X open-sources "For You" algorithm and launches "Under the Hood" shadowban checker — but key parameters stay hidden

On Thursday, August 13, 2026, X announced a significant expansion of its open-source transparency initiative by publishing the source code behind its "For You" recommendation algorithm and core ranking engine on GitHub under an Apache v2 license, alongside a new "Under the Hood" tool that lets users check whether their account or posts have been affected by visibility-limiting labels.

The open-source initiative

X released the full recommendation engine code — a ground-up rebuild in Rust and Python — to the xai-org/x-algorithm GitHub repository. The release includes model configuration, filtering systems, and core ranking code that determines which posts appear in users' feeds. According to X VP of Product Keith Coleman, parts of the ranking system can even be run independently outside the company. The codebase is now roughly 10 to 15 times larger than prior open-source releases.

The "Under the Hood" tool

The new tool, accessed via an "Under the Hood" page in the app's settings, lets eligible users download a JSON file of their aggregate account statistics from the past calendar month. The file reveals any labels applied by X's ranking systems that may limit a post's or account's visibility — essentially functioning as an official shadowban checker. The tool shows what each label means and how it affects reach. Early users have reported seeing labels such as "SPAM_HIGH_RECALL" that hide posts from recommendations to non-followers.

Eligibility requirements for the initial pilot:

• A test group of accounts that are at least a year old
• Accounts that have posted 10 or more times in the past month

Non-technical users can then feed the JSON data into an LLM with X's GitHub repo for interpretation.

What was excluded and why

Significant portions of the algorithm — including key operational weights and exact configuration settings — remain redacted. Crypto Briefing reports that expert John Thickstun noted these omissions limit the release's practical transparency, because independent researchers cannot fully reproduce the algorithm's behavior or audit its outputs without that data. X has not publicly explained why those specific components were withheld.

Broader context: X's shifting transparency since going private

TechCrunch noted the announcement arrives as X faces renewed scrutiny over its transparency practices since Elon Musk took the platform private in October 2022. The article references a previous January 2026 open-sourcing effort that occurred as X was simultaneously facing a transparency fine and controversies over its Grok AI model. As a private company, X is not subject to the same mandatory disclosures as publicly traded firms, which has allowed it to operate with far less formal transparency than the old Twitter — making voluntary releases like this one both notable and incomplete, as critics argue the redacted parameters undercut the stated goal of full openness.