The paper works in a mathematical setting called an "o-minimal structure with a generic derivation," which you can think of as a well-behaved geometric world where you also have access to rates of change (derivatives). O-minimal structures are prized because they rule out pathological behavior, keeping definable sets geometrically tame. Adding a derivation lets you ask calculus-style questions in this tame setting. The authors build a systematic way to break up these enriched spaces into simple, manageable pieces called "delta-cells," analogous to how ordinary cell decomposition chops up geometric spaces into cells like intervals and discs. Crucially, each delta-cell comes with "source cells" living in the original o-minimal world, meaning complicated differential questions can be translated into more familiar, finite-dimensional geometric questions.
The paper then constructs a metric space whose points are "definable curve germs," meaning the behavior of curves near a particular point, rather than entire curves. Thinking about germs is a standard technique when you care only about local behavior. The authors identify small neighborhoods in this germ space with products of a special algebraic object called the maximal ideal of a Hardy field, which tracks how definable functions grow or decay near a point. Hardy fields are a classical tool in real analysis for comparing rates of growth, so this connection gives the germ space a concrete algebraic flavor that makes it easier to analyze.
With this germ-space picture in hand, the authors prove a "curve selection lemma" for their enriched setting. Curve selection lemmas are fundamental results in real algebraic and analytic geometry: they say, roughly, that if a point is in the boundary of a set, you can reach it by a nice definable curve staying inside that set. Proving such a result in the differential setting is nontrivial because the extra structure coming from derivatives complicates topology. The authors handle this abstractly using their germ-space framework, and then go further in the special case of "closed ordered differential fields" to write down explicit, concrete curves (asymptotic representatives) that witness the theorem, making the abstract result computationally tangible.