Give a decision framework for choosing between prompting, RAG, and fine-tuning, with cost and latency attached.

Fine-tuning updates model behavior on task-specific examples. A sound plan defines the target behavior, cleans and splits data, chooses full or parameter-efficient updates, watches retained capabilities, and evaluates against prompting and RAG baselines.

Use equations or tensor shapes where they clarify the claim, then name an experiment or ablation that would distinguish competing explanations.