I think that it's for two reasons.
1. HackerNews originally started as a way to test Arc. It is still very much an 'experiment'. Since features are changed on the fly as part of the experiment (think orange names), it doesn't make that much sense to document them.
2. PG wishes to avoid an 'Eternal September' on HN. One way of doing this is to model the site directly towards a subset of users (hackers), and create barriers to entry/discouragements of use to the others. The design is simple, and a lot of the site's features are left undocumented. (user profile options, karma milestones for various abilities like poll-creation and downvoting) For those features, a 'hacker' would likely either a) ignore the features if the site already provided proper functionality and they were not relevant, b) be able to figure out what they do based on the names and some quick testing/playing with them, or c) be able to figure this out by Googling/searching.
They're described here, for example: http://ycombinator.com/newsnews.html
>Like email, social news sites can be dangerously addictive. So the latest version of Hacker News has a feature to let you limit your use of the site. There are three new fields in your profile, noprocrast, maxvisit, and minaway. (You can edit your profile by clicking on your username.) Noprocrast is turned off by default. If you turn it on by setting it to "yes," you'll only be allowed to visit the site for maxvisit minutes at a time, with gaps of minaway minutes in between. The defaults are 20 and 180, which would let you view the site for 20 minutes at a time, and then not allow you back in for 3 hours. You can override noprocrast if you want, in which case your visit clock starts over at zero.
> The reason for the distinction between killing and deleting is to avoid accusations of censorship. Anyone who does want to see the stuff killed by the editors can do it by setting showdead to yes in his profile. But it seemed right to offer submitters a more thorough sort of deletion for their own stuff.
So, basically, what I'm saying is that there's no reason to document or describe these features in a clearer manner than how they are currently documented.