Part 2 of Dissecting the X Algo Code That Decides Who Wins and Who Vanishes

You can drop a fire post and still get zero reach. The algo is not counting likes. It's reading who liked, when, and why.

Your trust layer decides if you pump or get shadow-buried.

The first hour tests your frens circle

That first hour after you post decides everything. From what we can see, the logic appears baked in. It watches who shows up, how fast, and how close they are to you.

EngagementDataRecordFeatures tracks the early moves: likes, replies, reposts. Relationship features like ViewerFollowsConversationRootAuthor and ViewerVerbsAuthor check whether those signals come from real frens or randoms.

If your inner circle hits early, you get lift. If the room stays quiet or the accounts look weak, the post sinks.

Think of it like a club door: ten people isn’t good or bad on its own. Ten loyal regulars who always bring a crowd? Green light. Ten tourists who never come back? The bouncer waves them through and forgets them.

What “Trusted Friends” actually means in code

TrustedFriendsFeatures is not a vibe. It is literal logic. In reverse-engineered descriptions and internal leaks, people refer to features like ViewerIsTrustedFriendOfTweetAuthor and ViewerIsTrustedFriendTweetAuthor. These appear to check whether a viewer is on your trusted list or owns it. When those conditions hit, the pipeline may treat early interactions as higher quality. It is the platform’s way of saying a real community is forming around your post, not a random flash mob.

Translation for practice:

Have a core circle ready to show up in the first hour. Likes, replies, and reposts from those people carry more weight than the same actions from total strangers. Keep it organic and human. Consistency beats volume.

Your social score is real, and it moves

Beneath the feed, the system is charting your trust profile. You never see it, but you feel it every time a post either flies or flatlines.

Every file you shared helps draw that picture.

• TweetSafetyLabels and ViolationLevel mark when your content brushes up against risk.
• ToxicReplyFilterFeature flags messy threads and notes when you are the one hosting them, because toxic threads drag reach.
• TweetPerspectiveFeatures logs reports. Every time someone hits that button, it adds friction to your flow.
• ModerationFeatures and TweetIsModerated stamp your account with a quiet caution tag.
• FosnrRelationshipFeatures checks whether your risky posts are being touched by followers or randoms. If it is mostly strangers, the system sees smoke.

Think of it like a credit score that updates in real time. Clean plays build trust. Spam, toxicity, or low-signal loops bleed points. Even when you post something brilliant later, your limit still depends on your recent history within each model window or time slice. The system adjusts, but the memory of your last few moves always shapes what happens next.

Bots and AI agents burn the score

Coordinated activity is detectable, and technically a violation of X policy. The codebase appears to include spam and safety labeling and filters for toxic reply patterns. Those catch loops, rings, and inorganic bursts. AI generated junk has the same footprint over time. It looks repetitive, low context, and engagement comes from other low trust accounts. That combination is a social score tax. It might pump a number for a day. It makes distribution harder for weeks.

Use this test before you post or join an “engagement group.” If the plan relies on people who never interact with you outside the swap, or on accounts that only touch farmed posts, the cost to your score is bigger than the boost.

Build a clean trusted circle

Start with 5 to 15 real people whose audiences overlap with yours. Trade signal, not spam. Reply with takes that add context. Share when something truly fits your crowd. Keep it varied so the pattern never looks mechanical.

If you do not have that yet, I set up a Creators Alliance on Telegram where we help each other the right way. You must be willing to show up for others.

Avoiding the echo chamber

Trusted circles are powerful, but they can quietly become traps. When the same few people always like, reply, and repost, your trust layer stops expanding. The algo sees it through RelationshipFeatures and ViewerInteractionGraph. It flags a tight cluster with no fresh edges. Safe, but stale.

Closed loops cap your reach. The model starts thinking your content only matters to one micro-crowd, so it stops testing you in new feeds. That is how accounts with solid engagement still plateau. The system wants discovery, not comfort.

To grow, you need overlap. Rotate who you engage with. Tap into nearby circles that share some DNA but bring different energy. Drop thoughtful replies outside your usual lane. Variety in tone and audience looks alive to the algo and keeps your score moving.

A healthy signal spreads like roots through soil, not vines choking the same post. Build bridges, not mirrors.

Rules for health:

• Keep the mix fresh. If the same five people are the only ones who ever engage, you look like an echo chamber.
• Avoid low quality accounts. One toxic chain can drag a week’s reach.
• Make it human. Answers with real thought do more than empty emojis.

Recovering when the score dips

Stop the bleeding. Cut the low quality loops and mute the toxic threads. Go back to original posts that actually say something and keep your media clean and hydrated. In that first hour, show up. Ask real questions, reply with substance, and let your circle set the tone. Those early human signals rebuild velocity faster than any pod. For the love of all things holy, do not use bots. Short term sugar, long term crash. Moderate your threads, feed the signal, not the trolls. Treat it like a detox after an all inclusive vacation. A week of clean input resets the baseline.

Closing and What’s Next

Part 1 covered frequency, timing, and pinned leverage. It showed why 1 to 3 posts a day, smart windows, and weekly pin rotation set your baseline.

Part 2 showed how the first hour and your inner circle shape the trust layer and your social score. It explained why bots and AI agents are a trap, and how to build signal the system respects.

Next up is Part 3. We will break down media types and metadata checks. You will see how DMCA, geo restrictions, NSFW flags, and clean hydration influence reach, and how to design the first frame of a video so the auto thumbnail earns clicks.

Make sure you follow and turn on alerts so you catch Part 3 the moment it drops.