That makes more sense. It’s still horribly non-intuitive. In particular, I don’t like the “qualitative” background shading technique, because it makes it appear as if a “satisfactory” performance is all of bad, good and satisfactory at the same time. Why not simply mark those thresholds under the scale of the x-axis? The background shading is disproportionately distracting, considering what little information it encodes.

That said, I can see maybe it would help when you are stacking multiple of these on top of each other and each has its own qualitative ranges.