I very often will find a post on HN where the conversation forked in numerous unrelated directions: one thread taking about the politics, one talking about the economics, one talking about the font used on the website, and another angry that the post got enough upvotes to be on the front page in the first place.<p>I consider it of immense value that I can skip most of the content I am not interested, and concentrate on the parts in which I am. Further, I will contend that it would be impossible to even have that many interesting discussions were every participant forced to struggle through all the other parts in a massive flat discussion.<p>Of course, you can then argue "let's have multiple flat discussions", at which point you start to see something more like a typical web forum: with categories, forums/topics (I will maintain "topic" going forward; to be clear, by "categories" I mean the section headings on the list of topics you often see), specific threads, and then linear posts within the thread.<p>But, if you think about how that maps to a site like HN, you find that the part with the flat posts isn't analogous to the comments on an article: the are more akin to individual thread trees, each one diverting off to talk about the politics, economics, etc. of the overall topic (the link). Now, the article does seem to realize this, as it explicitly is mentioned that capping the thread depth has value, but doesn't seem to understand that that's the world that most of the systems that he's operating in already have.<p>Take StackOverflow as an example: he says there is one level, but there are actually a bunch, from the site as a whole, to the individual interest areas, to the level of individual subjects (which are modelled as a DAG due to being structured with tags, but are of course used in the field as the next level in the tree, as that's how humans can conceptualize it) to questions to answers to individual comments on answers (where it stops).<p>That's a lot of levels of depth, and it lets the site help you weed out all of the stuff you care about from the stuff you don't. That depth was important: if you came to the site and you saw all of the questions at the same time, you'd be frustrated; if you came to the site and saw all of the comments, it would be worthless.<p>Now, go back to HN and attempt to remap that kind of depth: we have links, but there really are lots of sub areas that people like to talk about with regards to those links; as I mentioned: politics, morality, design, alternatives... and each of these is really a fairly high-level goal that tends to get rapidly paired down to "what you actually wanted to discuss".<p>It doesn't <i>feel</i> like these levels are occurring if you concentrate on the schema, but if you examine how the site is used they are clearly there; the exact boundaries, though, tend to get blurred depending on how many people are participating, what kind of article it is, etc.: there isn't a hard/fast set of rules like on StackOverflow, but the site serves a wider set of purposes.<p>Of course, when you get down in the trenches, things can get confusing. That's the only place this article has any meat: however, to turn this into a David v. Goliath "Discussions: Flat or Threaded?" kind of topic, claiming "Web Discussions: Flat by Design" misses the essential complexity of this field, throwing out all of the fascinating parts of discussion communities and bordering on linkbait :(.<p>Further, it then ignores the ways in which Hacker News attempts to do this (the reply delay limit that kicks in the deeper you get down in a discussion tree), and doesn't offer any enlightening solutions. That said, this is the same author (Jeff Atwood) who wrote a massive tirade about discussion systems in 2009 called "The Value of Downvoting, or, How Hacker News Gets It Wrong" without realizing that HN actually supported downvotes, so I'm not certain what I should expect here.<p>(Oh, and the idea that Twitter somehow makes it easy to follow conversations because it attempts to group replies back/forth is kind of ludicrous if you've attempted to dig back through a conversation on the site that involved more than four people after the fact: the mechanism often doesn't work, and it fails to deal with how Twitter is actually insanely NON-linear to the point where there is often a cacophony of discussion going in every direction at once by people who may or may not even be following other people who are involved in the same discussion... how that model--the least linear model in existence today--somehow proves that any linearity at all is valuable is confusing to me ;P.)