To flesh this out, look at the examples in the article where moderation decisions split or were reversed - that's the problem XCheck is designed to solve. Most are about politics:
- Saying "white paint colors are the worst" was classed as racism. Trying to define and then scrub all racist speech from the world is a left wing policy.
- A journalist who (we are told) was criticizing Osama bin Laden was classed automatically as supporting him, and then human moderators agreed. Scrubbing this sort of thing is (in the USA) historically either a right wing or bipartisan consensus. We don't know why the moderators agreed because the comments themselves are not presented to us. This was later overridden by higher ups at Facebook.
- Split decisions over Trump's post saying "when the looting starts the shooting starts" that got escalated to Zuckerberg himself.
- "The program included most government officials but didn't include all public candidates for office".
etc. If you try to ignore politics, the entire debate becomes meaningless, because indeed, you would not even perceive any problem with unequal enforcement in the first place.