Curation has been the quiet part of Ecency for years. It runs every day, it touches thousands of people and almost none of the rules were written down anywhere. That was our mistake. So here is the whole thing: the stake, the people, the weights, the numbers and what we are changing.
Web2 platforms are moving back toward human curation, because algorithmic feeds are good at surfacing volume and bad at surfacing quality. Medium wrote about this in July 2026 in How Medium is investing in the curation era and the number in their subtitle is the one worth sitting with: "for the average Medium reader, over 75% of stories on their feed are there because a human, exercising their taste, said they should be." They are paying publication editors through an Editor Partner Program, building tools for editors to reject low-effort submissions and giving readers mute and follow controls to shape their own feed.
The line from that post that applies most directly to us is this one: "curation is just as much about what's excluded as what's included."
On Hive that number is not 75%, it is 100%. Every reward and every ranking signal on this chain comes from a person voting. We have had what Medium is now building toward since the beginning and we have been running it on accumulated habit rather than stated rules. So we are catching up on the two things that post gets right: publish the rules and open the pipeline to people outside the project.
The stake is not ours. 
The DHF pays for servers/development, not votes. Our proposal pays 100 HBD a day, down from 396 and 300 in earlier years. It funds the apps.
People decide. A curator team plus two guest curators every month read and vote from their own accounts. The
The weights. Curators do not set the size of the
Our own posts. 
Who it reaches. Across the last 1000 votes, spread over 8 days, 429 different authors were curated. 43% of them hold under 2,000 HP and only 12% hold more than 50,000. The median author we voted holds 2,765 HP and was voted once. That is what we want this to be.
One weight for everyone. The front-end multiplier is gone. Every curated post now lands at the same weight, currently around 5.9%. We did this by raising the floor rather than cutting, so the total we hand out each day is unchanged: posts from other front-ends nearly double, from about 3.2% to 5.9%, and Ecency-published posts come down slightly, from about 6.3% to 5.9%.
Nomination opens to every front-end. Until now the curation queue was fed only by posts published through Ecency, which is the real reason our votes concentrated here. Curators now see posts from anywhere and Ecency-published ones stay visually marked so our own community is still easy to find in the queue.
The remaining rules, stated plainly. Curators receive a fixed 10% vote on their own posts as compensation for volunteer work. Our own accounts receive large votes and redistribute 99% and 80% of them as described above. Those stay and now they are written down.
Curators get more to go on. The queue used to show a single number labelled "AI". It now shows four things, none of which decides anything: how templated the writing reads, how many of the post's images sit on a Hive image host rather than being linked in from elsewhere, how often that author replies to people and occasionally a note that a post reads very differently from that author's own earlier writing.
That last one is deliberately rare. It is measured against the author against themselves rather than against some idea of what machine writing looks like, so it does not go stale as models improve and someone who has always written with help never trips it. Across 33 real posts in testing it fired zero times.
We renamed the first one from "AI" to "Formulaic" because it was not measuring what the label claimed. Tested against text written by a current model it scores near zero, while flagging ordinary human posts at 80%. What it actually detects is templated, boilerplate prose, which is a real and useful signal, just not the one the old label promised. Auto-generated activity reports score high and they are not AI-written at all.
We will consider visiting/checking these numbers quarterly and report. Distinct authors curated, share by front-end, share by author size and share going to comments. The same queries anyone else can run, so the report can be checked.
Curation should be an experiment with a scoreboard, not a habit. Four things we want to try, each with a number attached so we can tell whether it worked:
Weight by author stage instead of by app. More weight for newer and smaller authors. Baseline to beat: 43% of curated authors under 2,000 HP.
Wider spread. Baseline: 429 authors in 8 days, median one vote each. We would rather move that up than vote the same people harder.
More comment curation. Only 7.5% of our votes currently land on comments, and conversation is what keeps people here.
Signals that measure effort rather than authorship. The ones above are a start, not an answer. Authorship is the wrong question: a generated post can look considered and cost half a minute, while a short awkward one can be the only thing someone had time to write that day. What we want curators asking is whether a post contains something only that person could have produced, their own photographs, the specific details a participant would know, a stake in what happened. That test does not care which tools were involved and unlike polish it cannot be manufactured cheaply. Baseline: today only one of our four signals gets anywhere near it.
If a change does not move its number in a month, we will say so and revert it.
Guest curator seats open every month and anyone can apply in our Discord. It is the fastest way to find out whether any of the above is true and the invitation has been open the whole time.
