{
  "scope": "Bounded three-worker converged GPU QA; not full qualification, support promotion, performance or P/D fail-over acceptance",
  "gatewaySource": "1d51393773e4974b5e356377496c9d677dba9f9c",
  "gatewayMode": "source-built proxy-only, isolated container network; backend receipt instrumentation",
  "model": "deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B",
  "modelRevision": "ad9f0ae0864d7fbcd1cd905e3c6c5b069cc8b562",
  "engineVersion": "0.31.0",
  "engineImage": "sha256:a4a4c0437bf7240089da5f08aa370c4aee17ae5290f7a3b468825ee26c4c3a6b",
  "gpu": "3 x NVIDIA L4 24 GB",
  "evaluationManifestSha256": "253952ba951171a185cfb5a736d88438d134f9c3604493a5551a9a5ca4a3bb32",
  "unchangedSourceProfileTokenizerTemplateFixtures": true,
  "nativeChatAndSse": {
    "normal": 6,
    "offered": 6
  },
  "rr": {
    "normal": 12,
    "offered": 12,
    "backendCounts": [
      4,
      4,
      4
    ],
    "receiptErrors": 0
  },
  "exact": {
    "state": "READY",
    "reasonCodes": [],
    "goFenced": false,
    "normal": 12,
    "offered": 12,
    "prewarmedWorkerReceipts": 12,
    "cacheHitCounterDelta": 12,
    "receiptErrors": 0
  },
  "lifecycle": {
    "invalidModeRejected": 400,
    "invalidModeLeavesRuleUnchanged": true,
    "freshRequestsAfterWorkerStop": {
      "normal": 6,
      "offered": 6,
      "stoppedWorkerReceipts": 0
    },
    "workerRecovery": {
      "nativeHealth": true,
      "exactReady": true
    },
    "rrExactRrSwitchRequests": 6,
    "receiptErrors": 0,
    "finalRuleAbsent": true,
    "gatewayReadyAfterDeletion": true
  },
  "oldManifestNegative": {
    "declaredEngineVersion": "0.28.0",
    "runtimeEngineVersion": "0.31.0",
    "state": "PROFILE_VALIDATED",
    "reasonCodes": [
      "identity_mismatch"
    ],
    "goFenced": false,
    "chatHttpStatus": 200,
    "interpretation": "Successful ordinary/fallback serving; not Exact readiness or evidence of global fail-closed enforcement"
  },
  "remainingGates": [
    "uncached-prefix comparator",
    "post-recovery routing redistribution",
    "in-flight SSE interruption and client cancellation/resource return",
    "all original five-card tuples",
    "SGLang and TensorRT-LLM runtime cases",
    "sustained capacity and production SLO",
    "host eBPF, HA, P/D and release qualification"
  ],
  "publicEvidence": "Bounded functional summary; not support promotion, production performance or release qualification"
}