Fornex - what it is, what proves it.

FAQ

1. What stops the agent from breaking the leverage cap?

   The Anchor program. log_multi_agent_decision rejects any BULL vote with leverage > 3, any BEAR or ZEN vote > 2, and any consensus with confidence < 60%. update_nav rejects any write outside ±10% per cycle. record_trade_outcome rejects any single-trade PnL beyond ±50% of NAV. Caps live in the program, not the agent - see programs/fornex/src/errors.rs.

2. Why synthetic perps instead of just Drift?

   Drift devnet has been intermittent. We built a self-contained SyntheticPosition PDA that opens with a Pyth-stamped entry price, closes against a Pyth-stamped exit, and computes realized PnL directly in the Anchor program. No external dex dependency. The Drift executor is wired and switchable via FORNEX_EXECUTOR=drift when devnet is healthy.

3. Where does pay.sh money come from? Is it a self-transfer?

   Real on-chain SystemProgram::transfer from a separate treasury keypair to the agent wallet, on every executed trade. The treasury is funded once on devnet; if the env var is missing the code refuses to fake a self-transfer (see agent/src/paysh.ts). Click any pay.sh tx on the proof page - the From account is treasury, not the agent.

4. What does the agent wallet hold?

   On devnet, the agent wallet holds a small float of SOL it uses to execute Drift orders and to pay tx fees for on-chain logging. User deposits live in the Vault PDA, not the agent wallet. The mainnet plan is for the Vault PDA to CPI into Drift via signer seeds so the agent never custodies user funds.

5. Is the AI reasoning stored on-chain or just hashed?

   Stored, not hashed. Each AgentVote reserves 200 bytes for reasoning text, and every MultiAgentDecision PDA carries three of them - BULL, BEAR, ZEN - plus the consensus. That's the wall on /proof. Truncation is documented in agent/src/config.ts.

6. Why does the win-rate stat actually mean something?

   It's derived from record_trade_outcome, which only increments on closed positions with realized PnL. Deposits never count toward win-rate. The inception NAV is stamped on-chain at first deposit so every PnL number on the dashboard is anchored to a real on-chain start point.

7. Can I run this myself?

   Yes. git clone, npm install, anchor build, fund a devnet keypair, set the env vars listed in .env.local.example and agent/.env.example, then npm run dev and npm run agent. Single-cycle and force-direction flags are documented in DEMO_SCRIPT.md.

8. What happens if the LLM goes down?

   The cycle returns a safe FLAT vote per agent, the consensus comes out FLAT, and the on-chain log records that explicitly. No order is placed. Operational status (Pyth, RPC, agent heartbeat) is surfaced on the dashboard topbar.

9. Does the dashboard ever invent numbers?

   No. NAV, share supply, decision count, win-rate, executed-trade count, and pay.sh streamed are all decoded directly from the Vault PDA + program account list. Source: lib/chain.ts and components/TrustStrip.tsx.

10. How does the dashboard get live updates without a backend?

    The dashboard opens a logsSubscribe WebSocket directly to the configured Solana RPC and filters for log lines that mention Fornex instructions (`log_multi_agent_decision`, `record_trade_outcome`, `update_agent_reputation`, synthetic open/close). When a match arrives, the relevant page re-fetches. Auto-reconnects with capped exponential backoff. Source: hooks/useDecisionStream.ts.

11. Can I read Fornex state from my own app?

    Yes. The repo ships @fornex/sdk at packages/sdk/ - a tiny read-only TypeScript client with peer dep @solana/web3.js only. getVault, getDecisions, getNavHistory, getAgentReputation, and getVaultStrategy cover every Fornex PDA in one line.