The algorithm is open

Every number on this page is imported from the live ranking code — when the code changes, this page changes with it. No black box, no “trust us.” The one law everything below obeys: money never buys rank or reach.

How For You works

Candidates come from four pools, every post gets one blended score, then mixing rules shape the final page.

Your network

Recent posts from makers you follow or have recently sparked. Follows count double.

Your interests

Posts whose content embeddings match what you've engaged with — meaning, not keywords.

Proven this week

Site-wide engagement leaders from the last 7 days, re-scored by the same quality math as everything else.

Brand new

The newest posts from anyone — zero followers required. At least 5 slots per page are reserved for this pool, so new makers always get reach.

First creation

A maker's very first post, while it's still fresh and nobody has responded yet, gets at least 2 reserved slots. Silence — not criticism — is the #1 reason new makers leave, so a debut never dies unseen. A discovery slot, never paid reach.

The score

score = 0.3·affinity + 0.25·interest + 0.25·quality + 0.2·freshness

Affinity · 30%: How much you’ve engaged with this maker — your follows and sparks, nobody else’s.
Interest · 25%: Semantic match between the post and what you engage with.
Quality · 25%: What people did with the work per time it was seen — a Wilson 95% lower bound on weighted engagement per view. The lower bound means a post with 3 remixes from 10 views isn’t instantly crowned over one with 300 from 10,000; small samples earn trust gradually.
Freshness · 20%: Hacker-News-style gravity, 1/(hours+2)^1.5 — softened so day-old posts still surface.

Then the mixing rules:

Max 2 posts per author per page — nobody monopolizes your feed.
At least 5 of 50 slots go to the brand-new pool — the fairness quota that gives every new maker a shot.
At least 2 more slots are reserved for a maker’s firstcreation while it’s still fresh and unanswered — so a debut never lands in silence.
Posts you’ve already seen are demoted to 35% of their score.
Your own posts are filtered out.

The quota in practice

Not just a target — here’s the live supply of quota-eligible posts in the catalog right now (refreshed every few minutes). The back-fill reserves these slots on every feed whenever supply meets the floor, so these counts show whether the promise above can actually hold.

Brand new40newest makers reachable — floor of 5/page is fillable

First creation0fresh & unanswered in the last 48h — supply thin (0/2 floor); a quiet stretch, not a broken quota

“Eligible” means: a root post from a non-test maker the exploration pool can reach (the 40 newest — brand new), or that maker’s first-ever post still fresh and unanswered in the last 48h (first creation) — the exact pools the back-fill draws from. A shortfall here is an honest signal the catalog is quiet, never a hidden number.

What engagement is worth

The quality term weighs actions by how much they say about the work:

Remix×5someone built on your work

Building×4.5someone marked it building in their stack — intent to build

Save×4someone wants it in their stack

Reply×2it started a conversation

Boost×1.5someone vouched for it

Spark×1the baseline signal

How replies are ranked

Threads rank replies by what each reply contributes, never by who wrote it. The author’s own replies pin first (it’s their thread), then:

replyScore = 1·sparks + 2·replies + 2·boosts + 3·remixes + 4 if it ships code + 4 if it links a demo

Ties break chronologically. Bringing code or a working demo to a conversation outranks any amount of account status — because account status isn’t an input.

Never an input

Money. There is no paid placement, no premium tier, no boost-for-cash. Nothing you can buy moves rank.
Verification or account status. No checkmark math anywhere in the pipeline.
Follower count as status. A zero-follower account's post competes in the same exploration pool every page.
Who you are off-platform. No imported clout.
Engagement bait mechanics. Quality is engagement per view with small-sample humility — farming raw counts without reach doesn't work.
QA / test accounts. Flagged test fixtures are excluded from every discovery pool — they can see their own posts, but never occupy a real maker's reserved reach.

Brand new pool: checked live, not just asserted

Right now 19 of the 40 newest root posts are from test accounts. Run through the exact filter the For You discovery pools use, 0 of them reach a real feed — the exclusion holds.

And excluding them doesn’t starve the window: the pool reached past 34 test posts to assemble a full 40-post real-maker window — +35 above the 5-slot floor.

Interest pool: checked live, not just asserted

The semantic-match interest pool (embed similarity) queried the live match_posts_for_user RPC asking for 50 candidates. It returned 0 real-maker posts with 0 test posts — the exclusion holds.

No posts matched the embedding query right now (cold interest profile or sparse catalog). The interest pool returns zero safely.

Questions or think a weight is wrong? Post about it — the feed will rank your critique by the same rules. See also Community Guidelines.