Race to the Bottom is the most recent of a string of improvising machines involving bowed cardboard boxes, developed over the last decade. In all these systems, bowed cardboard is both the source of all sonic material and the ���control��� interface to software that occupies a changeable and turbulent role in the territories between algorithmic co-player, instrument and processor. Boxes, it turns out, yield a much more varied sound world, and more room for practised technique than I had imagined when I started exploring them (somewhat facetiously). They also present a range of interesting challenges to machine listening algorithms, such are the instabilities and varied points of interest in their sound. All of these improvising machines have explored different approaches to dealing with this, and finding creative ways of enjoying the software���s frequent ���misunderstandings��� of its input. Increasingly, these machines have also been a place for me to investigate ways of dealing with time in algorithmically mediated improvising, particularly when (as here) my hands are already busy, and I have to trust my software���s sense of time and musicality (or at least put up with it). In Race to the Bottom, these fronts are explored by abusing segmentation algorithms and beat trackers as (loose- ish) analogues for, respectively, oscillators and filters. A clutch of these run at different rates, latching on to different parts of different sounds, and interfering with each other, informing both the processing of sound and the unfolding of musical shape.