I extracted the upvote data from the full corpus of public articles from Inkhaven 2. It confirmed my intuitions. Inkhaven 2 had 1,789 public posts and 1,436 total upvotes. 57% of posts had no upvotes.

You can download the data from the Inkhaven Slack. DM me for a link. This data does not include private posts.

Why have upvotes? I’d be interested to hear the “mission statement” of Inkhaven’s upvoting system, from its implementers.
My take: For writers: To celebrate & elevate your good work. For readers: to make good content more visible. For gamers: to feed our competitive cravings.

Upvotes meant reaching the “Featured” listings at the top of each day’s collection. I call those 6 articles “above the fold” to borrow a newsprint term. I enjoyed moments when my posts peaked above the fold. I noticed when I upvoted an article and that article bounced above the fold and knocked my own out. I felt virtuous in those moments. And that was also a glimpse of the data that was later confirmed: getting above the fold - becoming elite - might take a single upvote.

The Reddit (and Digg before it) model: Upvoting elevates good content and buries bad content.
Reddit upvoting serves a distinctly different role - there is inherently bad content, tons of it. Downvoting helps filter it from the site entirely. It’s a matter of survival. Let’s presume we don’t need 90% of all submissions to fall out of visibility. Of the 55 posts a day, start with the benefit of the doubt that each has some value.
I’m not in favor of downvotes in our model. While there are a few bad posts, they would sink to the bottom in a better functioning upvote-only system.

What did upvoting do well? I invite your take.

I don’t question voters’ choices; presume that all upvotes are deserved. I actually believe the system works in the case of the handful of highly upvoted posts. If the intent of the ranking system is to mark a few “star” articles apart from the others then it succeeds. Upvoting systems inherently create a self-reinforcing “star” tier.

Some takeaways from the data:

There were 1789 public posts
There were 1,436 total upvotes

57% (1014 of 1789) posts got 0 upvotes.
26% (459 of 1789) posts had 1 upvote.
1 post had 10 upvotes, 1 had 15, and 1 had 16. There is your classic “Long Tail.”
This looks like the right half of a bell curve. How do we get the left half?

I suggest a more representative ranking system would in some way distinguish 1014 of the 1789 posts. Zero upvotes can mean so many things. For me, it could have meant “I don’t recognize that writer or article title - I didn’t read it”, or “I liked it & I’m too fried to go back to the portal and find the upvote button” or “I’m past my weekly quota of AI safety doom, and it’s only Tuesday.”

Clearly getting a single upvote separates a post from the majority. My intuition says that is a low quality signal, that many 0’s could have been 1’s and vice versa. I suggest that an upvoting system should do a better job of distinguishing the relative quality within those 1,473 posts with 0 or 1 upvote. For the small number of bad posts (where one might actively downvote) they should not be lumped in with the other 0-vote posts.

Solution: Increase Upvoting.

If more upvotes went to the 1,473 zero/one upvote articles, it would have separated them better. There simply weren’t enough upvotes cast. 1,436 upvotes is less than 1 upvote per day per person, and that doesn’t even include staff, alums & other portal users (if they get upvotes?). Perhaps people took the advice too much to heart about limiting their upvotes. If the system encouraged more upvotes, I claim it would yield more valid results. This encouragement could be via policy and/or by UX. An Inkhaven “Reader” site could show an upvote button alongside an article.

Alternative: Scrap Upvoting

If not ranked by upvote, how might the main page order articles - Random? grouped by category?
A no-upvote system is at least worth a trial. I question if much would be lost.
Perhaps article arrangement could be a user preference - recent/ upvotes/ alphabetical / etc. This gives you an A/B mechanism for testing different ordering.

Upvoting currently populates the “Featured” listings at the top of each day’s collection. I like this design, it gives the page a space to breathe, a few articles get a bit of extra text to entice a click. Without upvotes, some other selection method would be needed.

The Inkhaven upvoting system was created on the fly as we used it. It’s a new thing, and this is in no way a critique of that process. It’s right where it should be given the timeframe. Let’s think about where to take it from here.