Text this: mdok-style at SemEval-2026 Task 9: Finetuning LLMs for Multilingual Polarization Detection