What is TweepCred

Q: Does TweepCred change the For You score?

Not under that name, and not as UserCredV2 either. The weighted sum does not read it. home-mixer and visibility-filtering contain no reference to the type. Premium is not an organic ranking bonus in param.rs.

Last updated 2026-10-11. A search of xai-org/x-algorithm at commit cb45b55 finds no file, comment, or identifier named TweepCred. This page will not invent a calculator for a name that is not in the tree, and it will not copy constants from an older repository this site has not re-read.

People still use the word for “a hidden author reputation score.” The current tree has a different job, with a different name, and it is not wired into the For You sum.

What the current code publishes instead

user-cred-v2/ is the account-scoring job. The README’s August 13, 2026 section lists it among models that score accounts on various dimensions, next to Agatha and BDSM, in the group of systems that can feed visibility labels. The package is com.twitter.health.platform_manipulation.user_cred_v2.

What that job actually computes, from the Scala in that directory:

It runs a PageRank-style iteration (pageRankMain) over follow edges, after dropping users who are restricted, deactivated, suspended, or erased (ValidUserInfo.fromFlatUser).
Engagement edges are favorites and reposts (InteractionType.Favorite and InteractionType.Retweet) inside EngagementWindowDays = 7. Edges between linked accounts from the account-expansion dataset are removed.
The teleport mix (computeTeleportWeights) blends two distributions with engagement_teleport_beta, default 0.5. One side is uniform over accounts whose isPremium is true (blue, gray, gold, verified organization, or affiliate). The other side is those engagement edges, weighted by the engager’s prior mass.
Default PageRank arguments in UserCredV2Config.fromArgs are jump probability 0.2, difference threshold 0.001, and max iterations 50. Those are argument defaults, not a measured reputation of any account.
UserCredV2.fromMass turns mass into a score: 0 when mass is ≤ 0, otherwise 165.2 + 7.07 * ln(mass), clamped to 0–100.

A warm start reads the previous mass snapshot no older than 7 days. If a previous mass exists for that id, joinPreviousMass keeps the previous mass rather than the fresh one.

What it does not do

home-mixer/ does not reference UserCredV2. visibility-filtering/ does not either. The published For You weights in home-mixer/params/param.rs have no TweepCred column and no user-cred column. Premium is not a term in that sum. The premium flag above is only the uniform teleport set inside user-cred-v2.

So a TweepCred-style number, even under the new name, is not something this commit multiplies into a post’s For You score. The README says the job can matter to visibility labels. The code that would read UserCredV2 from inside visibility-filtering/ is not in this commit, so this page does not claim a specific label or a drop threshold.

SimClusters jobs mention a dataset named entityRealGraph. That is a graph input to embedding jobs. It is not TweepCred, and it is not a head in compute_weighted_score.

If you wanted a number on a draft

A draft score and an author-reputation score are different objects. tweet-factory’s draft score uses the published action weights and is described on How tweet-factory scores drafts. The browser draft score is tweetfactory.sh/score. It is not a TweepCred calculator, because the current ranker does not publish one.

FAQ

What is TweepCred?

In commit cb45b55, it is a name that does not occur. The account-scoring program in that tree is user-cred-v2: a PageRank over follows, with a teleport that mixes premium accounts and seven days of likes and reposts, then a 0–100 score from 165.2 + 7.07 * ln(mass).

Does TweepCred change the For You score?

Not under that name, and not as UserCredV2 either. The weighted sum does not read it. home-mixer and visibility-filtering contain no reference to the type. Premium is not an organic ranking bonus in param.rs.

Can I calculate my TweepCred from this repo?

No. The job needs production datasets (follows, server engagements, user snapshots) that are not in the GitHub checkout, and the For You request path does not consume the output. This site does not ship a TweepCred calculator on top of a missing input.