PASO: Prediction-Anchored Sequence Optimization for Latency-Constrained AMM Batch Execution on Blockchains

Transaction ordering in automated market makers (AMMs) directly determines price trajectories and user slippage, yet most on-chain rules remain insensitive to short-horizon predictive signals. This work introduces PASO - PredictionAnchored Sequence Optimization - a lightweight, latency-aware ordering framework that explicitly encodes short-term price predictions into a unified sequence-level objective. PASO searches feasible transaction permutations through heuristic adjacent swaps and incremental replay under constant-product AMM constraints. In replay-style experiments, PASO achieves the lowest anchor-tracking error (both median anchor gap and log-MSE) at window size w = 50, outperforming TWAP-300s and V3-fixed. On own-path execution metrics, its slippage is comparable to FIFO and slightly higher than CLVR, reflecting a deliberate stabilitytracking trade-off tunable by $\lambda$ and w. These results demonstrate that prediction-anchored optimization can substantially improve ordering quality within real-time blockchain latency limits without modifying AMM settlement rules.