Optimistic rollups provide scalable smart-contract execution but remain unsuitable for regulated financial applications due to three gaps: lack of semantic legitimacy checks, vulnerability to L1-L2 state divergence, and susceptibility to MEV-driven transaction reordering. We propose RegGuard, a unified framework that adds formal legitimacy guarantees to optimistic rollups. RegGuard includes: (1) a decidable semantic validator using the RegSpec rule language to encode and enforce regulatory and business constraints; (2) a state presynchronization validator that detects inconsistent cross-layer assumptions via a high-freshness L1 cache and differential dependency tracking; and (3) a verifiable fair-ordering protocol based on threshold encryption and binding commitments, achieving (α, β)-fair sequencing under standard cryptographic assumptions. We formalize correctness and fairness properties for each component and implement a 15k-LOC prototype integrated into an Optimism-based rollup. Experiments on a distributed testbed show that RegGuard reduces settlement failures by over 9 0%, prevents detectable ordering manipulation, and sustains more than 85% of baseline throughput, demonstrating that strong legitimacy guarantees can coexist with rollup scalability.

Sybil attacks increasingly threaten blockchain identity systems, especially in Layer 2 ecosystems. Existing methods relying on static statistical features are easily evaded by adversarial noise injection. We propose DeepSybil, an end-toend dynamic graph learning framework employing a SpatioTemporal Dual Encoder: Graph Attention Networks (GAT) for topological invariance and LSTM for sequential behavioral rhythm. Experiments on KYC-verified (BAB) and real-world Layer 2 datasets demonstrate that DeepSybil significantly outperforms baselines, retaining 92 % detection performance in crosschain zero-shot transfer while baselines degrade by over 20 %.