We describe Orbifoldr, an ongoing Haskell project aimed at classifying the wallpaper group of a user-submitted image. The 17 wallpaper groups are the complete classification of two-dimensional repeating-pattern symmetry, naturally encoded by Conway’s orbifold notation. Our approach represents images as functions Coord -> Color so that symmetry transformations are first-class values and symmetry testing reduces to function composition and comparison. The pipeline is entirely pure; image I/O via JuicyPixels is confined to the program boundary. This separation is not merely aesthetic: real images carry noise, compression artifacts, and imperfect periodicity, and the central engineering challenge is to bridge this messy, effectful input to the clean mathematical model. Every bridging decision (what threshold counts as “symmetric enough,” how to vote among candidate lattice vectors, whether to signal uncertainty when no group fits) has the character of an effect on the classification result. We describe the mathematical background, motivate the design choices around this effects boundary, and outline the open problems that remain.