Function radial_shell_occupancy 

pub fn radial_shell_occupancy(img: &CaptureImage, bg: [u8; 4], eps: u8) -> usize

How many of RADIAL_SHELLS concentric equal-radius annuli over the frame’s inscribed disc contain a meaningful share of lit pixels (0..=RADIAL_SHELLS) — a structural occupancy measure: at how many radii does this picture exist?

coverage counts lit pixels, which at capture size measures a thin-stroke figure’s halo rather than its geometry: at 96×96 the bare rosette and a 46×-denser four-ring mandala score identically, and 54 % more geometry moves the number 2.6 % (design-backlog 0072). This asks the question that actually separates them — the mandala exists at nine of ten radii, the rosette’s interlace band at one to three — and it cannot be bought with glow or trails, because inflating the halo around a stroke does not move which shells the stroke lives in.

Pixels outside the inscribed disc (the frame’s corners) are ignored: the measure is radial, and the corners exist at radii only a diagonal figure reaches. 0 for a frame with no lit pixels — a scene that renders nothing occupies nothing, which is the one conviction the coverage floor demonstrably gets right and this measure must preserve.