An Empirical Juridical-Digital Twin Integrating Blockchain Traceability, Multi-Agent Reinforcement Learning, and Demand-Signal Optimization for Micro-Enterprise Supply Chains in Pesisir Barat

Background: Micro-enterprises operating along the Pesisir Barat coastline experience chronic product spoilage and demurrage because logistics are fragmented and contractual enforcement under the Indonesian Civil Code (KUHPerdata) is weak. Objectives: This study aims to construct a juridical-digital twin that (1) optimizes logistical performance through multi-agent reinforcement learning (MARL), (2) guarantees compliance with Articles 1320, 1338, and 1243 of the KUHPerdata, and (3) allocates risk transparently via blockchain-based smart contracts. Methods: A three-layer architecture was implemented. The physical layer collects high-frequency IoT data (GPS, temperature, wave height, fuel flow). The digital-twin layer simulates the supply-chain using a Markov decision process whose state vector S_t^i=[GPS_t,T_t,Hs_t,F_t,D_t] and action set A_t^i={route, temp}. Reward R_t^i= -alpha·Spoilage_t^i -beta·Demurrage_t^i -gamma·P_legal^i + δ·ServiceLevel_t^i incorporates a legally calibrated penalty P_legal derived from smart-contract clauses. Policy-gradient updates Delta θ=alpha∇_θ logπ_θ(A_t|S_t)(G_t-V_θ(S_t)) were trained on 12 months of pilot data. Demand forecasting employed a calibrated LSTM (24-h window, 128-unit hidden layer, MSE loss). Legal constraints were encoded as Solidity smart contracts: Clause 1 triggers liquidated-damage payments for temperature breaches; Clause 2 defines force-majeure based on BMKG wave-height advisories; Clause 3 enforces a good-faith risk-sharing rule limited to 10 % of contract value. Published in Journal of Coastal Frontiers and Applied Development (JCFAD). CERN Zenodo Record: https://doi.org/10.5281/zenodo.24047039. Publisher: PT CHAIN ASIA CONECT Research and Development Division.