Sentinel — CRE AI Governance Oracle Defense

Sentinel is a risk-oracle defense system that protects oracle-driven protocols from reacting to manipulated or toxic market conditions.
Sentinel separates the pipeline into three layers:

Layer 1 — Signals (data ingestion): Ingests pluggable market + network signals (exchange prices, on-chain reads, gas/mempool indicators, and extendable liquidity inputs such as Uniswap pools). The demo supports SIM inputs and is designed to be wired to LIVE sources.

Layer 2 — CRE + AI governance (decision): A Chainlink CRE-style workflow computes a deterministic risk score and classifies the system into NORMAL / CAUTION / SHIELD / HALT. In borderline (GRAY) conditions, Sentinel optionally invokes a multi-node AI quorum to reduce false positives and decide whether to ACT. This produces an enforceable defense decision in near real time (typically sub-second, depending on consensus mode).

Layer 3 — Trust-minimized enforcement (on-chain): The decision is enforced by broadcasting transactions to an on-chain guard contract (e.g., PriceOracleGuard on Sepolia) to throttle oracle updates (delay) or halt updates (emergency stop). Any observable delay comes from block confirmation, not decision computation.

What it solves: Sentinel provides an automated, graded “circuit breaker” for oracle behavior—throttle, lock modes, or halt—reducing the blast radius of manipulation while keeping the system auditable and composable.

• CRE workflow (off-chain): fetch inputs → compute risk → decide target defense mode (ACT/SKIP)
• AI consensus (optional in GRAY): 5-node quorum aggregation to reduce false positives
• On-chain guard (Solidity): PriceOracleGuard enforces mode + oracle delay / emergency halt
• Runner/client (TypeScript): tx broadcast with cooldown + idempotency; clear “decision vs enforcement” logging
• Dashboard (Next.js): visualizes signals → risk → consensus → on-chain state; scenario-based simulation