[ICLR'26] Taming the Wild Hypothesis: Introducing Controllable Abductive Reasoning in Knowledge Graphs