FeedOracle Specifications

Grounding Receipt 2.0

Cryptographically signed proof that a tool call was actually executed by an MCP server. Provider-neutral — works with Anthropic, OpenAI, xAI, or any MCP-compatible LLM. Designed for DORA Art. 17, MiCA Art. 70, BaFin BAIT, and EU AI Act audit requirements.

Format: JSON · Signature: ES256K · Anchor: Polygon mesh-economics-v1 · Verification: offline via public JWKS

Read specification

Tier Classification (T0–T4)

Five-tier classification system for routing AI requests through appropriate verification layers. Balances latency cost against verification depth. Used as Layer 0 in the Ultimate Verification Oracle architecture.

Read specification

Approach: rule-based + ML hybrid · Calibration: AUROC ≥ 0.95 target

Ultimate Verification Oracle (UVO)

Seven-layer verification architecture between probabilistic LLM outputs and deterministic regulatory decisions. Three layers (Tier Classifier, Claim Decomposition, Neuro-Symbolic Verifier) are in production today.

Public MCP endpoint:

https://feedoracle.io/uvo/mcp/

Three tools available:

• uvo_verify — full pipeline, returns final_decision + per-atom verdicts + signed evidence chain
• uvo_status — health + version + active layers (free, no auth)
• uvo_pipeline_spec — architecture details + tier routing rules

Live test:

curl -X POST https://feedoracle.io/uvo/mcp/ \
  -H 'Content-Type: application/json' \
  -H 'Accept: application/json, text/event-stream' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call",
       "params":{"name":"uvo_verify",
                 "arguments":{"text":"USDT ist gemäß MiCA in der EU zugelassen."}}}'

Status: Layer 0 + Layer 2 + Layer 3 in production. Layers 4–7 forthcoming. Each call returns a Grounding Receipt v0.1 cryptographically signed (ES256K). Latency: T0/T1 < 1s, T2 ~ 7s, T3 async.