Fact-checking: the research version of "do not trust yourself"

DR agents add an explicit verification loop, which is Chapter 13's reliability law applied to facts. After drafting an answer, a good DR agent does not deliver it immediately. It cross-checks: it looks for independent sources that confirm a fact and searches for contradictions. Grok DeepSearch rates the credibility of every source and verifies key claims across multiple origins. Some systems (Zhipu's Rumination model) pause after concluding and keep searching to test whether the conclusion holds, then finalize. This multi-source cross-validation plus self-reflection is how DR agents drive down hallucination, and it is encouraged during training by correctness-oriented rewards.