TL;DR
We just added the ability for ❜embed users to order candidate pools in order to keep feeds fresh, high‑quality, and fair depending the desired use case.
Ordering happens at the candidate generation step, the first step in the feed building process.
Ordering helps select and pre‑sort candidates before the ranking function scores the final list of posts. In other words its a retrieval strategy that decides which items enter the candidate pool and in what preliminary order.
What is “feed candidates ordering”?
Feed ordering happens at the candidate generation step where we choose and pre‑sort candidates before the ranker decides the final feed. It doesn’t replace ranking; it shapes the pool that the ranker works with and therefore it’ll significantly influence the final output.
By ordering for chronological, popular, trending, and randomness feed creators can balance freshness, quality, serendipity, and fairness. This can help satisfy different goals like helping users consistently discover emerging creators, or showing guaranteed quality content with more traction.
4 types of feed candidates ordering
Chronological: newest posts come first within the specified time window.
Why use it: get the freshest post first
Trade‑off: quality can be variable here which can be mitigated by more intentional filters and opinionated ranking.
Popular: highest cumulative engagement/quality.
Why use it: best for guaranteeing consistent high quality.
Trade‑off: there are risks of staleness as most popular posts stay the same. Time decay can be used to make posts fresher.
Trending: fastest recent post velocity and momentum.
Why use it: this ensures new items reach the ranker quickly instead of being buried. Good for social trading apps for example.
Trade‑off: some posts that catch traction might feel spammy. Make sure to implement tight spam filters.
Random: controlled random sampling for long‑tail discovery and cold‑start fairness.
Why use it: Gives new creators/assets real exposure and maximizes discovery for users.
Trade‑off: quality can be more variable since posts haven't been curated by the network yet. Can be mitigated with engagement filters and AI labels.
Why it matters
As said earlier, the better the inputs to the ranker, the better its outputs and the better the feed.
When we see things from this point of of view, ordering is yet another lever we can use in order to obtain the best possible feed.
What changes for feed creators and end users
For end users this means even more accurate feed. Feed creators have even more control over what gets ranked and what gets served.
This feature is especially relevant when combined with feed mixing. For example you can create three different feeds:
One maximizing for discovery of emergent creators
A second one in charge of showing fresh content to the user
A last one ensuring users sees the top content in the network
You can then mix your feeds using the promotion feature with the desired % in oder to obtain the perfect mix.
Experiment using feed_1678
You can go to the ❜embed feed builder and access Farcaster Alpha (feed_1678), a public feed config that you can copy and remix. Change the ordering parameter and see how this impacts the feed in preview.
