Drop a new track into a 3D cloud of labelled records, arranged by tempo, spectral centroid and percussiveness, and watch a real k-nearest-neighbours classifier pick its closest neighbours and vote. Change the track, k, the distance metric and the vote weighting and see the decision slice, the confidence, the leave-one-out accuracy and the best k, with built-in numerical checks. Turntable, records and speaker modelled in Blender; the data are synthetic.