This paper investigates how people edit their episodic memory when incorrect information is later identified in a sequence of messages. The study finds that memory editing is ineffective when old, incorrect information plays a central role in the message, and even when corrections are recalled, inferences continue to be drawn from the outdated information. The paper concludes that memory contradictions are resolved locally, based on recent updates, but old information is harder to remove if it better fits the context of the message.