Guardian Intelligence Whitepaper
Protocol Whitepaper · Trust Artifact

Survivability-First
DeFi Intelligence
Guardian is not a bot, a script, or an APY ranker. It is a cognitive system that watches markets, protects capital, and compounds yield autonomously — without requiring a single human intervention.
Most DeFi interfaces optimize for one metric: APY. The highest number wins the click. But APY does not tell you whether a position will survive a 30% market move, whether the protocol's oracle is reliable, whether the funding rate will invert, or whether compounding requires human intervention to avoid liquidation.
Fifteen things Guardian does — or refuses to do — that the rest of DeFi can't or won't. Every one is real in the code today, honestly labeled MODELED until it is live on-chain. Tap any card for the detail.
Guardian supports five strategies built on top of YVIP. Each has a different risk profile, leverage level, and regime sensitivity. The per-regime APYs below are MODELED scenario projections — not realized returns — computed by the Scenario Engine for each combination.
| STRATEGY | BULL APY | NEUTRAL APY | BEAR APY | LIQ BUFFER | SCORE |
|---|---|---|---|---|---|
| Simple Hold | 32% | 28% | 24% | None | 84 |
| Loop 2× | 68% | 58% | 49% | 24% buffer | 72 |
| Loop 3× | 80% | 68% | LIQD | 15% buffer | 55 |
| Delta-Neutral | 26% | 26% | 28% | None | 84 |
| Bribe-Compound | 36% | 32% | 28% | None | 91 |
Guardian operates under a philosophical constitution — a set of non-negotiable principles that govern how it reasons, suppresses, and explains. These qualities cannot be overridden by market performance pressure, user requests, or protocol incentives.
By design, every yield sweep, basket rebalance, LP range adjustment, bribe compound, and risk response is executed by on-chain protocols — not by a human operator. The protocol owner's wallet holds no hot keys to the reserve and cannot sign a fund-movement transaction. Guardian is pre-mainnet, so there is no operating history yet; at mainnet every capital flow will be verifiable on-chain against the program's instruction set — provable, not merely claimed.