3 ms·
Your metric is not scale-invariant. It is trivially gamed by malicious actors. It does not reward higher-quality discourse. It rewards overzealous moderators.
by myWindoonn 8y ago
Your metric is not scale-invariant.
It is trivially gamed by malicious actors. It does not reward higher-quality discourse. It rewards overzealous moderators. It rewards edgy or controversial discourse which does not break the guidelines. It rewards users who contribute a bare minimum of information while punishing users who write longer, more informative or interesting posts.
Worst, by what objective measuring device can a post be compared to the guidelines? There are many extremely-subjective suggestions in HN's guidelines alone:
* "Don't say things you wouldn't say face-to-face."
* "Assume good faith."
* "Don't introduce flamewar topics unless you have something genuinely new to say." (Wanna give us a list of flamewar topics, mods?)
* "Please don't use Hacker News primarily for political or ideological battle." As I write this, there is an ideological battle between JS frameworks on the front page.
* "On HN, users should have an identity that others can relate to."
Could you write a script that measures whether an HN post breaks the guidelines?
- AnimalMuppet 8y ago> It is trivially gamed by malicious actors. It does not reward higher-quality discourse. It rewards overzealous moderators. It rewards edgy or controversial discourse which does not break the guidelines. It rewards users who contribute a bare minimum of information while punishing users who write longer, more informative or interesting posts. To the best of my ability to determine, every single thing you said in this paragraph is false. Could you explain how you think some of that works? Because, quite frankly, your comments here look like you just want to complain about HN. > Could you write a script that measures whether an HN post breaks the guidelines? No. I could evaluate a bunch of posts, though, and I suspect that my evaluation of that bunch of posts would not be statistically different from others' evaluation of them (assuming fairness and no malice on all the evaluators' part, of course). It's not automeable, and it's not perfect. But it's also not rocket science.