This talk presents a practice at the intersection of Critical AI Studies and Computational Design, one that treats software development, interface design, and algorithmic visualization as media for socio-technical investigation. Drawing on critical code studies and tactics such as slow computation, which brings hyper-fast, opaque processes back to a human scale, the work makes machine learning decision-making legible to non-technical audiences. The talk follows three projects that examine human motion modeling across three kinds of space. Labor Optics looks at construction sites as testing grounds for worker surveillance, where acts like drinking water are coded as unproductive time. A second, ongoing project moves into the home, re-enacting domestic labor as a motion-capture dataset to explore intimacy, gendered work, and the algorithmic gaze. The third turns to public space and climate protest in London, setting police and activist worldviews against each other through ontology design. Across all three, the talk develops the idea of the Unmodelled, the human values and practices that computational models cannot see, and proposes counter-datasets and counter-ontologies as ways to open model design to the communities it affects.