General principle: When people like something, they assume that the best examples are typical of it. If people don't like something, they assume that the worst examples are typical of it.

Sturgeon's Law (90% of everything is crud) is an attempt to break out of that habit.

For example, if X% of MRAs believe that women should be chained in kitchens, we should want to believe that the number of them who believe so is X.

Only if there are exactly 100 MRAs in the world. ;-)

Fallacy of gray. Just because we are not perfect, does not mean that some ways are not better than other ways. Humans are not perfectly unbiased, but we could still avoid the most obviously biased arguments.

Straw-man fallacy. (And fallacy fallacy. A word to the wise: Your arguments so far have been extremely vague and have only rarely intersected with my specific claims. You may be relying too much on general pattern-matching of vaguely similar argument types, rather than grappling specifically with what I've suggested here.)

Nowhere did I suggest that we should make 'obviously biased arguments.' But what is meant by 'biased' here? If admitting that we believe 'sex-based discrimination exists' and 'sex-based discrimination is bad' immediately makes us 'obviously biased' in the context of any discussion about sex and gender, then we seem to have committed ourselves to a rather untenably austere True Neutrality stance. Would admitting that I believe the Holocaust occurred, and that the Holocaust was bad, similarly call into question my credibility and objectivity as a reasoner? I think that's a bit over-the-top.

The problem for a rationalist would be if your liking or disliking motivated you to change your own perception of reality (for example by intentionally using non-natural categories) in a way that would make you more likely to believe false statements.

You're reifying categories too much. The way we group the world is almost never a completely neutral, interest-free sorting of empirical clusters. The take-away lesson from that isn't 'Despair of ever identifying any of Nature's joints,' nor is it 'Despair of ever being unbiased.' The take-away lesson is to beware of essentializing, i.e., of treating groupings we adopted for convenience as ultimately real. The advice you give is good in broad strokes, but unworkable if it requires that we simply stop having any terms we've defined as we do for purposes of convenience. Our choice of everyday terms is not (nor should it be) in all cases a deep worldly matter, even if the things we assert using those terms is indeed in all cases a deep worldly matter.

For example, if X% of MRAs believe that women should be chained in kitchens, we should want to believe that the number of them who believe so is X. Not X+1. Not X-1.

Yes. Nothing I've said suggests otherwise. Coining a new term specifically for people who care deeply about the welfare of men and think sexism (or female-directed sexism) doesn't exist would not help us better understand the world or its property clusters. Respond to my assertions, not to the most proximate schematic memes my words can be fitted into.

"...because I used a definition that only those who want to chain women in kitchens are the true MRAs."

Yes. Fortunately, my recommendation helps address and correct for precisely this problem.

Feminism is not clearly defined, so neither is anti-feminism.

I defined feminism in the very post you're responding to. I even explicitly defined anti-feminism using exactly this consideration. This again suggests to me that you aren't reading my posts with much care, but are just trying to pattern-match me to the nearest fallacy you can think of. That isn't a good way to persuade someone. Fortunately, I already agree with the good points you've raised, though not their relevance here.

By some people, yes. By other people, no. Should we also use the word "feminist" to mean "a woman who refuses to shave her legs, and talks about it all the time" just because some people use it this way?

If we're continuing the fallacy kick: Slippery slope fallacy. The fact that we shouldn't accept everything as a term of abuse doesn't mean that we shouldn't accept anything as one. 'Neo-nazi,' for instance, seems rightly pejorative. When the referent offends us, and for good reason, the term can acquire a pejorative character; but since it's the world and not the word choice that's causing the bad aftertaste, there's little to be gained via the euphemism treadmill.

This is also a distraction. My original argument corrects the problem of treating 'MRA' as a term of abuse. Whether we should take a principled stance against All Pejorative Words Forever is not my concern; I sidestep the issue when I propose the pragmatic resolution of adopting clearer and more natural terms for the subject matter in question. This conversation will be rather confusing if you continue to make my points for me while thinking they're repudiations of my practices or views.