Learn
How Social Feeds Learn About You
Explore how your interactions can influence what appears in a personalized feed.
Feed X-Ray is an educational simulation. Its weights and decision rules are fictional and are not the proprietary algorithms of Instagram, TikTok, YouTube, LinkedIn, X, or other platforms.
Your session
Your Session, Explained
Your simulation has not collected behavioral signals yet.
1 · Pipeline
The Recommendation Pipeline
A feed is not one magic score. It is a sequence of guesses, each one imperfect.
01
User interactions
People like, save, skip, watch, and tap through posts. Those actions are the raw material a feed can observe.
02
Behavioral signals
Actions such as likes, saves, and watch time can provide clues about what someone may find interesting.
Try liking or saving a post to see this stage update.
03
Interest profile
A recommendation system may summarize these clues into a representation of inferred interests.
Try liking or saving a post to see this stage update.
04
Candidate content
Large systems may first retrieve a smaller set of potentially relevant posts from a much larger collection.
- Real-world concept
- Candidate retrieval from a large content pool.
- Feed X-Ray implementation
- All 20 local posts are available to the ranking engine. Feed X-Ray does not implement production-scale candidate retrieval.
05
Ranking
Candidate posts can be scored and ordered based on multiple factors.
Try liking or saving a post to see this stage update.
06
Diversity / exploration
Some systems may introduce different content to avoid showing only one type of post, and may show unfamiliar topics to learn more.
07
Personalized feed
The visitor sees an ordered mix of organic posts, plus a separately chosen sponsored placement.
2 · Signals
What Do Your Actions Tell the System?
Behavioral signals are imperfect. Inferred interests are educated guesses — not verified personal preferences.
Like
A positive preference signal.
A like is still just one action. It does not prove a lasting preference.
Save
May indicate stronger interest or future intent.
People also save things to revisit later, not only because they enjoy them.
Watch time
Longer engagement can indicate interest, but does not always mean enjoyment.
Watching something for a long time does not necessarily mean you liked it. It might be surprising, confusing, or hard to look away from.
Quick skip
May indicate low interest, but can also happen for unrelated reasons.
A skip can mean “not now,” a mis-tap, or simply scrolling past — not a verified dislike.
Not Interested
Explicit negative feedback.
Even explicit rejection is about this content in this moment, not a person’s whole identity.
3 · Ranking
Why Some Posts Move Higher
Feed X-Ray uses a fictional educational ranking model. These factors are teaching weights, not a production algorithm.
Category affinity
Posts related to your inferred interests may receive a higher ranking score.
Technical detail
Feed X-Ray uses a deterministic weighted scoring function. Category affinity is the largest educational weight.
Freshness
Newer content may receive a small boost.
Technical detail
Freshness is a small demo factor so the catalog’s recency still has a voice.
Content quality
A post can rank a little higher when the demo treats it as higher quality.
Technical detail
Quality is a static teaching signal on each fictional post — not a trained quality model.
Popularity
Widely engaged content may get a modest lift.
Technical detail
Popularity here is a canned demo value, not live collaborative filtering.
Exploration
A system may show unfamiliar topics to learn whether you are interested.
Technical detail
Exploration can give weaker-affinity posts a chance to appear near the top.
Engagement
Saves, shares, and similar actions on a post can support its rank.
Technical detail
Post-level memory of likes, saves, and shares contributes a small engagement term.
Diversity penalty
A system may avoid showing too many similar posts consecutively.
Technical detail
Feed X-Ray uses greedy placement with diversity penalties so adjacent repeats are discouraged.
Negative feedback
Content similar to something you rejected may rank lower.
Technical detail
Not Interested and repeated skips reduce a post’s educational score.
Seen-post penalty
Posts you have already seen may be slightly less likely to stay on top.
Technical detail
A seen-post penalty reduces repetition after a post has already appeared.
Beginner vs technical
Beginner: Your interests influence which posts move higher.
Technical: Feed X-Ray uses a deterministic weighted scoring function and greedy placement with diversity penalties.
4 · Unexpected posts
Why Does My Feed Show Things I Didn't Ask For?
Exploration and diversity are related but not the same. Both can make a feed feel less predictable.
Exploration
A system may show unfamiliar topics to learn whether the user is interested.
Diversity
A system may avoid showing too many similar posts consecutively.
Illustrative · without diversity
- Technology
- Technology
- Technology
- Technology
Illustrative · with diversity
- Technology
- Travel
- Technology
- Food
The sequences above are labeled illustrative examples — not your actual feed.
5 · Feedback loop
The Feedback Loop
Recommendation systems can create feedback loops because what you see influences what you interact with next.
- 1
You engage with Technology
- 2
Technology interest increases
- 3
More Technology posts rank higher
- 4
You encounter more Technology content
- 5
Further engagement may strengthen that interest
- Recommendations can reinforce existing interests.
- Exploration may introduce new topics.
- Negative feedback can change future rankings.
- A recommendation is not proof of someone's identity or permanent preferences.
- Filter bubbles are a risk to watch, not an inevitable outcome.
6 · Two systems
Organic Recommendations vs Advertising
These are separate decision processes in Feed X-Ray. Real platforms may have more complex advertising systems.
Organic recommendations
Inputs
- Behavioral signals
- Interest profile
- Content ranking factors
Output
Order of organic posts
Sponsored content
Inputs
- Commercial signals
- Ad personalization setting
- Available fictional ads
Output
Selected sponsored placement
Disabling ad personalization does not necessarily remove advertising. It may change how ads are selected.
Feed off / Ads off
Current sessionOrganic posts stay in catalog order. Sponsored content can still appear, selected contextually.
Feed on / Ads off
Organic ranking can reorder posts from inferred interests. Ads stay contextual.
Feed off / Ads on
The organic feed stays neutral while sponsored selection may use the commercial profile.
Feed on / Ads on
Organic ranking and sponsored selection can both personalize — still as separate processes.
7 · Limits
What Feed X-Ray Simulates — and What It Doesn't
Feed X-Ray is an educational simulation. Its weights and decision rules are fictional and are not the proprietary algorithms of Instagram, TikTok, YouTube, LinkedIn, X, or other platforms.
Feed X-Ray implements
- Local behavioral events
- Deterministic interest scoring
- Organic ranking
- Diversity and exploration
- Commercial affinity scoring
- Fictional sponsored selection
- Ranking explanations
- Engineer observability
Feed X-Ray does not implement
- Real social platform algorithms
- Machine learning model training
- Embeddings
- Collaborative filtering
- Production candidate retrieval
- Large-scale distributed infrastructure
- Real advertising auctions
- Real advertiser integrations
- User identity tracking
- Backend behavioral storage