This paper provides new insights on the science of scaling. We study an educational mentoring program with a home visit component implemented at scale in Mexico, under different modalities (original and enhanced training for mentors) and different situations (field experiment and policy implementation). While the program was ineffective when implemented by the government in its original modality, the enhanced modality boosts children's outcomes, both in the field experiment and during the government implementation. Higher-quality home visits encourage parent/child and parent/community interactions, which in turn are found to promote the scalability of the program. Our work provides new knowledge on the socially determined nature of scaling educational programs.