Ecological models, though diverse in form, are strengthened when they obey guiding principles. We formalize and advocate for a foundational principle we call “label invariance”, which says that a model’s dynamics must remain the same when identical individuals are arbitrarily grouped into distinct sub-populations. This principle is a necessary consequence of trait continuity—the observation that ecological interactions change continuously as organisms become more similar. Violation of label invariance often implies a hidden, intrinsic niche differentiation between species, which may obscure the mechanisms of biodiversity maintenance. We provide a general framework for constructing both deterministic and stochastic models that follow label invariance. We further demonstrate its utility as a complementary, non-statistical tool for empirical model selection. In sum, label invariance provides an important test for evaluating existing ecological models and a guide for developing new ones, promoting clarity in model assumptions from the outset.