4c3a6b",
      "mode": "rr",
      "evaluationProfileDefinitionSha256": "63dbff121176520773005a84c05fde2dd94d23a246593e1fbb28783170de4bbd",
      "result": "PASS_BOUNDED_FUNCTIONAL",
      "rawResultSha256": "d553e96fd5e258e2e843506af3f2d11d07630289c9d7297b1ec5ef0fbc690f27",
      "configuration": "matrix-configurations/qwen25-vllm-rr.json",
      "remainingGates": [
        "release-image-and-host-dataplane",
        "capacity-and-application-SLO",
        "worker-loss-recovery-and-cancellation",
        "clean-replay-and-long-running-resource-return"
      ],
      "nativeNormal": 6,
      "offered": 6,
      "normal": 6,
      "workerReceipts": {
        "workerA": 2,
        "workerB": 2,
        "workerC": 2
      },
      "receiptHashAudit": "pass",
      "cacheHitDelta": 0,
      "invalidModeHttp": 400,
      "invalidMutationNoChange": true,
      "scopedRuleDeletion": "absent",
      "configuredServiceAndEndpointsUnchanged": true,
      "unknownModelRejectedWithoutDispatch": true
    },
    {
      "caseId": "qwen25-vllm-exact",
      "modelKey": "qwen25",
      "model": "Qwen/Qwen2.5-7B-Instruct",
      "modelRevision": "a09a35458c702b33eeacc393d103063234e8bc28",
      "engine": "vllm",
      "engineVersion": "0.31.0",
      "engineImageDigest": "sha256:a4a4c0437bf7240089da5f08aa370c4aee17ae5290f7a3b468825ee26c4c3a6b",
      "mode": "exact",
      "evaluationProfileDefinitionSha256": "63dbff121176520773005a84c05fde2dd94d23a246593e1fbb28783170de4bbd",
      "result": "PASS_BOUNDED_FUNCTIONAL",
      "rawResultSha256": "d553e96fd5e258e2e843506af3f2d11d07630289c9d7297b1ec5ef0fbc690f27",
      "configuration": "matrix-configurations/qwen25-vllm-exact.json",
      "remainingGates": [
        "release-image-and-host-dataplane",
        "capacity-and-application-SLO",
        "worker-loss-recovery-and-cancellation",
        "clean-replay-and-long-running-resource-return"
      ],
      "nativeNormal": 6,
      "offered": 6,
      "normal": 6,
      "workerReceipts": {
        "prewarmedWorkerA": 6
      },
      "receiptHashAudit": "pass",
      "cacheHitDelta": 6.0,
      "invalidModeHttp": 400,
      "invalidMutationNoChange": true,
      "scopedRuleDeletion": "absent",
      "configuredServiceAndEndpointsUnchanged": true,
      "enforcedState": "READY",
      "reasonCodes": [],
      "goFenced": false,
      "coldNormal": 3,
      "coldReceiptHashAudit": "pass"
    },
    {
      "caseId": "qwen25-sglang-rr",
      "modelKey": "qwen25",
      "model": "Qwen/Qwen2.5-7B-Instruct",
      "modelRevision": "a09a35458c702b33eeacc393d103063234e8bc28",
      "engine": "sglang",
      "engineVersion": "0.5.19",
      "engineImageDigest": "sha256:37bbbd3444732a464bbc68dee4fb0164e0ce9e18e2f027f3fc967f1152d3c262",
      "mode": "rr",
      "evaluationProfileDefinitionSha256": "63dbff121176520773005a84c05fde2dd94d23a246593e1fbb28783170de4bbd",
      "result": "PASS_BOUNDED_FUNCTIONAL",
      "rawResultSha256": "65448b3271e115731ec034095276a45b843ca40ff88c87c8211e615579aa4362",
      "configuration": "matrix-configurations/qwen25-sglang-rr.json",
      "remainingGates": [
        "release-image-and-host-dataplane",
        "capacity-and-application-SLO",
        "worker-loss-recovery-and-cancellation",
        "clean-replay-and-long-running-resource-return"
      ],
      "nativeNormal": 6,
      "offered": 6,
      "normal": 6,
      "workerReceipts": {
        "workerA": 2,
        "workerB": 2,
        "workerC": 2
      },
      "receiptHashAudit": "pass",
      "cacheHitDelta": 0,
      "invalidModeHttp": 400,
      "invalidMutationNoChange": true,
      "scopedRuleDeletion": "absent",
      "configuredServiceAndEndpointsUnchanged": true,
      "unknownModelRejectedWithoutDispatch": true
    },
    {
      "caseId": "qwen25-sglang-exact",
      "modelKey": "qwen25",
      "model": "Qwen/Qwen2.5-7B-Instruct",
      "modelRevision": "a09a35458c702b33eeacc393d103063234e8bc28",
      "engine": "sglang",
      "engineVersion": "0.5.19",
      "engineImageDigest": "sha256:37bbbd3444732a464bbc68dee4fb0164e0ce9e18e2f027f3fc967f1152d3c262",
      "mode": "exact",
      "evaluationProfileDefinitionSha256": "63dbff121176520773005a84c05fde2dd94d23a246593e1fbb28783170de4bbd",
      "result": "PASS_BOUNDED_FUNCTIONAL",
      "rawResultSha256": "65448b3271e115731ec034095276a45b843ca40ff88c87c8211e615579aa4362",
      "configuration": "matrix-configurations/qwen25-sglang-exact.json",
      "remainingGates": [
        "release-image-and-host-dataplane",
        "capacity-and-application-SLO",
        "worker-loss-recovery-and-cancellation",
        "clean-replay-and-long-running-resource-return"
      ],
      "nativeNormal": 6,
      "offered": 6,
      "normal": 6,
      "workerReceipts": {
        "prewarmedWorkerA": 6
      },
      "receiptHashAudit": "pass",
      "cacheHitDelta": 6.0,
      "invalidModeHttp": 400,
      "invalidMutationNoChange": true,
      "scopedRuleDeletion": "absent",
      "configuredServiceAndEndpointsUnchanged": true,
      "enforcedState": "READY",
      "reasonCodes": [],
      "goFenced": false,
      "coldNormal": 3,
      "coldReceiptHashAudit": "pass"
    },
    {
      "caseId": "qwen25-trtllm-rr",
      "modelKey": "qwen25",
      "model": "Qwen/Qwen2.5-7B-Instruct",
      "modelRevision": "a09a35458c702b33eeacc393d103063234e8bc28",
      "engine": "trtllm",
      "engineVersion": "1.3.0rc29",
      "engineImageDigest": "sha256:e9dab9fccc550dcc04356962ff152908d692e0e47ba143f57402a24a4ad64b77",
      "mode": "rr",
      "evaluationProfileDefinitionSha256": "63dbff121176520773005a84c05fde2dd94d23a246593e1fbb28783170de4bbd",
      "result": "PASS_BOUNDED_FUNCTIONAL",
      "rawResultSha256": "52100e2d11a3012f961186ab76fb08e1de35c9f37730073144cbad1537905c1e",
      "configuration": "matrix-configurations/qwen25-trtllm-rr.json",
      "remainingGates": [
        "release-image-and-host-dataplane",
        "capacity-and-application-SLO",
        "worker-loss-recovery-and-cancellation",
        "clean-replay-and-long-running-resource-return"
      ],
      "nativeNormal": 6,
      "offered": 6,
      "normal": 6,
      "workerReceipts": {
        "workerA": 2,
        "workerB": 2,
        "workerC": 2
      },
      "receiptHashAudit": "pass",
      "cacheHitDelta": 0,
      "invalidModeHttp": 400,
      "invalidMutationNoChange": true,
      "scopedRuleDeletion": "absent",
      "configuredServiceAndEndpointsUnchanged": true,
      "unknownModelRejectedWithoutDispatch": true
    },
    {
      "caseId": "qwen25-trtllm-exact",
      "modelKey": "qwen25",
      "model": "Qwen/Qwen2.5-7B-Instruct",
      "modelRevision": "a09a35458c702b33eeacc393d103063234e8bc28",
      "engine": "trtllm",
      "engineVersion": "1.3.0rc29",
      "engineImageDigest": "sha256:e9dab9fccc550dcc04356962ff152908d692e0e47ba143f57402a24a4ad64b77",
      "mode": "exact",
      "evaluationProfileDefinitionSha256": "63dbff121176520773005a84c05fde2dd94d23a246593e1fbb28783170de4bbd",
      "result": "PASS_BOUNDED_FUNCTIONAL",
      "rawResultSha256": "52100e2d11a3012f961186ab76fb08e1de35c9f37730073144cbad1537905c1e",
      "configuration": "matrix-configurations/qwen25-trtllm-exact.json",
      "remainingGates": [
        "release-image-and-host-dataplane",
        "capacity-and-application-SLO",
        "worker-loss-recovery-and-cancellation",
        "clean-replay-and-long-running-resource-return"
      ],
      "nativeNormal": 6,
      "offered": 6,
      "normal": 6,
      "workerReceipts": {
        "prewarmedWorkerA": 6
      },
      "receiptHashAudit": "pass",
      "cacheHitDelta": 6.0,
      "invalidModeHttp": 400,
      "invalidMutationNoChange": true,
      "scopedRuleDeletion": "absent",
      "configuredServiceAndEndpointsUnchanged": true,
      "enforcedState": "READY",
      "reasonCodes": [],
      "goFenced": false,
      "coldNormal": 3,
      "coldReceiptHashAudit": "pass"
    },
    {
      "caseId": "llama31-vllm-rr",
      "modelKey": "llama31",
      "model": "meta-llama/Llama-3.1-8B-Instruct",
      "modelRevision": "07eb05b21d191a58c577b4a45982fe0c049d0693",
      "engine": "vllm",
      "engineVersion": "0.31.0",
      "engineImageDigest": "sha256:a4a4c0437bf7240089da5f08aa370c4aee17ae5290f7a3b468825ee26c4c3a6b",
      "mode": "rr",
      "evaluationProfileDefinitionSha256": "e17366700032ab8cfc140c8f245d924b8312ab7746228b1fc0e69cf186ef0a9c",
      "result": "PASS_BOUNDED_FUNCTIONAL",
      "rawResultSha256": "a99cb062fad62a02e294a60b888ec845d6c932c105c3b7bec33d59ae2ead4bea",
      "configuration": "matrix-configurations/llama31-vllm-rr.json",
      "remainingGates": [
        "release-image-and-host-dataplane",
        "capacity-and-application-SLO",
        "worker-loss-recovery-and-cancellation",
        "clean-replay-and-long-running-resource-return"
      ],
      "nativeNormal": 6,
      "offered": 6,
      "normal": 6,
      "workerReceipts": {
        "workerA": 2,
        "workerB": 2,
        "workerC": 2
      },
      "receiptHashAudit": "pass",
      "cacheHitDelta": 0,
      "invalidModeHttp": 400,
      "invalidMutationNoChange": true,
      "scopedRuleDeletion": "absent",
      "configuredServiceAndEndpointsUnchanged": true,
      "unknownModelRejectedWithoutDispatch": true,
      "qaVerdictReview": {
        "initialHarnessResult": "FAIL",
        "originalFailurePreserved": true,
        "normalizedReadOnlyField": "fc_effective.reservation_epoch",
        "configuredMutationNegativeControl": "detected"
      }
    },
    {
      "caseId": "llama31-vllm-exact",
      "modelKey": "llama31",
      "model": "meta-llama/Llama-3.1-8B-Instruct",
      "modelRevision": "07eb05b21d191a58c577b4a45982fe0c049d0693",
      "engine": "vllm",
      "engineVersion": "0.31.0",
      "engineImageDigest": "sha256:a4a4c0437bf7240089da5f08aa370c4aee17ae5290f7a3b468825ee26c4c3a6b",
      "mode": "exact",
      "evaluationProfileDefinitionSha256": "e17366700032ab8cfc140c8f245d924b8312ab7746228b1fc0e69cf186ef0a9c",
      "result": "PASS_BOUNDED_FUNCTIONAL",
      "rawResultSha256": "a99cb062fad62a02e294a60b888ec845d6c932c105c3b7bec33d59ae2ead4bea",
      "configuration": "matrix-configurations/llama31-vllm-exact.json",
      "remainingGates": [
        "release-image-and-host-dataplane",
        "capacity-and-application-SLO",
        "worker-loss-recovery-and-cancellation",
        "clean-replay-and-long-running-resource-return"
      ],
      "nativeNormal": 6,
      "offered": 6,
      "normal": 6,
      "workerReceipts": {
        "prewarmedWorkerA": 6
      },
      "receiptHashAudit": "pass",
      "cacheHitDelta": 6.0,
      "invalidModeHttp": 400,
      "invalidMutationNoChange": true,
      "scopedRuleDeletion": "absent",
      "configuredServiceAndEndpointsUnchanged": true,
      "enforcedState": "READY",
      "reasonCodes": [],
      "goFenced": false,
      "coldNormal": 3,
      "coldReceiptHashAudit": "pass"
    },
    {
      "caseId": "llama31-sglang-rr",
      "modelKey": "llama31",
      "model": "meta-llama/Llama-3.1-8B-Instruct",
      "modelRevision": "07eb05b21d191a58c577b4a45982fe0c049d0693",
      "engine": "sglang",
      "engineVersion": "0.5.19",
      "engineImageDigest": "sha256:37bbbd3444732a464bbc68dee4fb0164e0ce9e18e2f027f3fc967f1152d3c262",
      "mode": "rr",
      "evaluationProfileDefinitionSha256": "e17366700032ab8cfc140c8f245d924b8312ab7746228b1fc0e69cf186ef0a9c",
      "result": "PASS_BOUNDED_FUNCTIONAL",
      "rawResultSha256": "13f0d55ade25d77c09b78373567a7d6f400d6b2bf167f4dd150f401a8be4db07",
      "configuration": "matrix-configurations/llama31-sglang-rr.json",
      "remainingGates": [
        "release-image-and-host-dataplane",
        "capacity-and-application-SLO",
        "worker-loss-recovery-and-cancellation",
        "clean-replay-and-long-running-resource-return"
      ],
      "nativeNormal": 6,
      "offered": 6,
      "normal": 6,
      "workerReceipts": {
        "workerA": 2,
        "workerB": 2,
        "workerC": 2
      },
      "receiptHashAudit": "pass",
      "cacheHitDelta": 0,
      "invalidModeHttp": 400,
      "invalidMutationNoChange": true,
      "scopedRuleDeletion": "absent",
      "configuredServiceAndEndpointsUnchanged": true,
      "unknownModelRejectedWithoutDispatch": true
    },
    {
      "caseId": "llama31-sglang-exact",
      "modelKey": "llama31",
      "model": "meta-llama/Llama-3.1-8B-Instruct",
      "modelRevision": "07eb05b21d191a58c577b4a45982fe0c049d0693",
      "engine": "sglang",
      "engineVersion": "0.5.19",
      "engineImageDigest": "sha256:37bbbd3444732a464bbc68dee4fb0164e0ce9e18e2f027f3fc967f1152d3c262",
      "mode": "exact",
      "evaluationProfileDefinitionSha256": "e17366700032ab8cfc140c8f245d924b8312ab7746228b1fc0e69cf186ef0a9c",
      "result": "PASS_BOUNDED_FUNCTIONAL",
      "rawResultSha256": "13f0d55ade25d77c09b78373567a7d6f400d6b2bf167f4dd150f401a8be4db07",
      "configuration": "matrix-configurations/llama31-sglang-exact.json",
      "remainingGates": [
        "release-image-and-host-dataplane",
        "capacity-and-application-SLO",
        "worker-loss-recovery-and-cancellation",
        "clean-replay-and-long-running-resource-return"
      ],
      "nativeNormal": 6,
      "offered": 6,
      "normal": 6,
      "workerReceipts": {
        "prewarmedWorkerA": 6
      },
      "receiptHashAudit": "pass",
      "cacheHitDelta": 6.0,
      "invalidModeHttp": 400,
      "invalidMutationNoChange": true,
      "scopedRuleDeletion": "absent",
      "configuredServiceAndEndpointsUnchanged": true,
      "enforcedState": "READY",
      "reasonCodes": [],
      "goFenced": false,
      "coldNormal": 3,
      "coldReceiptHashAudit": "pass"
    },
    {
      "caseId": "llama31-trtllm-rr",
      "modelKey": "llama31",
      "model": "meta-llama/Llama-3.1-8B-Instruct",
      "modelRevision": "07eb05b21d191a58c577b4a45982fe0c049d0693",
      "engine": "trtllm",
      "engineVersion": "1.3.0rc29",
      "engineImageDigest": "sha256:e9dab9fccc550dcc04356962ff152908d692e0e47ba143f57402a24a4ad64b77",
      "mode": "rr",
      "evaluationProfileDefinitionSha256": "e17366700032ab8cfc140c8f245d924b8312ab7746228b1fc0e69cf186ef0a9c",
      "result": "PASS_BOUNDED_FUNCTIONAL",
      "rawResultSha256": "5a7eca6a22e6722869928e70e341c7779fd2b67ecc2508cbd236554210b558d9",
      "configuration": "matrix-configurations/llama31-trtllm-rr.json",
      "remainingGates": [
        "release-image-and-host-dataplane",
        "capacity-and-application-SLO",
        "worker-loss-recovery-and-cancellation",
        "clean-replay-and-long-running-resource-return"
      ],
      "nativeNormal": 6,
      "offered": 6,
      "normal": 6,
      "workerReceipts": {
        "workerA": 2,
        "workerB": 2,
        "workerC": 2
      },
      "receiptHashAudit": "pass",
      "cacheHitDelta": 0,
      "invalidModeHttp": 400,
      "invalidMutationNoChange": true,
      "scopedRuleDeletion": "absent",
      "configuredServiceAndEndpointsUnchanged": true,
      "unknownModelRejectedWithoutDispatch": true
    },
    {
      "caseId": "llama31-trtllm-exact",
      "modelKey": "llama31",
      "model": "meta-llama/Llama-3.1-8B-Instruct",
      "modelRevision": "07eb05b21d191a58c577b4a45982fe0c049d0693",
      "engine": "trtllm",
      "engineVersion": "1.3.0rc29",
      "engineImageDigest": "sha256:e9dab9fccc550dcc04356962ff152908d692e0e47ba143f57402a24a4ad64b77",
      "mode": "exact",
      "evaluationProfileDefinitionSha256": "e17366700032ab8cfc140c8f245d924b8312ab7746228b1fc0e69cf186ef0a9c",
      "result": "PASS_BOUNDED_FUNCTIONAL",
      "rawResultSha256": "5a7eca6a22e6722869928e70e341c7779fd2b67ecc2508cbd236554210b558d9",
      "configuration": "matrix-configurations/llama31-trtllm-exact.json",
      "remainingGates": [
        "release-image-and-host-dataplane",
        "capacity-and-application-SLO",
        "worker-loss-recovery-and-cancellation",
        "clean-replay-and-long-running-resource-return"
      ],
      "nativeNormal": 6,
      "offered": 6,
      "normal": 6,
      "workerReceipts": {
        "prewarmedWorkerA": 6
      },
      "receiptHashAudit": "pass",
      "cacheHitDelta": 6.0,
      "invalidModeHttp": 400,
      "invalidMutationNoChange": true,
      "scopedRuleDeletion": "absent",
      "configuredServiceAndEndpointsUnchanged": true,
      "enforcedState": "READY",
      "reasonCodes": [],
      "goFenced": false,
      "coldNormal": 3,
      "coldReceiptHashAudit": "pass"
    },
    {
      "caseId": "r1-vllm-rr",
      "modelKey": "r1",
      "model": "deepseek-ai/DeepSeek-R1-Distill-Qwen-7B",
      "modelRevision": "916b56a44061fd5cd7d6a8fb632557ed4f724f60",
      "engine": "vllm",
      "engineVersion": "0.31.0",
      "engineImageDigest": "sha256:a4a4c0437bf7240089da5f08aa370c4aee17ae5290f7a3b468825ee26c4c3a6b",
      "mode": "rr",
      "evaluationProfileDefinitionSha256": "2802268c7d61580abd9c5e583b799301689b5c55f024d7a3fb29ac98461390d1",
      "result": "PASS_BOUNDED_FUNCTIONAL",
      "rawResultSha256": "5b1391529feacec69260e628cd89325b780abba79dfb41d877e0bf00dc04b4eb",
      "configuration": "matrix-configurations/r1-vllm-rr.json",
      "remainingGates": [
        "release-image-and-host-dataplane",
        "capacity-and-application-SLO",
        "worker-loss-recovery-and-cancellation",
        "clean-replay-and-long-running-resource-return"
      ],
      "nativeNormal": 6,
      "offered": 6,
      "normal": 6,
      "workerReceipts": {
        "workerA": 2,
        "workerB": 2,
        "workerC": 2
      },
      "receiptHashAudit": "pass",
      "cacheHitDelta": 0,
      "invalidModeHttp": 400,
      "invalidMutationNoChange": true,
      "scopedRuleDeletion": "absent",
      "configuredServiceAndEndpointsUnchanged": true,
      "unknownModelRejectedWithoutDispatch": true
    },
    {
      "caseId": "r1-vllm-exact",
      "modelKey": "r1",
      "model": "deepseek-ai/DeepSeek-R1-Distill-Qwen-7B",
      "modelRevision": "916b56a44061fd5cd7d6a8fb632557ed4f724f60",
      "engine": "vllm",
      "engineVersion": "0.31.0",
      "engineImageDigest": "sha256:a4a4c0437bf7240089da5f08aa370c4aee17ae5290f7a3b468825ee26c4c3a6b",
      "mode": "exact",
      "evaluationProfileDefinitionSha256": "2802268c7d61580abd9c5e583b799301689b5c55f024d7a3fb29ac98461390d1",
      "result": "PASS_BOUNDED_FUNCTIONAL",
      "rawResultSha256": "5b1391529feacec69260e628cd89325b780abba79dfb41d877e0bf00dc04b4eb",
      "configuration": "matrix-configurations/r1-vllm-exact.json",
      "remainingGates": [
        "release-image-and-host-dataplane",
        "capacity-and-application-SLO",
        "worker-loss-recovery-and-cancellation",
        "clean-replay-and-long-running-resource-return"
      ],
      "nativeNormal": 6,
      "offered": 6,
      "normal": 6,
      "workerReceipts": {
        "prewarmedWorkerA": 6
      },
      "receiptHashAudit": "pass",
      "cacheHitDelta": 6.0,
      "invalidModeHttp": 400,
      "invalidMutationNoChange": true,
      "scopedRuleDeletion": "absent",
      "configuredServiceAndEndpointsUnchanged": true,
      "enforcedState": "READY",
      "reasonCodes": [],
      "goFenced": false,
      "coldNormal": 3,
      "coldReceiptHashAudit": "pass"
    },
    {
      "caseId": "r1-sglang-rr",
      "modelKey": "r1",
      "model": "deepseek-ai/DeepSeek-R1-Distill-Qwen-7B",
      "modelRevision": "916b56a44061fd5cd7d6a8fb632557ed4f724f60",
      "engine": "sglang",
      "engineVersion": "0.5.19",
      "engineImageDigest": "sha256:37bbbd3444732a464bbc68dee4fb0164e0ce9e18e2f027f3fc967f1152d3c262",
      "mode": "rr",
      "evaluationProfileDefinitionSha256": "0e23bb89c028ec564440cf245966a5bddc31c5b5379413823ac90338270e6022",
      "result": "PASS_BOUNDED_FUNCTIONAL",
      "rawResultSha256": "8fed087cb63e4279d3fd84f7732e44d90573a43075783f73adee32063657267d",
      "configuration": "matrix-configurations/r1-sglang-rr.json",
      "remainingGates": [
        "release-image-and-host-dataplane",
        "capacity-and-application-SLO",
        "worker-loss-recovery-and-cancellation",
        "clean-replay-and-long-running-resource-return"
      ],
      "nativeNormal": 6,
      "offered": 6,
      "normal": 6,
      "workerReceipts": {
        "workerA": 2,
        "workerB": 2,
        "workerC": 2
      },
      "receiptHashAudit": "pass",
      "cacheHitDelta": 0,
      "invalidModeHttp": 400,
      "invalidMutationNoChange": true,
      "scopedRuleDeletion": "absent",
      "configuredServiceAndEndpointsUnchanged": true,
      "unknownModelRejectedWithoutDispatch": true,
      "profileInputReview": {
        "initialExactResult": "FAIL_ATTESTATION",
        "initialReason": "challenge_timeout",
        "initialEngineVersion": "0.5.18",
        "changedInputs": [
          "stock SGLang0.5.19",
          "SGLang-specific completionsBos=true metadata"
        ],
        "officialTokenizerAndTemplateUnchanged": true,
        "bothModesReplayed": true,
        "nativeOracle": {
          "rawPromptTokens": 37,
          "nativePromptTokens": 38,
          "bosTokenId": 151646,
          "nativeHelperMatchedExplicitBOSPlusRawTokens": true
        }
      }
    },
    {
      "caseId": "r1-sglang-exact",
      "modelKey": "r1",
      "model": "deepseek-ai/DeepSeek-R1-Distill-Qwen-7B",
      "modelRevision": "916b56a44061fd5cd7d6a8fb632557ed4f724f60",
      "engine": "sglang",
      "engineVersion": "0.5.19",
      "engineImageDigest": "sha256:37bbbd3444732a464bbc68dee4fb0164e0ce9e18e2f027f3fc967f1152d3c262",
      "mode": "exact",
      "evaluationProfileDefinitionSha256": "0e23bb89c028ec564440cf245966a5bddc31c5b5379413823ac90338270e6022",
      "result": "PASS_BOUNDED_FUNCTIONAL",
      "rawResultSha256": "8fed087cb63e4279d3fd84f7732e44d90573a43075783f73adee32063657267d",
      "configuration": "matrix-configurations/r1-sglang-exact.json",
      "remainingGates": [
        "release-image-and-host-dataplane",
        "capacity-and-application-SLO",
        "worker-loss-recovery-and-cancellation",
        "clean-replay-and-long-running-resource-return"
      ],
      "nativeNormal": 6,
      "offered": 6,
      "normal": 6,
      "workerReceipts": {
        "prewarmedWorkerA": 6
      },
      "receiptHashAudit": "pass",
      "cacheHitDelta": 6.0,
      "invalidModeHttp": 400,
      "invalidMutationNoChange": true,
      "scopedRuleDeletion": "absent",
      "configuredServiceAndEndpointsUnchanged": true,
      "enforcedState": "READY",
      "reasonCodes": [],
      "goFenced": false,
      "coldNormal": 3,
      "coldReceiptHashAudit": "pass",
      "profileInputReview": {
        "initialExactResult": "FAIL_ATTESTATION",
        "initialReason": "challenge_timeout",
        "initialEngineVersion": "0.5.18",
        "changedInputs": [
          "stock SGLang0.5.19",
          "SGLang-specific completionsBos=true metadata"
        ],
        "officialTokenizerAndTemplateUnchanged": true,
        "bothModesReplayed": true,
        "nativeOracle": {
          "rawPromptTokens": 37,
          "nativePromptTokens": 38,
          "bosTokenId": 151646,
          "nativeHelperMatchedExplicitBOSPlusRawTokens": true
        }
      }
    },
    {
      "caseId": "r1-trtllm-rr",
      "modelKey": "r1",
      "model": "deepseek-ai/DeepSeek-R1-Distill-Qwen-7B",
      "modelRevision": "916b56a44061fd5cd7d6a8fb632557ed4f724f60",
      "engine": "trtllm",
      "engineVersion": "1.3.0rc29",
      "engineImageDigest": "sha256:e9dab9fccc550dcc04356962ff152908d692e0e47ba143f57402a24a4ad64b77",
      "mode": "rr",
      "evaluationProfileDefinitionSha256": "2802268c7d61580abd9c5e583b799301689b5c55f024d7a3fb29ac98461390d1",
      "result": "PASS_BOUNDED_FUNCTIONAL",
      "rawResultSha256": "e9536524c849153538ea84261c8c43d6990ebe3e3ff077254f7ad94d845259ae",
      "configuration": "matrix-configurations/r1-trtllm-rr.json",
      "remainingGates": [
        "release-image-and-host-dataplane",
        "capacity-and-application-SLO",
        "worker-loss-recovery-and-cancellation",
        "clean-replay-and-long-running-resource-return"
      ],
      "nativeNormal": 6,
      "offered": 6,
      "normal": 6,
      "workerReceipts": {
        "workerA": 2,
        "workerB": 2,
        "workerC": 2
      },
      "receiptHashAudit": "pass",
      "cacheHitDelta": 0,
      "invalidModeHttp": 400,
      "invalidMutationNoChange": true,
      "scopedRuleDeletion": "absent",
      "configuredServiceAndEndpointsUnchanged": true,
      "unknownModelRejectedWithoutDispatch": true
    },
    {
      "caseId": "r1-trtllm-exact",
      "modelKey": "r1",
      "model": "deepseek-ai/DeepSeek-R1-Distill-Qwen-7B",
      "modelRevision": "916b56a44061fd5cd7d6a8fb632557ed4f724f60",
      "engine": "trtllm",
      "engineVersion": "1.3.0rc29",
      "engineImageDigest": "sha256:e9dab9fccc550dcc04356962ff152908d692e0e47ba143f57402a24a4ad64b77",
      "mode": "exact",
      "evaluationProfileDefinitionSha256": "2802268c7d61580abd9c5e583b799301689b5c55f024d7a3fb29ac98461390d1",
      "result": "PASS_BOUNDED_FUNCTIONAL",
      "rawResultSha256": "e9536524c849153538ea84261c8c43d6990ebe3e3ff077254f7ad94d845259ae",
      "configuration": "matrix-configurations/r1-trtllm-exact.json",
      "remainingGates": [
        "release-image-and-host-dataplane",
        "capacity-and-application-SLO",
        "worker-loss-recovery-and-cancellation",
        "clean-replay-and-long-running-resource-return"
      ],
      "nativeNormal": 6,
      "offered": 6,
      "normal": 6,
      "workerReceipts": {
        "prewarmedWorkerA": 6
      },
      "receiptHashAudit": "pass",
      "cacheHitDelta": 6.0,
      "invalidModeHttp": 400,
      "invalidMutationNoChange": true,
      "scopedRuleDeletion": "absent",
      "configuredServiceAndEndpointsUnchanged": true,
      "enforcedState": "READY",
      "reasonCodes": [],
      "goFenced": false,
      "coldNormal": 3,
      "coldReceiptHashAudit": "pass"
    },
    {
      "caseId": "exaone35-vllm-rr",
      "modelKey": "exaone35",
      "model": "LGAI-EXAONE/EXAONE-3.5-7.8B-Instruct",
      "modelRevision": "553ea250b9a5317231459279d5847d6cf955b9aa",
      "engine": "vllm",
      "engineVersion": "0.31.0",
      "engineImageDigest": "sha256:a4a4c0437bf7240089da5f08aa370c4aee17ae5290f7a3b468825ee26c4c3a6b",
      "mode": "rr",
      "evaluationProfileDefinitionSha256": "71c7ba2c29d91f5b9ab076707b4df89fae713148d6394b3dd22cb0b5a3cd86b8",
      "result": "PASS_BOUNDED_FUNCTIONAL",
      "rawResultSha256": "bf93f14c2e463bd5f87a8ac35fd4c56c7b08f481336fced497c46efe6c2d54ff",
      "configuration": "matrix-configurations/exaone35-vllm-rr.json",
      "remainingGates": [
        "release-image-and-host-dataplane",
        "capacity-and-application-SLO",
        "worker-loss-recovery-and-cancellation",
        "clean-replay-and-long-running-resource-return"
      ],
      "nativeNormal": 6,
      "offered": 6,
      "normal": 6,
      "workerReceipts": {
        "workerA": 2,
        "workerB": 2,
        "workerC": 2
      },
      "receiptHashAudit": "pass",
      "cacheHitDelta": 0,
      "invalidModeHttp": 400,
      "invalidMutationNoChange": true,
      "scopedRuleDeletion": "absent",
      "configuredServiceAndEndpointsUnchanged": true,
      "unknownModelRejectedWithoutDispatch": true
    },
    {
      "caseId": "exaone35-vllm-exact",
      "modelKey": "exaone35",
      "model": "LGAI-EXAONE/EXAONE-3.5-7.8B-Instruct",
      "modelRevision": "553ea250b9a5317231459279d5847d6cf955b9aa",
      "engine": "vllm",
      "engineVersion": "0.31.0",
      "engineImageDigest": "sha256:a4a4c0437bf7240089da5f08aa370c4aee17ae5290f7a3b468825ee26c4c3a6b",
      "mode": "exact",
      "evaluationProfileDefinitionSha256": "71c7ba2c29d91f5b9ab076707b4df89fae713148d6394b3dd22cb0b5a3cd86b8",
      "result": "PASS_BOUNDED_FUNCTIONAL",
      "rawResultSha256": "bf93f14c2e463bd5f87a8ac35fd4c56c7b08f481336fced497c46efe6c2d54ff",
      "configuration": "matrix-configurations/exaone35-vllm-exact.json",
      "remainingGates": [
        "release-image-and-host-dataplane",
        "capacity-and-application-SLO",
        "worker-loss-recovery-and-cancellation",
        "clean-replay-and-long-running-resource-return"
      ],
      "nativeNormal": 6,
      "offered": 6,
      "normal": 6,
      "workerReceipts": {
        "prewarmedWorkerA": 6
      },
      "receiptHashAudit": "pass",
      "cacheHitDelta": 6.0,
      "invalidModeHttp": 400,
      "invalidMutationNoChange": true,
      "scopedRuleDeletion": "absent",
      "configuredServiceAndEndpointsUnchanged": true,
      "enforcedState": "READY",
      "reasonCodes": [],
      "goFenced": false,
      "coldNormal": 3,
      "coldReceiptHashAudit": "pass"
    },
    {
      "caseId": "exaone35-sglang-rr",
      "modelKey": "exaone35",
      "model": "LGAI-EXAONE/EXAONE-3.5-7.8B-Instruct",
      "modelRevision": "553ea250b9a5317231459279d5847d6cf955b9aa",
      "engine": "sglang",
      "engineVersion": "0.5.19",
      "engineImageDigest": "sha256:37bbbd3444732a464bbc68dee4fb0164e0ce9e18e2f027f3fc967f1152d3c262",
      "mode": "rr",
      "evaluationProfileDefinitionSha256": "71c7ba2c29d91f5b9ab076707b4df89fae713148d6394b3dd22cb0b5a3cd86b8",
      "result": "PASS_BOUNDED_FUNCTIONAL",
      "rawResultSha256": "27963b41ed9244f313044d2256068ef42ca5aa32fcac5a1bdeec7724c4c734ea",
      "configuration": "matrix-configurations/exaone35-sglang-rr.json",
      "remainingGates": [
        "release-image-and-host-dataplane",
        "capacity-and-application-SLO",
        "worker-loss-recovery-and-cancellation",
        "clean-replay-and-long-running-resource-return"
      ],
      "nativeNormal": 6,
      "offered": 6,
      "normal": 6,
      "workerReceipts": {
        "workerA": 2,
        "workerB": 2,
        "workerC": 2
      },
      "receiptHashAudit": "pass",
      "cacheHitDelta": 0,
      "invalidModeHttp": 400,
      "invalidMutationNoChange": true,
      "scopedRuleDeletion": "absent",
      "configuredServiceAndEndpointsUnchanged": true,
      "unknownModelRejectedWithoutDispatch": true
    },
    {
      "caseId": "exaone35-sglang-exact",
      "modelKey": "exaone35",
      "model": "LGAI-EXAONE/EXAONE-3.5-7.8B-Instruct",
      "modelRevision": "553ea250b9a5317231459279d5847d6cf955b9aa",
      "engine": "sglang",
      "engineVersion": "0.5.19",
      "engineImageDigest": "sha256:37bbbd3444732a464bbc68dee4fb0164e0ce9e18e2f027f3fc967f1152d3c262",
      "mode": "exact",
      "evaluationProfileDefinitionSha256": "71c7ba2c29d91f5b9ab076707b4df89fae713148d6394b3dd22cb0b5a3cd86b8",
      "result": "PASS_BOUNDED_FUNCTIONAL",
      "rawResultSha256": "27963b41ed9244f313044d2256068ef42ca5aa32fcac5a1bdeec7724c4c734ea",
      "configuration": "matrix-configurations/exaone35-sglang-exact.json",
      "remainingGates": [
        "release-image-and-host-dataplane",
        "capacity-and-application-SLO",
        "worker-loss-recovery-and-cancellation",
        "clean-replay-and-long-running-resource-return"
      ],
      "nativeNormal": 6,
      "offered": 6,
      "normal": 6,
      "workerReceipts": {
        "prewarmedWorkerA": 6
      },
      "receiptHashAudit": "pass",
      "cacheHitDelta": 6.0,
      "invalidModeHttp": 400,
      "invalidMutationNoChange": true,
      "scopedRuleDeletion": "absent",
      "configuredServiceAndEndpointsUnchanged": true,
      "enforcedState": "READY",
      "reasonCodes": [],
      "goFenced": false,
      "coldNormal": 3,
      "coldReceiptHashAudit": "pass"
    },
    {
      "caseId": "exaone35-trtllm-rr",
      "modelKey": "exaone35",
      "model": "LGAI-EXAONE/EXAONE-3.5-7.8B-Instruct",
      "modelRevision": "553ea250b9a5317231459279d5847d6cf955b9aa",
      "engine": "trtllm",
      "engineVersion": "1.3.0rc29",
      "engineImageDigest": "sha256:e9dab9fccc550dcc04356962ff152908d692e0e47ba143f57402a24a4ad64b77",
      "mode": "rr",
      "evaluationProfileDefinitionSha256": "71c7ba2c29d91f5b9ab076707b4df89fae713148d6394b3dd22cb0b5a3cd86b8",
      "result": "BLOCKED_NATIVE_ARCHITECTURE",
      "rawResultSha256": "4a7d4adac0cd0fea271bda240179e36346f9ae8727a99da60af16f5d16135a88",
      "configuration": "matrix-configurations/exaone35-trtllm-rr.json",
      "remainingGates": [
        "release-image-and-host-dataplane",
        "capacity-and-application-SLO",
        "worker-loss-recovery-and-cancellation",
        "clean-replay-and-long-running-resource-return"
      ],
      "architecture": "ExaoneForCausalLM",
      "inferenceAttempted": false,
      "reason": "Stock rc29 PyTorch native executor rejects architecture before checkpoint loading",
      "retestRequirement": "Obtain a separately supported backend/image tuple, then repeat both native and Gateway modes"
    },
    {
      "caseId": "exaone35-trtllm-exact",
      "modelKey": "exaone35",
      "model": "LGAI-EXAONE/EXAONE-3.5-7.8B-Instruct",
      "modelRevision": "553ea250b9a5317231459279d5847d6cf955b9aa",
      "engine": "trtllm",
      "engineVersion": "1.3.0rc29",
      "engineImageDigest": "sha256:e9dab9fccc550dcc04356962ff152908d692e0e47ba143f57402a24a4ad64b77",
      "mode": "exact",
      "evaluationProfileDefinitionSha256": "71c7ba2c29d91f5b9ab076707b4df89fae713148d6394b3dd22cb0b5a3cd86b8",
      "result": "BLOCKED_NATIVE_ARCHITECTURE",
      "rawResultSha256": "4a7d4adac0cd0fea271bda240179e36346f9ae8727a99da60af16f5d16135a88",
      "configuration": "matrix-configurations/exaone35-trtllm-exact.json",
      "remainingGates": [
        "release-image-and-host-dataplane",
        "capacity-and-application-SLO",
        "worker-loss-recovery-and-cancellation",
        "clean-replay-and-long-running-resource-return"
      ],
      "architecture": "ExaoneForCausalLM",
      "inferenceAttempted": false,
      "reason": "Stock rc29 PyTorch native executor rejects architecture before checkpoint loading",
      "retestRequirement": "Obtain a separately supported backend/image tuple, then repeat both native and Gateway modes"
    }
  ]
}