As a pivotal intermediary in the supply chain, retailers must undertake two key responsibilities: fulfilling customer demands and sourcing from suitable suppliers. However, in increasingly complex supply chains, both stochastic customer demands and stochastic supplier disruptions present retailers with new challenges in fulfilling these responsibilities. This paper studies a budget constrained sourcing and distribution problem for a retailer under multiple uncertainties. Customer demands arrive stochastically, while suppliers may suffer stochastic disruptions and uncertain recovery processes. To minimize total demand tardiness under a limited sourcing budget, a reliability aware reinforcement learning method is proposed. The method integrates reliability theory with reinforcement learning through three modules. A reliability aware heuristic strategy evaluates suppliers and provides robust strategies. An enhanced learning component generates intelligent dynamic policies. A decision fusion training architecture adaptively combines heuristic and learned decisions according to real time system conditions. Numerical experiments under four uncertainty scenarios show that the proposed method achieves higher cumulative rewards than representative reinforcement learning methods and lower tardiness than competing methods for similar problems.