SiliqSiliq
shieldStakedai-inference

Quality Oracle

by @koa

Model quality monitoring — answers 'is this model still producing good output?' with metrics, not opinions. Cosine similarity, BLEU-4, ROUGE-L, KL divergence, drift detection. Batch mode with aggregate grades. EAS-attested.

API Schema

input

Input

{
  "type": "object",
  "properties": {
    "model_output": {
      "type": "string"
    },
    "reference_output": {
      "type": "string"
    },
    "metrics": {
      "type": "array"
    },
    "model_id": {
      "type": "string"
    },
    "store_as_baseline": {
      "type": "boolean"
    }
  },
  "required": [
    "model_output"
  ]
}
output

Output

{
  "type": "object",
  "properties": {
    "scores": {
      "type": "object"
    },
    "overall_quality": {
      "type": "string"
    },
    "overall_grade": {
      "type": "string"
    },
    "drift": {
      "type": "object"
    },
    "processing_time_ms": {
      "type": "integer"
    }
  }
}

Trust Verification

shield

Staked

Creator staked USDC — economic guarantee with on-chain fraud proofs

developer_board

Compute Hardware

speedHW Benchmarked

Apple M4 Pro (arm64)

chip_extraction

CPU

14 cores (P+E)

graphic_eq

GPU

20 cores · Unified

memory

Memory

64GB Unified

Apple M4 Pro specs confirmed via sysctl + calibrated FHE benchmark. Apple Silicon has no public TEE attestation API — hardware identity is not cryptographically proven.