●builderOptimizing your evaluation pipeline for English-centric prompting may lead to unreliable or biased results in multilingual deployments.
●researcherThe interaction between judge model architecture and prompting strategy is a critical variable that must be controlled in multilingual evaluations.