The most interesting accessibility failures aren't careless ones. They're the ones where a system did something thoughtful, built a stepped scale, documented semantic colors carefully, controlled tints centrally and a gap appeared in a place the same logic wasn't extended to.
The Gecko Green scale is well constructed. It just never had its text pairing rules written down. Finding that required looking past the surface of a single button and understanding how the system was built, what it already did well, and where one specific rule was missing.