CoT vs ToT vs GoT: Does Reasoning Topology Actually Improve AutoML Decisions?

Four ways to make the same decision: answer directly, reason in a chain, explore a tree, or build a graph of claims. Same local 7B model, same measured evidence, 9 unseen datasets. We did not detect an accuracy difference between any pair; the tree and the graph cost 8.8× and 12.1× the compute of a direct answer. On a stronger hosted model, every topology was less accurate than a direct answer.