Both knowledge graph strategies passed the Zork1 bottleneck and achieved comparable scores, but KG-A2C-chained converged faster and was more sample efficient.