We analyze how the ordering of in-context examples affects causal language model performance and propose strategies to reduce order sensitivity, improving robustness across downstream tasks.