svg's primitives are shapes and you program it with js and dom.
canvas primitive is a 2D array of pixels, and you program it with js and an api.
webgpu's primitive are pipelines (or whatever) and you program it with js and an api.
Maybe you're hung up on using the dom instead of an api. But the dom is an api, and it is further specialized by svg. Mutating the screen is exactly equivalent to mutating the dom within SVG. The mental model is that of a persisting scene that you nudge around with changes. This, versus the mental model of canvas where shape persistence is up to you. My understanding of low-level 3d graphics (limited to that of a hobbyist) is that driver commands setup a pipeline and then nudge around a scene graph in a similar fashion to svg. So it is doubly ironic that she would omit it.