Frequently Asked Questions

Is Under The Hood Analyzer affiliated with X?

No. Under The Hood Analyzer is an independent website and is not affiliated with, endorsed by, or officially connected to X Corp. or X. We use publicly available information, including X's published technical material and open-source code, for research and analysis.

Is X's recommendation algorithm open source?

X has published substantial parts of its recommendation-system code publicly on GitHub. The repository describes the recommendation system as a collection of services and jobs used across X surfaces, including the For You timeline, Search, Explore, and Notifications. It also documents components such as user-action signals, SimClusters, TwHIN, candidate generation, and ranking systems. (GitHub) However, open source does not mean every part of X's production system is publicly available.

What does the open-source X algorithm actually reveal?

The published code provides insight into the architecture and mechanisms used by X's recommendation systems. The newer open-source x-algorithm repository describes a For You recommendation pipeline involving in-network and out-of-network content, candidate pipelines, retrieval, filtering, and ranking. It also documents a Grok-based transformer used for recommendation ranking. (GitHub) This gives researchers and creators considerably more technical information than simply guessing how the algorithm works.

Does the open-source code represent exactly what X is running today?

Not necessarily. Open-source repositories are snapshots of published code and may not contain every production component, configuration, model, experiment, business rule, or continuously changing parameter used by X. Therefore, Under The Hood Analyzer treats the source code as technical evidence, not as a guarantee that every recommendation decision on X currently follows exactly the same implementation.

Why does my post appear in the For You feed?

X's recommendation system can use both in-network content from accounts you follow and out-of-network content discovered through recommendation systems. The published architecture includes candidate-generation and ranking stages that help determine what content may be shown. (GitHub) This means a post can reach people who do not follow its author.

Why do I see posts from people I don't follow?

The For You feed is designed to recommend content beyond your existing network. X's published recommendation architecture includes systems for retrieving and ranking out-of-network content, allowing posts from accounts you don't follow to become candidates for recommendation. (GitHub)

Does a Like guarantee more reach?

No. A Like is one user action among many signals that recommendation systems can consider. X's published code describes a centralized user-signal system containing both explicit actions, such as Likes and replies, and implicit signals, such as profile visits and post clicks. (GitHub) A single engagement action should therefore not be interpreted as a guaranteed reach multiplier.

Are replies more important than Likes?

There is no universal rule saying that every reply is worth more than every Like. Different actions can be modeled separately, and recommendation systems can predict multiple possible user actions rather than relying on one engagement metric. The newer open-source X algorithm describes multi-action prediction as part of its design. (GitHub) Under The Hood Analyzer therefore avoids presenting simplistic claims such as “one reply = X Likes.”

Does engagement velocity matter?

Engagement timing can be important to how a post develops, but creators should be careful with claims about an exact “first 30 minutes” rule or a guaranteed viral window. The open-source code provides insight into candidate generation, scoring, ranking, and user-action prediction, but it does not establish a universal public formula saying that every post is evaluated according to a fixed viral timer.

Why can a post with fewer Likes get more reach?

Reach is not determined by Likes alone. A recommendation system can consider multiple signals, predicted actions, relevance, candidate sources, filtering, and ranking. X's published architecture explicitly includes multiple user signals and ranking stages. (GitHub) Therefore, two posts with the same number of Likes can receive very different distribution.

Does follower count determine reach?

Not by itself. The recommendation architecture allows content to move beyond an author's existing follower network through out-of-network candidate retrieval and ranking. (GitHub) A smaller account can therefore receive significant distribution when its content becomes relevant to users outside its follower base.

Can Under The Hood Analyzer predict whether my post will go viral?

No tool can guarantee virality. Under The Hood Analyzer can analyze available signals and explain patterns that may be associated with recommendation and engagement, but future distribution depends on a complex recommendation system and changing user behavior. Any score or prediction should be treated as an analytical estimate, not a promise.

Does Under The Hood Analyzer know X's complete ranking formula?

No. We do not have access to X's private production systems, internal experiments, private datasets, or proprietary configurations. Our analysis combines publicly available documentation, open-source code, technical research, and observable behavior.

What is candidate generation?

Candidate generation is the stage where a recommendation system identifies potential posts that could be shown to a user. X's open-source architecture includes multiple candidate sources and a candidate-pipeline framework before content reaches later ranking and selection stages. (GitHub) In simple terms: Find possible posts → filter them → score them → rank them → select what to show.

What is ranking?

Ranking is the process of estimating which candidates are most relevant to a particular user and determining their position or likelihood of being selected. X's published systems include ranking components and models that use user context and predicted engagement or other actions. (GitHub)

What is Phoenix?

Phoenix is a major component of the newer open-source X recommendation architecture. The published repository describes Phoenix as the Grok-based transformer system used for recommendation ranking, with retrieval and ranking components within the broader For You pipeline. (GitHub)

What is Grok doing in the recommendation system?

The newer open-source X recommendation system describes a Grok-based transformer being used to understand user context and rank candidate posts. The system is designed to predict multiple possible user actions rather than reducing relevance to a single engagement metric. (GitHub)

Does X only care about Likes, Replies, and Reposts?

No. X's published code identifies a broader collection of explicit and implicit user signals, including actions such as profile visits and post clicks. (GitHub) The exact importance of individual signals can depend on the recommendation system, model, context, and configuration.

Can I use Under The Hood Analyzer to manipulate X's algorithm?

Under The Hood Analyzer is intended for analysis and understanding, not manipulation. We encourage creators to use analytics to understand their content, improve its quality, and make better decisions rather than attempting to manufacture artificial engagement.

Does Under The Hood Analyzer have access to private X data?

No. Unless a user explicitly provides information through an available feature, Under The Hood Analyzer does not have access to X's private internal databases, private recommendation models, or confidential user information.

Why does Under The Hood Analyzer sometimes say “may” instead of “does”?

Because there is an important difference between what the source code demonstrates and what we can conclusively claim about every production recommendation. For example: “The open-source code contains this mechanism” is a technical statement. “This mechanism caused your post to receive 10,000 impressions” is a much stronger claim that requires information we generally cannot observe. We prefer evidence over certainty.

How accurate are Under The Hood Analyzer's analyses?

Accuracy depends on the information available, the version of the underlying open-source code, the analysis method, and changes made by X. We aim to distinguish between:

• Confirmed: directly supported by published source code or documentation.
• Observed: supported by measurable or repeated behavior.
• Inferred: a reasonable interpretation based on available evidence.
• Unknown: information that cannot currently be established.

This distinction is important because recommendation systems are complex and continuously evolving.

Why does my reach suddenly drop?

A change in reach does not automatically mean that your account has been shadowbanned. Reach can change because of content relevance, audience behavior, competition for attention, recommendation decisions, engagement patterns, filtering, ranking, or changes to the platform itself. A single low-performing post is not enough evidence to identify the cause.

Can Under The Hood Analyzer detect a shadowban?

No tool should claim to definitively detect a “shadowban” without access to X's internal enforcement and recommendation systems. We can analyze observable patterns and identify potential distribution changes, but that is different from proving that X has applied a specific internal restriction to an account.

Does posting more always increase reach?

No. Posting more creates more opportunities to publish, but it does not guarantee that every post will receive greater distribution. Recommendation systems evaluate individual pieces of content in context, and increased posting frequency can also change audience behavior and engagement patterns.

Does deleting a post hurt my account?

There is no universal public rule establishing that deleting a post automatically causes a permanent reach penalty. Individual account behavior, content history, and recommendation eligibility can involve factors that are not completely visible from the public source code. Avoid treating isolated anecdotes as proof of a platform-wide rule.

Why does a post sometimes stop getting recommendations?

A post may stop receiving additional distribution for many possible reasons. Its relevance to the available audience can change, new candidates can compete for attention, predicted engagement can change, or the post can become less likely to be selected by later ranking stages. The open-source architecture demonstrates that recommendation involves multiple stages rather than a single permanent “viral” switch. (GitHub)

How often does X change its algorithm?

X's recommendation systems can evolve over time through new code, models, configurations, experiments, and infrastructure changes. That is why Under The Hood Analyzer should not be treated as a permanent rulebook for X. When discussing algorithm behavior, always consider the version and date of the underlying evidence.

What is the difference between X's 2023 and newer open-source algorithm code?

The original public repository released in 2023 documents a broad recommendation architecture containing services, user signals, community detection, embeddings, and ranking components. (GitHub) The newer x-algorithm repository provides a more recent For You recommendation architecture, including candidate pipelines, Phoenix retrieval/ranking, and a Grok-based transformer. (GitHub) For current analysis, it is therefore important to distinguish between older Twitter-era source code and newer X-era recommendation code.

Where can I read the source code myself?

The primary public repositories are available on GitHub:

• X Recommendation Algorithm repository (2023)
• X AI Recommendation Algorithm repository (Newer)

We encourage readers to examine the source directly rather than relying solely on third-party interpretations.

Is Under The Hood Analyzer's information official X information?

No. Where we discuss X's open-source code, we may quote or interpret publicly available technical material. However, our explanations and analyses are independent and should not be interpreted as official statements from X.

Where does Under The Hood Analyzer get its information?

Our analysis may use:

• X's publicly released source code
• X engineering documentation
• Public technical documentation
• Publicly available datasets and information
• Observable platform behavior
• Independent technical analysis
• Statistical and analytical methods

The underlying evidence will vary depending on the topic.

Can the information on this website change?

Yes. X can modify its recommendation systems, models, ranking logic, policies, and product features. We may update our articles and analyses when new public evidence becomes available. For important decisions, readers should verify information against the latest primary sources.

What is the main purpose of Under The Hood Analyzer?

To replace algorithm myths with evidence-based analysis. Instead of asking: “What secret trick makes X go viral?” we want to ask: “What can the available evidence actually tell us about how X recommendations work?” That is the idea behind Under The Hood Analyzer.