ELSEIF
Your brief EB
334 stories from 97 feeds 287 clusters Refreshed 4 minutes ago next pull 18:36

SECURITY Signal 395

X publishes 'For You' ranking code on GitHub under Apache v2, adds in-app tool to surface ranking effects on accounts

Illustration only Photo by Alan Bowman on Unsplash

X has published the source code for its 'For You' feed ranking system on GitHub under the Apache v2 license and is rolling out an in-app transparency tool that lets users download a JSON file showing whether ranking systems have applied labels to their account or posts.

WHY IT MATTERS

For a working engineer, this is one of the larger public drops of a production social-media ranker, shipping with model configuration, filter logic, and the weighting parameters that decide what gets shown. The Grok-based system that predicts rule-violating posts is deliberately excluded to keep spammers from gaming it, so the open code is the ranking half, not the full enforcement stack. X has become less transparent in other areas since going private, user metrics, growth, revenue, and government takedown requests are no longer routinely reported, so the open-sourcing here does not extend to the company's overall disclosure posture.

Written by elseif from the cluster below · every claim links back to a source

The three things worth knowing

01

X released the 'For You' timeline source code on GitHub under Apache v2, including model configuration, filters, and the parameters that weight different ranking signals, and the codebase is roughly 10 to 15 times larger than X's prior open-source drop.

02

A new 'Under the Hood' section in app settings lets users who have posted at least 10 times in the past month download a JSON of their aggregate stats, including any labels applied to their account or posts in the past calendar month, with the pilot initially limited to accounts that are at least a year old.

03

External researchers had already run the per-post 'score' outside of X before launch, and X says it will accept pull requests on the repo, though the Grok-based system that flags potentially rule-violating posts is excluded to prevent adversarial tuning.

THE READ

What the cluster adds up to.

ORIGINAL ANALYSIS

The headline change is that X has published the source code for its 'For You' feed on GitHub under the Apache v2 license, including the model configuration, filter, and core ranking system details, meaning the parameters that weight the different signals used to score and order posts. By the company's own framing, the codebase is roughly 10 to 15 times larger than what it has open-sourced before. X also previewed the repo to outside researchers familiar with recommendation systems, who got the per-post 'score' running outside of X before launch, and the company is publicly inviting pull requests against the repo, even though not all contributions will land.

For engineers, the practical entry point is twofold: read the GitHub repo to understand how the ranker is assembled, and download a JSON file from a new 'Under the Hood' page in the app's settings that records any labels applied to your account or posts in the past calendar month. That JSON is meant to be readable on its own but is also designed to be fed into an LLM alongside the repo to get an interpretation, which lowers the bar for non-specialists. The pilot is limited to accounts at least a year old, so for the moment the new transparency tool is not a general-purpose right any user can exercise, and adoption is gated by both an account-age floor and an active-posting threshold.

The release is deliberately not full-stack. X is withholding the Grok-based system that predicts whether a post could be violating a rule, citing the need to keep bad actors from using the information to spam the network. That means the open code covers ranking and filtering, not the upper layer that decides which content is moderated, and the trade-off X is making is reproducibility in exchange for some adversarial headroom. The excluded component is not named as a separate open release, so an auditor can verify what gets shown but not what gets removed, and any future attempt to game the ranker from the open source has to assume the closed layer is still in play.

The release comes against a backdrop in which X has become less transparent in other areas since being taken private. The company no longer reports user metrics, growth, revenue, or government takedown requests with the regularity it did when it was filing with the SEC, and prior public-company disclosures have not been replaced. So this is a focused expansion of transparency on the ranking side, not a broader shift in disclosure practice, and engineers reading the repo will get a much clearer picture of what surfaces in the feed than of how the platform as a whole is operating.

Written by elseif from the cluster below · checked for specifics the sources never contained

THE CLUSTER

Same story, 1 feed.

ORDERED BY FIRST SEEN
TechCrunch X open sources its ranking algorithm, letting users see if they’ve been ‘shadowbanned’ Open ↗