Hacker News new | past | comments | ask | show | jobs | submit login
Pending Comments Update
152 points by pg on Mar 27, 2014 | hide | past | web | favorite | 126 comments
Since some people were alarmed about pending comments, I thought I should explain what's going on with them. Sorry for the delay, but between implementing them and now I've had to help get the current batch of startups through Demo Day.

To recap, pending comments are a feature we hope might mitigate or eliminate the nasty comment threads that sometimes occur on HN. It essentially turns HN into the sort of hard-moderated forum where comments have to be approved by moderators before being seen, except to make it more democratic (and less work for us), any user over some karma threshold can approve comments.

Since the implementation touches many parts of the code, and I am at this stage still the one who understands the HN code best, I was encouraged to write it before I left.

The plan from the beginning was to make it possible to turn pending comments on or off per thread. Since the simplest v1 was to have pending comments sitewide, I released it in that form first. That code didn't break, fortunately. The next day I finished making it work per thread. Or more precisely per item tree; a moderator can turn on pending comments for anything from the whole site to an individual comment.

When I announced pending comments I said the threshold for endorsing comments would be 1000 karma. Some people have been alarmed by that. But 1000 was nothing more than a plausible initial value of a variable meant to be tuned by the moderator. In fact, the threshold never was 1000 even at first; I set it to 500 in the first release. Maybe it will end up being 50. The only way to figure out what works is empirically.

Indeed, the only way to figure out if pending comments will work at all is empirically. The plan is for the moderator to experiment with turning pending comments on for individual threads. If that works, maybe after tuning the parameters he'll gradually expand to more and maybe eventually just turn on the feature sitewide. Or maybe that will never be necessary. Or maybe software will decide when to turn on pending comments.

Nor does the threshold for endorsing comments have to be karma. That was the obvious choice for a v1, but it would be easy to incorporate or substitute other things like account age or average comment score, or even introduce randomness.

It's impossible to predict exactly how pending comments will end up working, but one thing you can safely predict is that whatever happens with them won't ruin HN. The main moderator is extremely sensitive to the state of HN, and if something he did was making the site worse, he'd be quick to notice.

I'm concerned that giving comment-approval power to people with x amount of karma will reinforce many of the existing problematic aspects of HN instead of opening it up to better discussion. It means that people with currently-underrepresented voices will have to "play nice" with the current mainstream of HN in order to be represented on the site.

For example, discussions here already have a pattern of being mostly men, and most people with significant amounts of karma are men - I don't know the data here, but I don't think that's a controversial observation. If you're a woman who is new to HN and trying to explain your work experience as a woman in a relevant thread, right now you can give commenting an honest effort and know that everyone can at least read your words and consider them. But under pending comments, instead women will have to write comments that men approve before those women's comments are even visible.

The pending comments system seems worryingly likely to reinforce HN's existing systemic biases in silent/hidden ways that will be hard to analyze and improve after implementation.

Edit to suggest an improvement instead of just criticism: I've been moderating forums and IRC professionally for six years (for del.icio.us and now for Cydia), and to change the culture here, I'd first try expanding the community guidelines with much more detail and several specific examples of unacceptable comments. It clearly has not been enough to only say "be civil", no name calling, and "no classic flamewar topics" - it's a good start, but vague and incomplete. Expanded guidelines would go along with clearly-indicated removals of unacceptable comments to show that the new guidelines are serious and not just suggestions. I would also try implementing a Metafilter-style flagging system: make the "flag" button consistently visible, with a "pick a reason to flag" menu that has one option per rule category (http://i.imgur.com/Aw03Tl2.png). This serves as an integrated (and "just in time") reminder of the rules, with the bonus of flag counts helping moderators find problem spots.

Don't worry, the moderator is very sensitive to this issue. There are going to be fairly explicit standards about what should be endorsed and what shouldn't. Roughly things shouldn't be endorsed if they are gratuitously uncivil or content-free, but there will presumably be a fully detailed version of that policy. And there will probably also be a version of showdead for pending comments. (Already anyone who can endorse can see them at /pending.)

We're hoping that this feature will make the site more hospitable to women. That wasn't the only reason for doing it, because users get stupid and/or nasty about lots of topics (or any topic, if they happen to start insulting one another), but certainly some of the most cringe-worthy threads we've cringed over have involved female programmers. So it would be hard to imagine a version of "working" for which pending comments working would not thereby make HN more welcoming to women.

I'm glad this is being considered, but I'm worried that it's easy for non-majority opinions to be interpreted as "gratuitously uncivil", especially on emotion-generating topics.

For example, if a person has had a bad experience being harassed by a coworker and writes a clear and honest comment about this as a problem, that can be easily interpreted (such as by a person unfamiliar with getting harassed) as an uncivil thing to say about your coworker. There is a known pattern of "tone policing" - that people have a pattern of asking less-privileged people to be "not so angry-sounding" when talking about something problematic or upsetting from their lives. I think of this pattern as a defense mechanism to protect yourself from understanding somebody else's pain (especially if it challenges something about your status), and it seems likely to be a bias here, even an unconscious bias.

Instead, I'd love to see explicit standards for comments on a visible community guidelines page, and an efficient flagging system that reinforces these guidelines.

The sort of incivility we're worried about is the more explicit type where someone replies to a comment with "You are an idiot. Don't you realize that x y z?" when they could have simply said "x y z." I'm pretty sure if HN cuts down on that type of comment, people who are earnest but upset about something will be net ahead of people they might be arguing with who are merely thoughtless jerks.

So, a lot of people's first introduction to all of this was the comment Sam made in his article "What I've Learned From Female Founders So Far" about how "we're working on something to improve the quality of Hacker News comments". It is thereby not surprising that people are trying to analyze whether this feature will help that specific issue ;P.

FWIW, I think that this is a different problem than the "you are an idiot" problem: I think a lot of the threads that people are finding issue with when it comes to attitudes towards under-represented groups are not (at least entirely) filled with this "explicit type" of negative comment: it is instead the more insidious, implicit type being stared at.

Regardless, I don't see why the endorsement system will somehow work against either of these problems when the voting mechanism hasn't: the comments that start with "you are an idiot, don't you realize that" are currently getting upvotes, so why would we presume they won't also get endorsements? Are high-karma users voting differently?

I would love to see that kind of comment cut out too, but I am concerned that pending comments won't actually achieve that goal - that biases (unconscious and not) will sneak into the patterns of what gets approved.

The moderator is concerned about that too. But if it did start to happen it would be pretty obvious. So it seems worth trying to see if pending comments can be tuned to cut obvious crap without eliminating stuff that's merely controversial.

That sort of thing needs a flag mechanism, not an a priori vetting of each comment. Enough flags = lock replies, a few more flags = hide comment.

Exactly! And really, we already have a flagging system: the downvote. My understanding is that comments on HN should only really be downvoted for the same reasons as pending comments would not be approved. So why not improve on that existing system? When a comment is downvoted, it should ask the reason for the 'flag', with choices representing the various commenting rules. This should help ensure that people are using downvotes correctly, which should allow them to be acted on more strongly. Sufficient downvotes (or sufficient downvotes with the same reason) could hide the comment entirely.

Mmm. This smells a bit like the League of Legends flagging and Tribunal system, which has worked very well for them. I wonder if HN has enough active readers to make a Tribunal work well.

I've decided to start using a throwaway for political discussions...

I would personally approve any comment that discusses a lived experience in concrete terms.

That being said re: tone policing, there is a segment of the social justice crowd on HN that will make inflammatory generalizations about large groups of people. Generally this will take the form of "X happens because of goddamn white cishet men." Inevitably someone will call these comments out and the response will be that it is justified to speak this way about the "oppressor class" and that anyone who "tone polices" these comments is ignoring the "power dynamics" of the situation.

In my opinion these comments do nothing but add fuel to the fire and I will never approve them. They are wrong for the same reasons gross generalizations about minorities are wrong.

And no, I'm not white, but yes, I am a man.

My take on tone policing is it is a mechanism (whether conscious or unconscious) to tell people they shouldn't be as passionate and concerned about something as they are.

As such, it's relative to the tone of the forum (be it HN, tumblr, irl, wherever). If it's acceptable for me to say "goddamn NSA fucking ruining the internet" then it's acceptable for me to say "goddamn cis people who fucking ruin the internet" [1]

There are plenty of people on HN who speak passionately about many things, and who do so with expletives and curses, blunt and punchy statements. Personally I hope they can continue, but either way I hope it is explicitly allowed or discouraged in the guidelines.

Speaking to over-generalisations; it's a hard one. There is undoubtedly a need to talk about the commulative effect of oppressive groups, and there is definitely a need to talk a about subsections of oppressive groups who are actively doing opressive things (again, whether consciously or unconsciously, knowingly or unknowingly).

Unfortunately we haven't yet figured out/settled on the language to use that differentiates between the two that doesn't involve a whole bunch of clauses and caveats. And given a forum where people are permitted to express their passion, extra clauses and caveats weakens the impact. When you are trying to communicate how fucked up a situation is to someone, it's exhausting and derailing to have to continually validate them and say "of course I don't mean everyone here, implicitly from the context I mean everyone who is engaging in this behaviour".

[1] I had to think a bit then for the phrasing where I could most easily argue "of course I don't mean every cis person" without also weakening the statement with extra clauses.

Excellent points. Are there any examples of guidelines you think would be good starting points?

It's interesting to look at the format of the Django Code of Conduct, although the content would be different since this is for a project instead a forum: https://www.djangoproject.com/conduct/ - including detailed explanations of what it considers important, examples of types of problems, a FAQ explaining the code of conduct, a reporting guide, an enforcement manual, and a changelog.

The Flickr Community Guidelines are also interesting in format: https://www.flickr.com/help/guidelines/ - with many very specific details about acceptable and unacceptable behavior.


I agree with your general sentiment. However, it only takes one member of the underrepresented group (with endorsing capability), or any member of a centrist or neutral party, to approve the a comment. (And as of so far, it doesn't seem like pending comments can be disapproved, e.g. by a hostile party.)

Granted, it's true that by nature of being underrepresented, the probability of that one member showing up to approve the comment may be unacceptably low. But I doubt it. In other words, to use your example, as long as the probability is high enough that at least one woman will show up to approve another woman's comment, there shouldn't be a problem.

Personally, I think that this probability is pretty high, though, as pg mentions, this'll have to be determined empirically. For example, it's entirely possible (as discussed in previous threads), that the population of active endorsers are skewed toward certain groups. Or that even a slightly <100% probability of legit comments being seen can ultimately compound systemic biases in the long run.

I think it is very important to not discount the emotional complexity of forcing someone into a position where their comments must be "endorsed" by a group of people they are perceiving as "hostile" to them.

Even if there are people who would endorse the comment, that is not what the person leaving the comment is likely going to be "feeling" in that moment: they are reading a bunch of comments that they are bothered by--ones which none of the other people around seem to be taking serious issue with (which is a key part of the original problem statement)--and those are the people they are going to perceive as being the ones who must endorse their comment: the ones to whom they are effectively submitting themselves for "endorsement".

(Which, frankly, I think is a separate reason why this endorsement system doesn't make sense as a solution to an endemic problem: most of those issuematic threads are filled with users--many of which have high karma--who are reinforcing the negative comments; it is unclear to me why the negative comments aren't going to have an easy time getting endorsements given that they currently don't have a difficult time getting upvotes.)

I thereby feel like even if this feature "worked" (and again, it isn't really clear to me why this would help, given that it isn't like these users are currently being downvoted or flagged out of the conversation) it still might not be an "appropriate" way to solve these underlying problems (which I would claim are inherently messy and emotional).

I agree with you and your parent. If this system goes into place I'll probably spend far more time on HN than I usually do these days -- at least at first -- specifically trying to approve comments made by other women, in the fear that others with enough karma will dismiss their point of view.

However, that won't stop many of the negative comments you mention, posted by and/or endorsed by high-karma members, also showing up in the thread. It only takes one high-karma approver to allow the negative stuff through and we're back to square one, but possibly excluding some underrepresented groups.

I like the idea in general, that is, I like the drive to improve the quality of HN comments. I'm not sure that flipping 'flag' around into 'approve' is the solution, but sadly I don't have any better ideas. Well, other than post emotion analysis and multiple-tiered karma systems (+1 Insightful, anyone?).

I agree completely. It appears that the problem behind the newly announced commenting changes can be simplified to an "innocent until proven guilty" argument. The new system assumes that the users commenting on HN are immediately guilty of offensive comments.

Another potential problem is that people will begin to purposefully tone-police themselves in an effort to appease the gatekeepers and ensure their comments are approved. This will likely result an atmosphere in which moderators and those with high karma create - one in which many unpopular opinions are silenced via unapproved comments.

As someone who is an inactive user, with absolutely no karma, I'm not looking forward to having to jump through hoops in order to have comments approved. Why not wait to see if I'm an actual troll or jerk before deciding that I need to be policed?

I'm concerned that giving comment-approval power to people with x amount of karma will reinforce many of the existing problematic aspects of HN instead of opening it up to better discussion

Given that endorsement is a mechanism which can only reduce the quantity of discussion on HN, it is difficult to imagine that it is designed to open up a discussion to more viewpoints.

I think your choice of "female work experience in technology fields" is interesting because it is one of the few legitimate topics where there isn't widespread agreement on basic foundations around the topic.

Others topics that fall into that category include "The Economics of Bitcoin", "Is Google Evil", "Startup X was bought by large company Y".

There's also a wide range of political & economic topics which are always controversial, but sometimes I question their legitimacy as a topic for HN.

I wonder if these controversial topics should have special handling so far as endorsement (and maybe even voting) on comments goes?

Given that endorsement is a mechanism which can only reduce the quantity of discussion on HN

That seems a mistaken assumption. The presence of existing comments often causes me to refrain from posting comments. Often I've been about to reply to something, then notice someone else has already said substantially the same thing, and as a result don't.

And this is not a contrived example. As I said when I first launched pending comments, if someone says something important in a nasty way, it may well turn out to be safe not to endorse it, because someone else will probably show up and make the same point without nastiness.

Plus if HN gets more civil, it may encourage people to comment who might not have before. I know from my own experience that incivility has decreased the number of comments I make. I often find myself about to say something, then decide I have work to do and don't have time to spend the afternoon fighting, and so don't say it.

Only experience can tell us whether one or both of these factors will increase the number of comments, but we can't assume that pending comments can only reduce discussion.

Often I've been about to reply to something, then notice someone else has already said substantially the same thing, and as a result don't.

Hm. Interesting...

I guess unique insights aren't, generally.

Amusingly, this already happens with dead users as well. I remember one thread where several European users were chiming in with agreement, all dead and invisible to anyone who hadn't set the flag. There was nothing nasty about their posts either, they were just the brutally honest type.

> It means that people with currently-underrepresented voices will have to "play nice" with the current mainstream of HN in order to be represented on the site.

Surely there's more to talk about on HN than one aspect of someone's life that puts them out of the HN mainstream?

We should be well rounded enough to be able to participate in some conversation on the site without swimming against the tide. That's not "playing nice", that's just not going against the grain in every single thread.

I love the idea of these guidelines and hope that if they happen, they'll be public! Thanks for outlining the idea so clearly.

I don't think it will be a problem in practice, both because women will be able to endorse comments and because during the test period people endorsed most comments. The only comments that weren't endorsed were troll comments.

The most recent test period was brief (just a couple hours) so it isn't representative of how it will work out in practice, but that's why I think we should turn it on for 24 hours and see what happens.

EDIT: On reflection, I apologize for how dismissive my comment must've sounded. It wasn't my intention.

> The only comments that weren't endorsed were troll comments.

I have always browsed HN with showdead on; I occasionally see dead "troll comments" and I hardly ever see not-dead "troll comments" that are not at least slightly downvoted. It would seem to me that the downvote and flag mechanisms are already sufficient, therefore, against this class of "troll comments".

The real question is whether serious comments--maybe even ones some users would consider "highly informative"--that also happen to have "negative affect" (or, even more difficultly, which are subtly dismissive) will actually have a difficult time getting the endorsements required to be posted on the site.

(Separately, there is then the issue as to whether one's own biases alter what they even consider "negative affect". There are some other comments on this thread that talk about "tone policing" which point out some subtle issues in how people perceive the tone of others when they are in disagreements.)

I also will note that even if getting endorsements is not a problem, it seems important to not discount the emotional complexity of forcing people to submit themselves for "endorsement" to the group of people they may already feel are being dismissive of their viewpoint. I wrote more on this here:


I just wanted to thank you for the comment. You make some good points.

I don't see that a brief test period is much of a model for sustained use. I think people will get tired of endorsing any strangers and mostly just look for friends' postings to let in.

I feel my previous comment [1] on the design of this feature is still valid. The word "endorse" carries the wrong connotations and increases the risk of subjectivity and groupthink. A better link label might be "tolerable".


IIRC that name was chosen by the moderator, but it will be easy enough for him to change if he wants.

$0.02 regarding the name of the feature, which I think anchors too much in the "censorship/free speech" principle discussion when it seems you have much more tactical goals (moderator workload, foster specific topics, etc.).

Consider renaming the feature from "pending comments" to "civility filter." In prose one would say Moderators now can require civility via special-purpose moderation on specific threads.

Similarly, consider changing the "[pending]" text to "[Pending civility check]" and changing the "endorse" link to "Affirm civility" (or just "Civil" to keep with one-word nav elements).


"All speech must be endorsed" --> "Bad Thing"

"All speech must be civil" --> "Good Thing"

could it just be that if the 'privileged user' upvoted or downvoted it, it would be considered an endorsement whereas not touching it would leave it pending?

But that's the problem. Upvoting and approving/moderating are different things. Upvoting is a subjective exercise, whereby you signal substantive agreement. Approving/moderating is an objective exercise, whereby you signal that a comment meets minimal community standards of tone and non-triviality. Conflating those will inhibit dissenting opinion.

That said, the direction you're implying might be an improvement. Rather than a separate Pending Comments list, mod-level users would be able to see pending comments in the context of the parent item/tree, but with some kind of adornment or different color. Pending comments will also have a 'tolerable'/'civil'/'endorse' link next to it, but no up/down vote buttons. Only after it's been deemed tolerable enough, will a pending comment become live/approved and have up/down buttons.

"Endorse" means "I support this item, and I put my name behind that support". It is not the right term for this task.

Indeed, that seems a much more sensible interaction-model to me. Take the absolute number of upvotes as the "endorsement value" of a post. Make all posts start at zero "endorsement", and require them to reach some N before they appear.

Prior restraint solutions sit very badly with me: there's a reason that "prior restraint on speech" is near or at the top of the list of First Amendment no-nos (yes, the First Amendment applies to US government entities rather than private entities; the lesson is still valuable). It's a gratuitously bad idea to enable the prior restraint feature sitewide and by default. Besides the principle of prior restraint, that has all the hallmarks of being a feature that scales very poorly because it requires human involvement. Requiring human involvement and discretion is the point, so I'm not sure how one could eliminate that. Further, the part of it that say "if a user has a comment in the pending-approval status, they may not post new comments at all," seems very, very likely to have adverse consequences of large scope. People participate in multiple discussions over time and change their minds about them and behave differently in different contexts in response to different prompts. Holding all of their participation hostage to something they've said in the most contentious available context, seems like a great way to exert a chilling effect on participation as a whole.

I strongly, strongly object to and oppose turning on the prior-restraint feature sitewide by default, and especially to the "if a user has a comment in the pending-approval status, they may not post new comments at all" part of the feature.

That said, the history of moderating large discussions, especially in digital media, demonstrates that giving moderators programmatic tools to enforce their judgment is a Good Thing. So if you consider the proposed feature as "Allow moderators to say that any reply to a comment in the tree rooted at Comment Foo, must be human-approved before going live," sounds like a great tool for a moderator to have at their disposal. It is limited in scope and its effects can be judged and known. I think that's a reasonable thing to add, with the caveat that that shouldn't include the "users who have pending comments can't create other comments" part.

If we didn't have prior restraints on people speaking about places where you could buy cheap Ugg boots, HN would be so overrun with spam that it would be unusable.

That's a good observation, but doesn't really get at the point the parent was making. If I may continue the example of First Amendment rights, consider the well-founded prohibition on yelling 'fire' in a crowded theater, which seems more or less in line with an ad for boots on the Internet.

Could you address the issue of only allowing a user to post one comment at a time? I think the parent made a good point on the potential chilling effect that may have on parallel comment threads.

That's a variable the moderator can tune as well. We actually ended up setting the limit at 5 when we launched, which I don't think inconvenienced anyone.

I didn't know that part, thank you for bringing that up again (I imagine you've had to repeat it). That reduces my worry level a lot.

Yeah, I think we're all pretty clear that the pending-comments feature is a different thing from conventional anti-spam measures, which have fairly reliable programmatic solutions in this kind of scenario.

My instinct is the hidden scores are the bigger problem.

There are so many posts where the top comment seems relevant and interesting, and I up-vote it blindly, even though it already has like 50 up-votes, and the more interesting discussion is buried. As a 1000+ karma user, I'll make the same mistake, under the new system.

Hiding the scores was a popular change because comments that would normally generate +5, all the sudden were getting +60. It made everyone feel better. None complained, but the site got worse.

I really miss the old days, when I could check the comments page, and the top comment would be a new, interesting, nuanced, balanced, comprehensive, perspective on the issue. Now the top comment is the first popular opinion followed by a bunch of trolling.

Can we bring back showing scores for a week? Maybe as a april fools joke.

Why not put a new mental rule in place: Do not upvote top level comments unless they are extremely new <10 minutes ago. Then you get the benefits of hidden scores and you do not contribute to burying the interesting discussions. It is also worth rethinking blindly up-voting. How reasonable is it to expect a comment scoring system to succeed when users blindly upvote things?

That is a good suggestion for me, but it doesn't fix anything.

I want to read interesting, comprehensive, and nuanced comments. My voting won't change that.

The reason HN moved to blind voting was because there was ugly complaining about high karma users. HN should have hidden the usernames rather than the votes. The Economist works this way.

My new theory is that HN should exploit the Hawthorne effect. Basically change the rules constantly to get the best results.

An alternative could be that the impact of an upvote is inversly proportional to its position on the page. You'd probably have to combine it with other metrics like age.

For what it's worth, I disagree with pre-moderation. I'm not frequent enough to ever have hundreds of karma points, and it's very rare for me to see a "nasty comment".

What's wrong with reactive moderation? Or the existing system where unpopular comments fade away literally?

Pre-moderation kills the flow of conversation. It's a massive bottleneck, is more work for moderators, and for what benefit? Sanitation?

There's nothing wrong with a bit of grit. We're not all sensitive flowers, reduced to tears at the sight of a nasty spray by Mr X. I hardly see that anyway, I'm really suprised to see pending comments even being contemplated here.

Why does the site reward high karma rather than high average karma?

I don't have any experiences miderating a forum, but intuitively it seems to me that those who spend a lot of time contributing low quality posts gain more moderation privileges than those who are either too busy to do so or who spend time writing a few quality posts.

Mostly because high-quality posts hidden in low-traffic subthreads "look like" low-quality posts under such a system. People who go out of their way to help some individual without the expectation of a status-gain would be punished, while sycophants would be rewarded. This is the opposite of the incentives you'd want.

I worked on the moderation team at Quora, and though you and your moderators know HN better, I'd be happy to share some of what I learned there with respect to hairy comment threads. One of the main focuses of my work was pragmatic solutions that reduce the necessity of human (moderator) intervention without eliminating it entirely.

Some concepts are controversial, but the resulting discussion hashed out from them can often be more insightful and meaningful than the original article.

Some concepts are odd on the first take, and it might take a child comment to flesh out what may be a very good but difficult to understand idea. And if most people do not understand the point right away, they may not endorse it.

Some people are creative with their use of language, or are non-native English speakers, and may not be able to effectively articulate what may well be a very interesting idea or important concept that adds to the conversation. Alone such a comment may not be useful, but child comments exploring it further may yield some excellent discussion.

Some comments may seem flippant, but solicitations for back-story reveal what the original commenter was really getting at.

One of the best things about HN is the comments, and it follows that one of the best things about any particular comment is a another comment reply.


There's a problem you can peek at in academia, it doesn't have a name, but it's the reason that 125 Harvard students were caught cheating a few years ago. The problem isn't that the students were cheating - Harvard students are not dumb and I doubt they're particularly lazy. The problem was that grading had become such a low priority for a professor that the take-home final answers, that only a gradable subset of possible answers were really accepted for any question, so that grading could be done more on the number of citations than how effectively you communicated an answer. Implicitly and usually explicitly it's understood that you cannot just give an original answer, you must give an answer relating to something that was taught in the course. This does not optimize for interesting or insightful answers, it optimizes for regurgitation. Originality becomes dangerous, since it demand deviation.

I fear we will run into a similar issue here. People will tailor their comments to please the endorsers that be. We will be turning inward mentally, and we will never know how much.

And for what? It's true that sometimes there's a controversial comment and it ends in a 60-comment emotive goose-chase.

Is that such a bad price to pay?

> Indeed, the only way to figure out if pending comments will work at all is empirically.

We cannot know what interesting conversations and discussions might be lost. No comment is an island, and I think this concept ignores that.


I do hope that if you turn it on it is as laser-focused as possible. The smaller the unit you damage though, the more you put the the ability to shut down/delay conversation in what might be an otherwise interesting thread into the hands of one/a few people.


[1] There are probably other problems at play, such as Harvard students being pressed for time and giving low priority to an intro to government class. That Harvard has an intro to government class is also probably a (separate) issue.

People will tailor their comments to please the endorsers that be. We will be turning inward mentally, and we will never know how much.

This might be the goal.

YC has been fairly heavily criticized for comments on certain topics which don't toe the party line. Mitigating such comments may be beneficial for YC (the business) in order to avoid negative publicity and attacks from powerful adversaries looking to score a symbolic victory [1]. Remember "The [mis]Information" interview with PG?

Discussions of such topics are mostly a distraction from what most of us come here for anyway, so it likely would not hurt HN much, particularly if it is turned on in sensitive threads only. In my view banning such topics would be a better solution, but I'm aware that HN has been attacked by critics for the soft ban (which pushes such stories down quickly), so this might be the best compromise for the business.

(It's important to remember that YC is a business and HN is just advertising for the business.)

[1] YC is a good target for such a symbolic victory since they are weak supporters of the cause and are unlikely to fight back too strongly but are also highly visible.

If our goal were to decrease criticism of YC, this wouldn't work. In fact, it wouldn't work as a way to eliminate any specific type of critical comment, because all it takes to make a comment visible is a small number of the many people who can endorse comments. To ensure a comment would be suppressed, the endorsers would have to be unanimously opposed to it, and it's hard to think of opinions that many HN users are unanimous about. Certainly an uncritical admiration for YC is not one of them.

What you say about YC being a good target for a symbolic victory is very insightful though. That is a real conundrum. I think it explains why any company over a certain size tends to express itself pretty blandly.

I don't believe the goal was to decrease criticism of YC on HN, I believe that part of the goal might be to reduce HN comments which have provided fodder for criticism of YC in other media.

Being specific, assorted feminist critics of YC/technology have complained strongly about HN comments being skeptical, not toeing the party line, etc. In the mainstream media, it's not too hard to blame YC/you/Altman for views that commenters here might express.

Similarly, such critics have also complained about cursing/mysogyny/racism in github repos and implicitly attributed them to Github, Inc. (For example, I don't work for github, but github could be blamed for the fact that "I have no fucking clue why this speeds things up": https://github.com/scalanlp/breeze/blame/master/src/main/sca... )

You suggested in some previous discussion on this that people who use the endorsing power incorrectly (I don't believe this was clarified but I haven't followed this endorsement thing closely [1]) might lose the power to endorse. Combining endorsement with filtering the "bad" endorsers could certainly work to steer the sort of comments that are endorsed. Rather than filtering all comments one need only filter the much smaller group of endorsers.

If you say this is not the goal and the plan is not to do this, I will drastically reduce my estimate of the probability that this is the goal. But at the moment I don't find it implausible.

[1] Most of the time I waste here is waiting for computations to finish and code to compile. But I imagine if I stopped commenting here I might be more productive, so I'm not personally strongly opposed to changes that might push me away from this site.

Oops, I get it now. It should decrease the sort of comment people attack HN for, but that wasn't the purpose of it. Those comments are just a subset of the more general problem of users saying mean and/or stupid things about any topic.

The real motivation for pending comments, incidentally, was simply that I found, as a user, that I didn't like reading comment threads as much as I used to. HN was an instance of me following the advice I often give founders: to build something you yourself want.

maybe you liked them better when scores were displayed? have you thought about displaying them again to see if that changes your perceived quality?

Wow, this is very well written. I apologize that I am not nearly as eloquent a writer as you are.

I think that your theory in general sounds good, but the problem is that you assume that the emotive comments can be easily ignored. I have seen too many cases where the number one comment is a simple casual dismissal of a new product or idea. This problem is compounded by the layout of this site such that a -1000 scored comment will still appear above the number two comment if it is a reply to the number one comment. Therefore, the amount of attention given to any comment below the number one comment (that is not a child) is very very minimal. To even get to the number two comment requires scrolling past very large amounts of replies and replies to replies.

This is not to say that I think the pending system will fix all these flaws as I dont know. I am waiting to see how it will play out. My point here is simply that there is a need.

> the amount of attention given to any comment below the number one comment (that is not a child) is very very minimal

This seems more like a UX issue. Maybe folding replies after n levels will fix that problem?

Oooh, or maybe folding when the response's value (value meaning the score + initial + decay over time, etc) is below some fraction of the parent it gets folded out.

The situation you describe is the exact reason I read the comments less and less these days. I still like the content that floats to the Hacker News front page, but the fact that the term, "top dismissive comment" exists, and is damn near ubiquitous with the site is a pretty big problem. On Hacker News, rather than simply writing "First!!!" in the comment box, people seemingly write why OP is an idiot for trying, how X was solved years ago, and nothing their start up solves things better.

It's a sad state and annoying of affairs.

> I have seen too many cases where the number one comment is a simple casual dismissal of a new product or idea.

The comment must be agreeable to a large crowd and that is why it is number one. "Pending comments" wouldn't solve it at all.

What it may solve is the moderation of comments which are not voted number one, but still hog your attention span by being children of high ranking comment.

However, I don't think "pending comments" is the right solution. A simpler and more effective solution would be folding of low ranking comments (ala Slashdot).

But comments do not need many people to endorse them, they only need one.

That means all but the worst comments will be approved.

I'd rather have people tailor their comment to please endorsors than tailor it to please the site wide community.

> Indeed, the only way to figure out if pending comments will work at all is empirically. The plan is for the moderator to experiment with turning pending comments on for individual threads. If that works...

If this is going to be empirical, then what is the measurement for 'it works' and how quickly do you expect to determine efficacy of the endorsement system?

Since all threads are different it would be really cool if there could be some A/B tests. Every thread or day you round-robin which users have to go through the pending system. The pending and non-pending groups see completely different sets of comments. Then we could see what type of effect it had on comments. Does it raise the average karma per comment? How does it impact the number of comments? etc...

Probably too much work to implement but it'd be really cool to see.

Did you give any thought to the "showpending" idea posted in the previous discussion? Account config for those of us who don't mind reading the full, uncensored firehose.

> Nor does the threshold for endorsing comments have to be karma. That was the obvious choice for a v1, but it would be easy to incorporate or substitute other things like account age or average comment score, or even introduce randomness.

I don't have a sense anymore for the quality of comments on Slashdot but Slashdot has had a form of random moderation for long time. I do wonder how the goals and ideals are (were?) different between HN and Slashdot and how these differences manifest themselves in moderation policy.

A good overview of Slashdot's history and though process on moderation can be found here: http://slashdot.org/moderation.shtml

Slashdot is a lot better, however, because it is user customizable. Comments are rated on various attributes like if they are funny or insightful, then users can adjust the scoring of posts based on their preferences. It is also social - you can adjust posts of people you mark friends or foes. It is also policed - meta moderation moderated the moderators. As someone who has run internet forums and picture sharing sites before, I know the problem isn't finding moderators, the problem is finding moderators who don't get ban happy and enforce rules just for the fun of enforcing.

I fully support making discourse more productive and agree the problem exists, but the idea of giving a small subgroup the power to determine whose voices can be heard is hardly "more democratic." I do agree it is less work for you.

You can set the value of the karma threshold automatically, to some multiple of the standard deviation of karma for all of HN. So you won't have to constantly revise it.

I assume the distribution of karma here is exponential, so take the distribution's parameter lambda to be the highest value of karma, and set the threshold to half of the inverse of lambda. This page explains it better: http://www.phy.ornl.gov/csep/mc/node18.html

I like the idea of a hidden karma threshold, because it prevents any obvious attacks like "get to 50 karma and then approve bad comments".

I am confused about the "main moderator." When pg says "main moderator" is he referring to a specifc person other than himself, himself in the third person or a functional role (possibly rotating like "pager duty")? Elsewhere in this discussion he says "IIRC that name was chosen by the moderator, but it will be easy enough for him to change if he wants." The first mention of endorsing comments I can find is in pg's initial announcement of the pending comments change.

> I am at this stage still the one who understands the HN code best

Hopefully not sounding offensive, but objectively: this is probably because the HN codebase is a complex ball of mud written in an obscure, original language, further obfuscated by a rabbit-hole filled journey of macros/DSLs.

Ok, maybe that was a little offensive. And, I'm not one to talk about language choice (much less language design), as I use PHP heavily every day. But if PG is going to back off from the code ("I was encouraged to write it before I left"), then maybe it's time for a rewrite in a more familiar language, and maybe it's time to open source it properly and actively. While I think the current Arc situation epitomizes the hacker culture, I disagree that an open source rewrite in a more popular language could not also function in the same manner (if not more so).

Edit: "ball of mud" isn't necessarily all negative. I'm borrowing from a supposed Joel Moses quote: "APL is like a beautiful diamond - flawless, beautifully symmetrical. But you can't add anything to it. If you try to glue on another diamond, you don't get a bigger diamond. Lisp is like a ball of mud. Add more and it's still a ball of mud - it still looks like Lisp."

It's not nearly as bad as you make it sound. Most people just don't give it a chance by forcing themselves to work exclusively in it for a week while trying to implement features, etc.

The ice skating analogy is pretty apt. You'll fall and hurt yourself a few times, but once you get the hang of it you get around much faster.

That still doesn't address the apparent issue of PG being the only person capable of implementing this feature.

It seems like a lot of flame wars could be prevented without a pending comment feature, or at least a watered down one.

If a comment is particularly offensive/offtopic, it is already downvoted, eventually greyingout. To prevent flame wars, the reply button could simply be disabled on those comments, or in comment threads with high vote volatility (high numbers of up and down votes on parent comment). Vote volatility could indicate a flamebait worthy topic, and the reply depth could be limited based on the volatility.

Similarly the reply button could be pulled when two users are chaining long reply threads alone.

Perhaps with higher levels of karma, the reply thread depth is relaxed for those particular users.

Modifying reply access should prevent flame wars, but if there is still a desire to police the content more, then I think a "flag" button is more effective. If enough people flag your post (regardless of your karma), it could be placed in a type of quarantine area and then be pending until someone with the karma privileges decides to moderate it back in or not. I believe this is what reddit does, though the comments must be approved by mods rather than users.

What about connecting pending comments (or posts) to an initial upvote?

- Anything posted by K- users is only visible to K+ users and the K- guy who posted it.

- Nobody can comment on a K- post/comment until it gets the first upvote by a K+ user.

- Once a K+ user upvotes the K- post, it clears for everyone to read and respond, even if it drops under 0 later.

I imagine this could easily reduce the most obvious friction.

I think this feature could have a very positive impact on HN.

What happens if somebody replies to my comment, but I don't get around to responding to that person for a few days. Now there are hardly any eyeballs hitting that thread, so probably a low chance of my comment getting approved by somebody. Is it possible that my reply will never get to be seen?

The solution may be to have pending comments only enforced on recent threads. That would be about 30 characters of additional code.

I haven't actually been on HN when pending was turned on, but there's one question I have:

How do I tell when pending is on? As far as I know, it will be in the top bar -- if you qualify to approve comments. Does it appear if I can't approve comments? How else could I tell whether pending is on, or if I am able to approve comments?

You can tell when a comment of your own is pending because the text is preceded by [pending]. You can tell when you can endorse a comment because there is link to do so. The link is currently small and gray, so it should probably be made a brighter color or replaced by a button or something.

First Question:

Does an endorsement also provide one karma point in favor of the user that posted the comment?

If yes, why not just bind endorsements to the upvote link? So that it becomes a natural behavior, in keeping with existing norms on this site. (perhaps emphasize the upvote button with color-coding, to indicate that the upvote also endorses the comment)

If not, why not? Why shouldn't an endorsement count as an upvote that provides karma?

Second Question:

Do pending comments also apply to users with endorsement powers, or are their comments always visible by default, due to their endorsement privileges? I'm assuming not.

What if a privileged user were also required to be endorsed by a fellow karma-enabled user who was not the author of the comment under review, as a form of peer review? In other words, just because you have enough karma to review someone else's comments, still does not guarantee that your own judgement of your personal statements might hold universal, objective merit.

I think the suggestion of tying the upvote to an endorsement may be a bad idea. It would be conflating the fact that you want the comment to be seen with the possibility that you might agree with it. The comment could be tangential/adversarial to your opinion, and so you might not upvote it, however it could still add to the discussion, and hence you would endorse it.

To answer your second question I believe pg has stated that only users with over 10000 karma have their comments automatically endorsed.

"The comment could be tangential/adversarial to your opinion, and so you might not upvote it, however it could still add to the discussion, and hence you would endorse it."

I was under the impression that upvoting was to reward comments that added to the discussion, rather than for ones that you happened to agree with. I have upvoted comments that expressed a view that I thought was wrong, and that said that I was wrong, because they were good comments. There is always a chance that I'm wrong, however infinitesimal that might be.

Sir, your response is well-reasoned and satisfactory.

  "I do not agree with what you have to say, but I'll 
   defend to the death your right to say it." 

  - Voltaire

Would a moderator care to turn on pending comments for this thread, so we can get an idea of how it works?

Are there several moderators or just one "main moderator"? I can't work that out from the text.

There is one main moderator (who is not me) as well as several other people who could do something like that but wouldn't.

Why not just bury new comments at the bottom of the thread until they get an upvote (essentially equivalent to "endorsement".) It's a terrible idea (new comments never get seen), but it's functionally the same and only a small modification rather than a whole new system.

You can also have an option to "hide" comments below a certain threshold (i.e. ones that haven't been "endorsed" yet) if that is the desired system. Or an option like showdead for low karma comments (perhaps count only total upvotes or "endorsements" rather than upvotes-downvotes.)

I'm not sure I'm a fan yet but the main benefit of pending comments will be that it should make people think more before commenting. If pending comments can be enabled per-thread, will there be a visual warning when commenting? I know mods will still have to approve the comment but that seems the lesser of the two benefits to me. If not then pending comments need to be enabled for the majority of threads to make everybody operation under the assumption that they are enabled.

Why not turn on pending comments site-wide for 24 hours and see how it works out? The initial test worked fine, and it'd be a nice experiment.

The main reason not to is that comments sitewide come too fast for a human to watch what's happening in real time.

As for a karma rating to pend, then perhaps some scaling, if backlog of pending comments then the limit lowered and if low then raised, Base on say 5-15 minute average, or even hoursly or daily...

Also thing the initial poster should automaticly get some level of control upon the posted comments.

Random is what /. did for a last time I used it and that is pretty random in results, garbage in, garbage out as they say.

It seems like many/most sites with content issues use a "flag bad stuff" option rather than an "endorse good stuff" option-- which makes sense if most stuff is good. Why not give people with karma level of X (5000?) the ability to insta-kill anything that is rude? And perhaps people with lesser karma the ability to flag, where X flag points = comment death?

I really like the idea of introducing some form of randomness to prevent gaming, social stratification, and other unintended consequences.

My immediate thought is to suggest some type of majority voting and maybe even random bagging -- e.g., show the "endorse" link to a few random users (or groups of users) and endorse any comment that gets majority vote.

How do you even determine results empirically? Is there some objective measure of utility, or is it just up to whether HN mods/admins think that things seem better with the new feature? If it's something like the latter, I wouldn't call that an empirical approach.

This may or may not be a bad change to HN, it's too early to tell. Instead of having karma as the sole criterion of whether or not to endorse comments, why not have additional criteria such as age of the account, or number of threads started that netted x amount of karma?

You may have never worked in a content industry before, but there are certainly empirical methods you can use to find out if something is more appealing to people along some axes.

To further the endorsing aspect, how about we link the endorser to the endorsee? If the user endorses a comment which gets downvoted or flagged, there should be some negative reinforcement for the endorser. Similarly, if the user endorses a post that then goes on to get lots of upvotes, the endorser should also receive some positive reinforcement.

In that case, then nobody would dare endorse even the most remotely controversial statements.

Result: Total echo chamber.

I've been thinking about pending comments and some of the issues I foresee from it. I have a major suggestion on how to change pending comments functionality to make it more useful and keep comments productive:

The basic idea is to move away from the model of censorship/moderation and toward a model around rate limiting, as well as to integrate more seamlessly into the existing workflow of HN rather than to bolt on new things for people to do that cause the site to break down if they're not done.

So, instead of hiding pending comments and making them visible only on the pending comments page until they've been upvoted enough to graduate to real comments you show them inline, but not 100% of the time. There are lots of algorithmic options on this part but one simple idea would be to display them to X% of users with over 500 karma who have been active within the last few days. The posts would simply appear as normal, and have normal up/down vote buttons on them.

If the post gathers enough upvotes then it gets promoted to universally visible. If not then it continues to be only partially visible until some timeout where it becomes visible.

Additionally, the amount of upvotes/downvotes could change the rate limit timeouts on the post. If a post receives a few upvotes then it could be automatically promoted out of pending status faster than one with no upvotes. If a post receives a few downvotes then it could be promoted slower than normal. If a post receives enough downvotes then perhaps it disappears, or simply takes a very long time to show up (days? a week?)

The advantages of doing it this way are many fold. It blends into the normal functioning of the site. People won't even know they're voting on pending comments, they'll just use the site as usual. More importantly, it gets rid of a lot of the clunkier problems of pending comments, such as the potential for comments to languish in the pending state. In that case this model simply falls back to the rate limiting model. The site would still continue to function just fine even if nobody voted on pending comments, but the pace of discussion would be slower. And it avoids the censorship problem. Moderation is a fundamentally hard problem even with humans in the loop, by defaulting to allowing all comments to surface eventually you solve the problem of denying people a voice while also putting the brakes on overheated emotionally fueled flame wars and so forth.

Also, I'd propose a few tweaks. First, make it so that high karma users' comments usually skip the pending step. To me this just seems logical and a good way to avoid excessive friction. If necessary you can add a toggle which forces pending comments behavior on a per thread basis. Also, you can add functionality which pushes a comment into pending status if it's downvoted. Second, consider making direct replies always visible to the parent poster, perhaps gated based on karma of the parent/child or both. Third, perhaps make it impossible to reply to a pending comment unless it has at least one upvote. That would preserve constructive back and forth comment threads while discouraging flame wars and vitriol.

I don't want to see commentary, even controversial commentary, squashed in a misguided effort to turn HN into the digital equivalent of "Pleasantville". Sometimes controversy leads to extremely valuable discussion. But I think discouraging and quieting non-constructive commentary is well within the realm of possibility.

PS. Release the HN Code.

Need changes to the code, ask for the HN community :)

pg, how many moderators are there?

I sincerely apologize for saying this, but should all this have been made clear since Day 1?

Not making public even a very basic whitepaper as to how to implement a site-wide change (and such a major one as well) just seems... off, given the high quality of this site otherwise.

But thanks for the explanation. Duly noted.

I wrote most of the code the day it launched, and I had to get it done that day because I had to help the startups prepare their Demo Day pitches.

Thanks for saying the site is high quality, but if so it's despite (or possibly because of) the fact that much of the development happens in the repl of the live site. There are not a lot of whitepapers around here. We tend to think by writing code.

> much of the development happens in the repl of the live site

Wow, I guess that gives me an excuse when I'm too lazy to test on a dev server. "Well, PG live edits HN"

>"Well, PG live edits HN"

But HN has a persistent bug where the next button on the first page expires and doesn't work. That's the difference between doing something for work and doing something for fun.

The willingnes to tolerate malfunction is higher

> HN has a persistent bug where the next button on the first page expires and doesn't work.

I don't think that's a bug, but a feature - a type of cache invalidation that forces you to refresh to ensure paging is accurate

There's the same 'feature' which works against insightful commentary. Spend too long crafting a response, get "unknown or expired link". Not particularly endorsing of a considered, pensive response.

Surely there are better ways. For example, you keep track of which posts a user has already seen, and when they hit the next page button, show them the X most highly ranked posts that they haven't yet seen. Breaking the link is just the easiest way to deal with the problem.

He's also caused significant downtime by doing this, so this may not be a good lesson to take for your own sites.

Actually I don't think any of the outages have been caused by this. The prospect does add an edge of excitement to programming though. Gamification for programmers!

There was one outage caused by a runaway which testing would have easily caught; I think about last year.

Are you thinking of this outage from last month?


There was something from >1y ago.

Probably this one:


(I remembered "new user" so was able to get a search to work...)

Good catch. But strictly speaking that wasn't doing development in the repl, but repairing data. The live server is the only place you can repair data on the live server.

Yep. That's the kind of thing which is fun for hn but wouldn't be a good idea at all on a financial system -- where you optimize for least bad likely worst case, not best best case.

That's interesting insight. I wonder what would happen if Bank of America took that same approach to pushing changes...

...what about the White House?


If there is any world where the 95th percentile programmer at BOA is as good as pg, perhaps.

This is my all time favorite HN comment!

For everyone who claims there is one true way to write successful software, it should be noted that one of the most popular (or least influential) sites frequented by developers was built on the live server using the REPL!

Maybe PG is the exception that proves the rule, but damn that's awesome.

I think PG's company is the exception that proves the rule. It's still probably a bad idea.

Don't know. It has been shown to work.

One of the major benefits touted by the proponents of Agile was that it modeled how successful software teams worked. Well here's a successful software team and this is how it works.

For most teams it is probably a truly bad idea. But it shows there is more than one way to do it.

Update: The more I think about it, maybe this isn't so crazy. Wasn't Erlang built to work this way, so engineers could modify a telecom switch without taking it offline? As systems get gigantic it gets very difficult to have a staging environment that mimics the live site.

"We tend to think by writing code."

That should get added to a quote collection somewhere :)

Great! So it will be even more left-wing, white knights oriented, immature behind big books kind of an affair. Write me off today Paul! Have a nice day!

I know we're not allowed to complain that HN is becoming Reddit. But with this feature..

HN is becoming Slashdot.

Applications are open for YC Summer 2019

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact