Study design determines what a result can mean
Randomised controlled trials establish causation but are short and small in nutrition, because you cannot randomise people to a forty-year diet. Prospective cohort studies follow hundreds of thousands of people for decades and can detect long-run disease outcomes, but cannot fully exclude confounding. Mechanistic and animal studies explain how something might work but predict human outcomes poorly. Strong conclusions come from agreement across all three, not from any single paper.