Research questionHow can we evaluate LLM knowledge updates over time without contamination or inconsistent facts?Static pretraining leaves LLM knowledge increasingly outdated. Existing evaluations of knowledge editing can become contaminated or introduce counterfactual facts that conflict with the model’s established knowledge.