3D curve inference from 2D sketch strokes typically employs a curve fitness metric [Schmidt et al. 2009; Xu et al. 2014], with both discrete (for example to capture geometric regularity or curve-type constraints) and continuous terms (for example to model curve smoothness or minimal variation from the 2D stroke). This mix of prioritized discrete and continuous constraints makes fitness optimization for sketch understanding a challenging problem. Solutions such as combinatorial enumeration [Schmidt et al. 2009] or iterative least squares refinement [Xu et al. 2014] of discrete constraints, are tailored to exploit properties specific to the sketch problem domain. In our context of layered sketch strokes, prioritized discrete constraints take the form of admissible curve-types (Section 3.1), that we are able to efficiently classify from the 2D strokes and underlying 3D geometry. While we can similarly cast 3D curve inference as the optimization of a fitness function over a family of candidate 3D curves, our curve-type classification algorithm ( Figure 4) allows us to find a suitably fit 3D curve directly.