Four hundred seventy-six nuclear power plants are in operation or under construction around the world. Are concepts for designing and operating plants safely sufficient? Conventional approaches are premised on expectations of predictability and control of radiation release and on assumptions that plant operations are closed systems. Field observations in the industry find, however, that the periodic necessity to refuel, test safety equipment, and continuously upgrade plant designs introduces challenges to control not originally calculated. The social and cultural contexts of markets, (...) regulation, and work systems create additional contingencies. In this dynamic environment, reactor operations produce invaluable data about the states of the entire technology and organizational system. Rereading these dynamics as "experiments" suggests an alternative epistemol ogy that synthesizes science's values of doubt, discovery, and documentation with engineering's values of efficiency, optimization, and problem solving. This redefines plant-based risk handlers as ethologists, geologists, and anthropologists of an ecology of visible and invisible hazards. (shrink)