Quantified self-experimentation with personal diets is a popular activity among health enthusiasts, diagnosed patients, as well as "life hackers" pursuing self-optimization goals. In this paper, we reflect on self-experimentation practices in the context of amateur citizen science communities. We report findings from 11 month-long qualitative fieldwork in a community of nutrition hobbyists experimenting with a powdered food substitute "soylent". Our respondents customized the soylent powders to their personal needs, tracked their metabolic reactions to the diet, and discussed their findings with the online soylent user community. Although the data and knowledge sharing within the community positively impacted respondents' nutrition literacy, these activities created risks regarding their health safety and data privacy. We define soylent selfexperimentation as a form of "extreme citizen science". Based on the limitations identified in the soylent community, we suggest a set of design recommendations for extreme citizen science projects.