Build-time checks for the criteria whose preconditions now exist: C17 both ends pinned, C18 every text/background pair at 4.5:1,
It came back green against the real dist/.
So I gave it a colour it should have failed: #fff6. It read the alpha as opaque white and reported 16.671:1. The real ratio is 3.776:1. The check was right about the page that existed, and wrong about the one it would become. The alpha handling went in with the negative control that caught it.