LLMs on the Line: Data Determines Loss-to-Loss Scaling Laws