A Careful Review of the Current State Before the Release of the Latest Algorithm Source Code (AI Illustration from March 29, 2026)

Elon and Nikita have announced that Grok will be fully deployed for algorithmic control and that the source code for the latest algorithm will be released soon.

Perhaps in connection with this, the following information about the algorithm has once again become frequently visible on X.

• “Posting external links in the body text causes deboosting”

• “Links should be placed in the reply section, not in the body text”

• “One reply is equivalent to 75 likes”

• “Replying is equivalent to 150 likes”

• “The weight of likes has been reduced”

And things like that.

To give you the conclusion first,
they are all wrong.

For some reason, this kind of incorrect information is more believable and spreads more easily than the correct information.

Why do you believe it?

So, immediately after the source code for the latest algorithm was released in January 2026, Ipersonally checked all the code and shared the correct information,but I would like to summarize it again.

💡Conclusion first

• The For You timeline is created through a process of candidate collection, information enrichment, filtering, scoring, and selection.

• The specific values for the weights of each action are not public.

• There is no verifiable evidence that posting external links in the body text causes deboosting for that reason alone.

Now, please see the details.

📝Specific weight values are private

Regarding the score calculation for the For You timeline, it is explained in the public documentation as a Weighted Scorer.

It is organized there such that the final score is a weighted synthesis of multiple behavioral predictions.

Final Score = Σ (weight_i × P(action_i))

The structure is such that the "weight_i" included in the calculation formula is applied as a weight to each action, such as favorite, reply, and repost.

This is what is known as "weighting."

However, the specific values of the weights have not been disclosed.

In the structure of the released code, home-mixer::params is “Excluded from open source release for security reasons”, and is excluded from the release.

At this point, looking only at the released code, it is clear that one cannot

• "1 reply is equivalent to 75 likes"

• "Replying is equivalent to 150 likes"

• "The weight of likes has become lighter"

or similar claims, cannot be asserted.

You cannot definitively state something that has not been disclosed based on the released code.

In other words, these claims have no basis.

Why is incorrect information being spread?

Because people are taking Grok's answers at face value.

Since the introduction of agents with the update to Grok 4.20 Beta, accuracy has improved compared to before, but it still sometimes answers by citing old open source code, user posts, and information from the internet.

Most incorrect information is based on old open source code.

📝 How the Recommended Timeline is Created

Looking at the released README and the description of the Candidate Pipeline, it can be seen that the recommended timeline is roughly created through the following flow.

1. Collect candidates

Candidates are mainly collected from two sources.

• Thunder
  Posts originating from accounts you follow

• Phoenix Retrieval
  Candidates found via machine learning, including those from non-followed accounts

In other words, the recommended timeline is structured so that not only posts from accounts you follow, but also posts from accounts you do not follow are included as candidates.

2. Adding necessary information to candidates

Next, we supplement each candidate with information such as the body text, author details, video information, and subscription status.

In the public documentation, this process is described as Hydration.

3. Removing unnecessary candidates

After that, we filter out posts that should not be shown.

The filters that can be confirmed in the public documentation include the following:

• Removal of duplicates

• Removal of posts that are too old

• Removal of your own posts

• Removal related to blocks and mutes

• Removal of muted keywords

• Removal of posts already seen

• Removal of posts already displayed

• Removal of posts that do not meet subscription criteria

and so on.

4. Scoring

The main components here are the Phoenix Scorer and the Weighted Scorer.

The Phoenix Scorer makes multiple predictions about how a user is likely to react (engage) to that post.

The public documentation lists the following behavioral predictions.

• favorite

• reply

• repost

• quote

• click

• profile_click

• video_view

• photo_expand

• share

• dwell

• follow_author

• not_interested

• block_author

• mute_author

• report

Then, the Weighted Scorersynthesizes them with weights to create a final score.

Furthermore, the Author Diversity Scorer makes adjustments to ensure that posts from the same author do not appear too frequently.

5. Select the top results

Finally, they are sorted by score and the top candidates are selected.

After that, post-filters are applied as necessary to determine which posts are displayed. For example, items with visibility issues such as deleted content, spam, or violence/gore are excluded.

📝 What can be confirmed in the public code

As far as the list of filters and score structures in the public documentation show, the explanation that 'posts with external links in the body are penalized' cannot be confirmed.

At least within the publicly available scope,

• a penalty rule specifically for pasting external links in the body

• a filter that uniformly lowers scores because of external links

• code that can be read as 'should be pasted in the reply section instead of the body'

cannot be confirmed.

Therefore,

Pasting external links in the body is deboosted. Therefore, they should be pasted in the reply section.

cannot be asserted based on the public code.

📝 Why one might feel that 'posts with links are less likely to grow'

From here on, this is not about what is written in the public code, but my interpretation based on the publicly available algorithm structure, so it includes speculation.

What can be understood from the source code is not a mechanism that imposes a penalty on the presence of a link itself, but a mechanism that predicts how users are likely to react to that post and determines the ranking based on the combined results.

Considering this structure, for example,

• posts with almost no explanation

• posts where the reason to view them is not conveyed

• posts with thin context

• posts that look like advertisements

will, as a result,

• reply

• repost

• quote

• favorite

• dwell

There is a possibility that predictions for positive actions like these may become harder to increase.

Depending on the content of the link,

• not_interested

• mute_author

• block_author

there is also a possibility that predictions for negative actions like these may increase.

In other words,

it is not that 'it is deboosted because there is an external link,' but rather that the final score is unlikely to rise if the post is in a format that is difficult to react to.

It is considered more consistent with the source code to interpret it this way.

📝 What is important is not the 'presence or absence of a link' but the 'context'.

This is also an interpretation based on the structure of the released code.

Even if an external link is attached,

• it is clear what the link is about

• the meaning is conveyed by the post alone

• there are images or supplementary information

• There is a reason to read or a reason to react

If it is a post with context like this, the prediction for positive behavior is more likely to increase.

Conversely,

• links only

• no explanation

• no images

• no context

Posts like these are less likely to see growth in engagement predictions, and as a result, it is quite conceivable that they will be less likely to appear in the For You timeline.

It is not that external links themselves are the problem, but rather that the lack of context as a post is likely influencing the results

.

Looking only at the structure of the code, this understanding is more natural, isn't it?

To summarize up to this point, what can be confirmed from the code is as follows.

• The For You timeline is composed of candidate sourcing, information enrichment, filtering, scoring, and ranking.

• The score is created by synthesizing multiple behavior predictions with weights.

• The specific values of those weights are not public.

• In the public scope, there is no evidence to confirm that external links are directly deboosted because they are pasted in the body of the post.

Therefore,

because pasting external links in the body is deboosted, you should always paste them in the reply section instead

—this claim is not a conclusion that can be drawn from the public code, at least.

💡Summary

What can be confirmed from the source code of X's latest algorithm is that the For You timeline is created through the following flow.

1. Collect candidates

2. Supplement information

3. Exclude unnecessary candidates

4. Score based on multiple action predictions

5. Select the top ones

And here, the following two points are particularly important.

• The specific values for the weight of each action have not been disclosed

• No explicit evidence can be confirmed that directly deboosts pasting the body of an external link

Based on these points,

• "Pasting an external link in the body will cause a deboost"

• "Links should be placed in the reply section, not in the body"

• "One reply is equivalent to 75 likes"

• "Replying is equivalent to 150 likes"

• "The weight of a like has become lighter"

I think you can see that these claims are groundless and incorrect.

It is easy to believe what people considered influencers say, or things that are convenient for us (because we want to believe them).

"Who says it"

rather than

"What is said"

I believe it is important to scrutinize information from the perspective of.

The subject of "who is saying it" also includes "Grok." It is a major mistake to assume that because it is an AI created by X, all answers regarding X are correct.

As those in the AI illustration community know well, even the slightest difference in a prompt results in a significantly different illustration. Furthermore, even with the exact same prompt, you will not get the exact same illustration.

Grok's answers are the same.

If the question (prompt) is different, the answer (output) will also be different.

It is strange to adjust prompts and use trial and error when AI illustrations don't match your intent, yet blindly accept Grok's answers at face value.

As I wrote at the beginning, the source code for the latest algorithm will be released soon.

Let's be careful not to be swayed by shallow, incorrect information meant for farming impressions.

Regarding the algorithm, I am referring to Fukubun-san's GPTs. They have published useful know-how in addition to X analytics tools, so please check them out.

Tennen's Subscription
Tennen's note
Tennen's YouTube
Tennen's Pinterest
Tennen's Instagram
Tennen's Threads
Tennen's𝕏

◆ Mecha Girl

An illustration of a mecha girl with vivid red hair and striking golden eyes. Her design centers on a teal bodysuit, with black and yellow armor parts arranged three-dimensionally across the chest, waist, and arms, balancing lightness with mechanical density. The compact headgear and hard parts around the neck enhance the futuristic impression, while her gentle expression and slightly leaning pose add a sense of warmth and approachability. The vivid color scheme and flowing hair create a bright and refined mecha girl design.

An illustration of a mecha girl with vivid red hair and striking golden eyes. Her design centers on a teal bodysuit, with black and yellow armor parts arranged three-dimensionally across the chest, waist, and arms, balancing lightness with mechanical density. The compact headgear and hard parts around the neck enhance the futuristic impression, while her gentle expression and slightly leaning pose add a sense of warmth and approachability. The vivid color scheme and flowing hair create a bright and refined mecha girl design.

＼ Dancing, Posing, Approaching ／
See all past Mecha Girl works at once!

List of Mecha Girl posts

◆ Action Mecha Girl

An illustration of a mecha girl in a gray sailor uniform fused with metallic frames and finely engineered joint mechanisms. The exposed mechanical structure from the back to the waist and the heavily detailed armor integrated into her limbs are rendered with impressive density, while her sleek black bob haircut and golden eyes add a soft contrast to the inorganic design. The blade-like headgear and the sword on her back sharpen the overall impression, emphasizing a quiet tension and a distinctly Japanese cybernetic atmosphere.

An illustration of a mecha girl in a gray sailor uniform fused with metallic frames and finely engineered joint mechanisms. The exposed mechanical structure from the back to the waist and the heavily detailed armor integrated into her limbs are rendered with impressive density, while her sleek black bob haircut and golden eyes add a soft contrast to the inorganic design. The blade-like headgear and the sword on her back sharpen the overall impression, emphasizing a quiet tension and a distinctly Japanese cybernetic atmosphere.

List of Action Mecha Girl posts

◆ Relaxed Mecha Girl

A profile illustration of a mecha girl unified by pure white hair and a soft, pale color palette. She wears a large blade-like headgear and circular mechanical unit, while white wings mechanically connected to her back create a quiet fusion of inorganic structure and angelic imagery. The minimal composition, restrained exposure, and gentle lighting enhance the sense of transparency and sanctity, resulting in a fantastical scene where delicate expression and precise mechanical detail exist in perfect harmony.

A profile illustration of a mecha girl unified by pure white hair and a soft, pale color palette. She wears a large blade-like headgear and circular mechanical unit, while white wings mechanically connected to her back create a quiet fusion of inorganic structure and angelic imagery. The minimal composition, restrained exposure, and gentle lighting enhance the sense of transparency and sanctity, resulting in a fantastical scene where delicate expression and precise mechanical detail exist in perfect harmony.

List of glowing Relaxed Mecha Girls

◆ Armored Schoolgirl

An illustration of a mecha girl wearing a white sailor-style top with a red scarf and denim-like short bottoms. Her left arm and both legs are replaced with white-and-gold mechanical prosthetics, rendered with precise armor plating and detailed joint structures. The dual-sword setup strengthens the armed, combat-ready impression, while the contrast between the inorganic mechanical elements and the softness of exposed skin stands out vividly. Her upright pose and strong gaze create a striking mecha girl design that combines refinement with clear martial presence.

An illustration of a mecha girl wearing a white sailor-style top with a red scarf and denim-like short bottoms. Her left arm and both legs are replaced with white-and-gold mechanical prosthetics, rendered with precise armor plating and detailed joint structures. The dual-sword setup strengthens the armed, combat-ready impression, while the contrast between the inorganic mechanical elements and the softness of exposed skin stands out vividly. Her upright pose and strong gaze create a striking mecha girl design that combines refinement with clear martial presence.

List of Armored Schoolgirl posts

"High School Girl" Summary

◆Samurai Mecha Girl

List of Samurai Mecha Girl Posts

"Samurai Mecha Girl" Summary

◆Mecha Girl

An illustration of a mecha girl wearing a white sailor uniform with a blue collar and a red scarf. In contrast to her soft expression as she smiles while resting her chin on her hands, large blue mechanical units are densely arranged on both sides of her head, with lenses, wiring, and joint structures rendered in high density. Her clear blue eyes and natural hair texture contrast beautifully with the inorganic devices, creating an impressive composition where approachability and precise mechanical depiction coexist.

An illustration of a mecha girl wearing a white sailor uniform with a blue collar and a red scarf. Her soft smile and relaxed pose with her chin resting on her hands contrast beautifully with the large blue mechanical units densely arranged on both sides of her head, where lenses, cables, and articulated structures are rendered in high detail. Her clear blue eyes and natural hair texture create a striking balance with the inorganic machinery, resulting in a composition that blends warmth with precise mechanical design.

◆List of Mecha Girl Posts

◆Elf Girl

An illustration of a blonde, green-eyed elf girl wearing a white flower crown, sitting surrounded by plants. Her light white dress and pale blue outer layer create a soft atmosphere, and her gentle smile with flushed cheeks enhances her approachability. Her long ears, floral accessories, and plant-themed necklace add a fantastical impression, and her natural, barefoot pose makes for a lovely and open-feeling piece.

An illustration of a blonde, green-eyed elf girl wearing a white flower crown and sitting among blooming plants. Her light white dress and pale blue outer layer create a soft, airy mood, while her faint blush and gentle smile add a warm sense of approachability. Her long ears, floral accessories, and plant-themed necklace enhance the fantasy atmosphere, and her relaxed barefoot pose gives the piece a graceful and open feeling.

List of Elf Girl Posts

◆Heartbroken Girls

An illustration of a schoolgirl wearing a white short-sleeved shirt, a pale blue ribbon, and a gray pleated skirt, smiling as if looking back. Her soft brown hair and flushed expression are approachable, and her lightly swaying skirt adds natural movement. One speech bubble reads "Is there still more?" and the other features the large character for "Pulse," creating a composition that drifts with a quiet sense of unease through the contrast between her calm expression and the striking text.

An illustration of a schoolgirl wearing a white short-sleeved shirt, a pale blue ribbon, and a gray pleated skirt, smiling as she looks back over her shoulder. Her soft brown hair, lightly blushing cheeks, and gently swaying skirt create a warm and approachable mood. The speech balloons read “Is there still more?” and a bold “Pulse,” and the contrast between her calm expression and the striking text gives the scene a subtle sense of unease.

List of Heartbroken Girls Posts

"Heartbroken Girls" Summary

◆X Revenue Amount Publicly Available

I am publishing my bi-weekly creator earnings along with a detailed analysis. 47,000 followers, 5 illustration posts every day, and absolutely no engagement in reply threads. I hope this serves as a reference for those who are monetizing on X or aiming to do so.

Natural's𝕏