Agree, and this goes along with Twitter's desire to curate feeds by showing, for example, tweets that were 'liked' by people you follow. By doing this, they broaden the amount of content they can show you and therefore increase the discretion they have in surfacing content they want to spread. (It's also super annoying — I follow someone so I can see what they tweet and RT. I do not care one bit about the things they like.)

Hidden downvotes are the flip side of the coin — instead of being used to boost the spread of certain tweets, they can be used as an excuse for why certain information is hidden.

Basically, these changes remind me of p-hacking. As more variables are added, you can figure out ways to structure your algorithm so that you promote the information that you already wanted to promote and demote the information that you wanted to hide. This could be for profit-seeking reasons, political reasons, or other reasons. In any event, it decreases transparency and reduces trust in the systems.

edit: added last paragraph

> Hidden downvotes are a terrible idea

First and foremost, I agree that this will probably end poorly.

However, I could see this being a good idea in the right hands, because it _could_ be used as _part_ of a more comprehensive auto-mod tools, including self-shadow-banning.

Imagine you had a human reviewing highly ratioed comments, providing a final judgment on "is this comment actually unhelpful, or is it just being brigaded?". (And let's imagine one judgment option is "unclear, do nothing".) You could then keep a per-user metric of their (mis)alignment with that judgment across all their votes. That metric could then be used to weight the votes of that user.

High alignment then classifies that user as a "trustworthy unpaid mod" behind the scenes, which can then be used as a signal to determine future human reviews. Similarly, if someone has an absolutely stellar alignment, you can use the occasional disagreements to do meta-mod reviews of your mods' judgments. If you go really wild and draw some correlation between mis-votes and hashtag/topic/channel/whatever, then you can use that.

Low alignment could then be used to devalue votes from that user, leading to a slow shadow ban, and potentially even other more visible effects.

All of which can be done regardless of the visibility of the votes, of course. But hiding "down" votes should help to reduce the pile-on effects of brigading ("a ton of other people downvoted it, so I am safe in the crowd") and thus the temptations leading to slowly shadow banning yourself. But unlike completely taking away down votes, like YouTube, you leave enough rope for bad actors to hang themselves.

It's equivalent to the system Slashdot has had for decades (IIRC).

Do I trust Twitter to put in place something with that amount of utility, feedback, and nuance? Nope. And even if they did, they'd throw away the value of having invisible scores for the opportunity to add "gold checkmarks" or some other BS.

I disagree that hidden downvotes are generally a terrible idea. Downvotes should probably usually be hidden and used as a measure of whether to surface information - except when that information is factual. Twitter contains absolutely nothing except opinions and thus visible downvotes are useless there.

There are other platforms (like YouTube) where downvotes make a lot of sense because users post guides and tutorials where the downvote ratio is a good indicator of how accurate or useful that guide is.