Unfortunately - especially when dealing with non-technical ideas - some of these "falsehoods" veer into opinion territory. It's an interesting resource but perhaps the title of it is a little egoistic/presumptuous in that regard.
You just inspired an idea from me regarding HN comments. I think Slashdot still has one of the most elegant commenting systems around, even if the quality of the comments (due to the site's ownership) is worse. I would love to tag HN comments with certain attributes and be able to filter them out. For example, this comment and the parent could be tagged "meta-discussion". Then those who want to discuss the content of…
The simplification to up and down hurts discussion on Reddit (and probably on hacker news too) because it promotes widely comprehended and accepted content. This is why on Reddit pictures do so well unless they are banned, they are very quick to review and choose to up or down vote. However a 10k document will be reviewed by a lot less people, it could fit exceptionally well with the content and get 100% upvotes but be completely buried by all the pictures because a lot less people had time to view it and upvote it.
There is a deeper challenge here with how comments and upvoting and content filtering via users is done. It isn't just the lack of information from tags but also once a document is reviewed will an upvote occur. Much of Reddit shows users will widely upvote on the basis of the title of say news without ever reading it. The longer the article is the less people will return to Reddit to upvote it. Both HN and Reddit promote popularity over fit because low numbers of upvotes on content equates to low engagement and that isn't valuable to them, but it might be the singularly most important piece of content for that audience its just long and badly titled.
There are a bunch of unsolved issues that Reddit and HN have surfaced around the model for content and comment voting and where these two sites differ is mostly only in the rules to try and mitigate the problem with the chosen model. The problem is only more complicated models can improve it and the moment you do that you potentially lose customers, but at the same time you potentially don't want those customers either.
That is my unfinished thought chain on this.