Reduced emissions from deforestation and forest degradation (REDD+) promise to deliver performance-based, cost-effective climate change mitigation. 15 years after REDD+’s conception, we analyse the rigorous counterfactual-based ev-idence for environmental and welfare effects from national and subnational initiatives, along a REDD+ Theory of Change. Using machine-learning tools for literature review, we compare 32 quantitative studies including 26 primary forest-related and 12 socioeconomic effect sizes. Average environmental impacts were positively significant yet moderately sized, comparable to impacts from other conservation tools, and mostly impermanent over time. Socioeconomic impacts were welfare-neutral to slightly positive, especially at outcome stage (e.g. rising incomes). Moderator analysis shows that REDD+ environmental additionality was likely restricted by project proponents’ ‘high-and-far’ spatial targeting of low-threat areas (adverse selection bias). Disappointingly scarce funding flows from carbon markets and ill-enforced condi-tionality probably also limited impacts. Hence, important policy and implementation lessons emerge for boosting effec-tiveness in the current global transition towards larger-scale, jurisdictional REDD+ action.