Smart contracts

Security model

Known AMM attacks and how Neura V1 handles each one.

Unaudited reference code. The mechanisms below are not guarantees against all attacks. See testing for executable coverage. Threat Mechanism and limits in V1 Re-entrancy lock modifier on every state-changing pool function LP share inflation (first depositor) 1,000 volatile units; decimal-aware larger stable lock Stable-pool dust initialisation Balanced first deposit; stable lock and post-update invariant floor Sandwich / MEV Surcharge on all later swaps; no bot detection or guaranteed protection Rounding exploits Bisection retains minimal safe reserve; bounded arithmetic and fuzz checks Fee-on-transfer tokens Pool reads real balances; Router checks real received amount Non-standard ERC-20 (no bool) Low-level call accepting empty return data Oracle manipulation Volatile raw-ratio cumulatives only; consumers need reviewed observations and risk controls Overflow Checked math, uint112 raw cap and 10^28 stable normalized cap Admin abuse No owner, no pause, no upgrade, immutable fees Stale transactions deadline on all Router calls Known limits Rebasing tokens (balance changes without transfers) are not supported — use wrapped versions. A pool cannot protect against a malicious token contract; Neura AI flags unknown tokens. skim lets anyone collect donated excess; never send tokens directly to a pool. Before mainnet Independent audits and a deployment review remain required. Local regression/fuzz tests are included, but formal verification and longer adversarial campaigns remain outstanding. No completed audit, bounty program or deployment date is claimed.