Both knowledge graph strategies passed the score-40 bottleneck and achieved comparable scores; KG-A2C-chained converged faster and was more sample efficient.