I'd say you have the option between using either sizing, and should decide based on which sizing your population will play worse against. Try taking a look and how the solver responds to each sizing and compare with how you think villains will actually respond in game. Often you can find that one sizing prompts more mistakes than another.
Another thing worth trying is experimenting with different calling ranges, for the BTN in your scenario. I ran SB v BTN 3B pot sims in Pio against a moderate 10% BTN calling range, and also against a potential loose recreational player's 22% calling range. On T92ss for example, you can see vs the looser range the solver only uses a larger sizing, but splits sizings vs the tighter range.
(from an aggregate report comparison)