Combining literature grounding with explicit, traceable reasoning operators improves AI-generated research ideas, but the quality gains come more from iterative refinement than from the specific operators themselves—the real value is in inspectability and source attribution.
This paper presents CAR, an AI framework that generates research ideas by grounding them in literature, iteratively developing concepts, and explicitly tracking the reasoning process. CAR retrieves relevant papers, builds a knowledge graph, and uses four conceptual operators (bisociation, analogy, exploration, elaboration) to transform ideas while recording each step and its sources.