Since the early 2000s, a movement within development economics has sought to bring the randomized controlled trial (RCT), long a fixture of medical research, to the study of poverty. Rather than debating from theory whether, say, microcredit or free bed nets reduce hardship, researchers randomly assign the intervention to some communities and withhold it from others, then compare outcomes. Because assignment is random, differences that emerge can, in principle, be attributed to the intervention itself rather than to preexisting differences between the groups. Advocates hail the method as a corrective to a discipline long prone to grand, untested generalizations. The approach has yielded sobering findings. Several large trials of microcredit - the extension of small loans to poor entrepreneurs, once promoted as a near-panacea - found only modest effects on business investment and none on broader measures of household welfare, puncturing inflated claims. Here the RCT's value is largely negative but real: it disciplines enthusiasm by subjecting popular interventions to a demanding test. Critics, however, raise a problem of external validity. An RCT establishes that an intervention worked in a particular place at a particular time, among a particular population. It does not, by itself, establish that the same intervention will work elsewhere, where institutions, culture, and complementary conditions differ. A deworming program that raised school attendance in one Kenyan district may fail in a region with a different disease burden or different schooling incentives. The very feature that gives the RCT its internal rigor - its focus on a specific, controlled comparison - is what limits the reach of its conclusions. Defenders of the method reply that this criticism, while fair, is not unique to experiments; every empirical finding, however generated, must be extrapolated with care. Moreover, they argue, replication across varied settings can gradually map the conditions under which an intervention succeeds. A single trial proves little about the wider world, but an accumulation of trials, interpreted alongside a theory of why the intervention works, can support cautious generalization. The disagreement, on this view, is less about whether RCTs are useful than about how modest their claims should be. A deeper objection questions the movement's implicit priorities. By design, RCTs favor interventions that can be randomized and measured over short horizons - a fertilizer subsidy, a cash transfer, a teacher incentive. They are ill-suited to evaluating the large, systemic changes - industrial policy, the building of legal institutions, the structural transformation of an economy - that historically account for the greatest reductions in poverty but that cannot be switched on and off at random across villages. A method that can rigorously assess only what it can randomize may, critics warn, quietly narrow the field's sense of which questions are worth asking. Defenders concede the point but deny the implication: a tool need not answer every question to be worth using, and rigor about small things is preferable to confident ignorance about large ones. The debate, still unresolved, is ultimately about the proper division of labor between experiment and other forms of evidence.