X open-sources its ranking algorithm and launches transparency tools

On August 13, 2026, X significantly expanded the open-source release of its ranking algorithm, publishing source code for its “For You” timeline feed under the Apache 2.0 license. The release includes the core ranking engine, model configuration parameters, filter details, and the Phoenix scoring system — the last of which external researchers had already independently validated in tests.

Alongside the code release, X introduced a transparency feature called “Under the Hood,” available to users who post at least ten times per month. These users can download their aggregate account data as a JSON file showing which labels X’s systems have applied to their account or posts. The company explicitly designed the tool so that non-technical users can feed this JSON data into an LLM alongside the open-sourced GitHub repository and receive a plain-language explanation of how the system handled their content.

The move addresses a category of product decisions that most social platforms have historically left opaque: what determines whether a post reaches its audience, and whether a user’s reach has been reduced by automated systems without notification. X frames the transparency tool as a direct response to long-standing shadowbanning concerns — the practice of limiting a user’s distribution without their knowledge.

For product managers, the significance sits at two levels. At the pattern level, platforms competing for creator and publisher attention will face increased pressure to explain their ranking and distribution systems more explicitly, now that a major platform has established a concrete example of what such transparency can look like. At the design level, the specific architecture X chose — open source code plus read-only personal data export, rather than user-controlled ranking parameters — represents a deliberate decision about where to place the accountability boundary. That trade-off is worth examining for any product team building AI-influenced distribution, feed curation, or recommendation systems.

Systems used by Grok for rule violation prediction remain proprietary. X described this as a security boundary to prevent bad actors from optimizing around content moderation. The tension between that carve-out and the broader transparency narrative is worth watching as the policy develops.