Abstract
Variation is a foundational biological principle that has historically been marginalized-both due to limited experimental accessibility and because of idealized, stereotypic blueprints rooted in essentialist thinking. With the advent of genetics and quantitative biology investigating environmental influences on the phenotype, variation was redefined from mere noise to a fundamental property. Modern light sheet microscopy now enables high-resolution, long-term imaging of dynamic processes across large populations, making it possible to systematically study phenotypic variation in vivo. Yet, the resulting high-dimensional datasets overwhelm traditional modes of analysis and visualization, risking the loss of biological insight. The transition from qualitative representation to quantitative measurement demands new epistemic practices-shifting from selective human interpretation to computational abstraction. Instead of relying on either very limited sampling or exhaustive scanning, we advocate for representative sampling of phenotypic variation: adaptive, model-guided systems that dynamically sample biological variation using real-time feedback, directing attention towards biologically relevant events and rare or extreme phenotypes. The underlying models act as the interface to human insight, constructing navigable, queryable representations of variation as a multidimensional manifold shaped by genetics, environment, and stochasticity. Crucially, adaptive systems call for new methods of visualizations-interfaces that encode uncertainty, consensus, and distributional structure. Such visualizations should preserve the interpretability of historical illustrations while fully embracing biological variation. The future of biology lies not in acquiring more data, but in developing smarter ways to sample, represent, and understand it.