In the past two years alone, at least six systematic reviews or meta-analyses have examined the interventions that improve learning outcomes in low- and middle-income countries. However, these reviews have sometimes reached starkly different conclusions: reviews, in turn, recommend information technology, interventions that provide information about school quality, or even basic infrastructure (such as desks) to achieve the greatest improvements in student learning. This paper demonstrates that these divergent conclusions are largely driven by differences in the samples of research incorporated by each review. The top recommendations in a given review are often driven by the results of evaluations not included in other reviews. Of 227 studies with student learning results, the most inclusive review incorporates less than half of the total studies. Variance in classification also plays a role. Across the reviews, the three classes of programs that are recommended with some consistency (albeit under different names) are pedagogical interventions (including computer-assisted learning) that tailor teaching to student skills; repeated teacher training interventions, often linked to another pedagogical intervention; and improving accountability through contracts or performance incentives, at least in certain contexts. Future reviews will be most useful if they combine narrative review with meta-analysis, conduct more exhaustive searches, and maintain low aggregation of intervention categories.