24. Security Model
24.1 Defensive Components
The repository contains security managers and checks for transaction validity, consensus conditions, network behavior, time drift, DDoS controls, Sybil/eclipsing signals, VM limits, storage isolation, replay protection, and slashing evidence. These are implementation controls, not a complete security proof or production audit.
| Area | Code evidence | Qualification |
|---|---|---|
| Transaction safety | Nonce, chain ID, timestamp and signature checks | Must be tested across replay and serialization cases |
| Consensus safety | Quorum accounting, checkpoint validation, duplicate-signer and equivocation paths | Depends on committee configuration and multi-node behavior |
| VM safety | Wasm validation, compute-unit limits, host-function bounds and call-depth controls | Requires bytecode fuzzing and deterministic cross-platform tests |
| Network safety | Peer, time, message and admission protections | Deployment and topology dependent |
| Economic safety | Stake and slashing data structures | Monetary policy and attack cost are not finalized |
| Cross-chain safety | L0 verifier/relay strategies | Trust model differs per chain and adapter |
24.2 Cryptographic Scope
The official repository uses ML-DSA-65 for signing, ML-KEM-768-compatible Kyber for KEM operations, Rescue-Prime + STARK for the hash-based VRF path, SPHINCS+ support, and SHA3/BLAKE3 commitments. Security depends on concrete parameters, implementation correctness, domain separation, proof soundness, key custody and side-channel resistance.
24.3 Slashing and Recovery
Slashing code validates evidence, deduplicates submissions, computes configurable penalties, records history, and can jail validators. Penalty percentages, burn policy, reporter rewards, appeals and unbonding must be fixed by network configuration before mainnet.
24.4 What Is Not Claimed
The repository does not justify a projected dollar attack cost, a guaranteed BFT fault threshold for every deployment, a universal front-running solution, or a production security certification. Those claims require a canonical genesis, threat-model review, independent audits, fuzzing, formal analysis where appropriate, and adversarial multi-validator testnets.
For L0, an external chain inherits only the verification and trust model actually used by its adapter. RPC oracle attestation is not equivalent to a cryptographic light client.