Research Principle: “Our feeds shouldn’t value confrontational engagement over complementary content, nor should they obsequiously feed us only content we agree with. They should strike a balance and mend social disagreements rather than make them worse.” — Michael Bernstein, Stanford HAI Senior Fellow

1. Algorithm Objective Tuner Balanced Complementary Bridging

Bridging / Cross-Group Consensus 0.50

Rewards posts liked by diverse ideological clusters rather than single factions.

Ideological Affinity (Homophily) 0.35

Prioritizes in-group agreement and familiar network confirmation.

Confrontational Outrage Engagement 0.15

Boosts high-velocity replies, heated quotes, and viral disagreement.

Polarization Index
0.28
Low Division
Cross-Ideology Exposure
64.5%
Healthy Exchange
Outrage Amplification
0.32
Damped Hostility
Community Health Score
88/100
Constructive
Simulated Network Graph 3 Clusters • 12 Bridge Posts
Cluster A
Cluster B
Bridge Post

2. Live Curation Feed Stream

Displaying top 12 posts scored by active multi-objective model
Real-Time Ranking Active
Select a post above to inspect ranking weight breakdown
Click on any post in the feed to see how bridging consensus, ideological affinity, and outrage dynamics calculate its exact position.

3. Stanford HAI Algorithmic Policy Brief Export

Durable summary of simulated objective weights, exposure telemetry, and mitigation insights.


    

How the saved weighted ranking objectives change a synthetic feed

Read the explanation

The saved simulator ranks synthetic posts by a weighted sum of three hand-assigned features: bridging, affinity and outrage. For its default weights point five, point three five and point one five, example record p seven scores point six one two. Record p eight scores point four six five. The bars use one thousand pixels per score unit. They show this pairwise comparison, not a claim that either record is highest among all twelve. The source scores are authored examples; they are not measured truth, civic quality or predictions of real users. The outrage preset changes the weights to point zero five, point two and point seven five. Keeping the same two feature vectors, p seven now scores point two zero two five and p eight point seven nine five. The relative order reverses because the second example has an outrage feature of point eight five versus point zero eight. These bars keep the same one thousand pixels per score unit. Changing the objective changes what the ranking rewards; it does not supply new evidence that a post is accurate or that a real community becomes polarized. The page also labels an index polarization. Its source computes point four five times affinity weight plus point six five times outrage weight, minus point four times bridging weight, plus point one five, then clamps the result. The default weights give about point two zero five before display rounding; the outrage preset gives about point seven zero seven five. The bars use one thousand pixels per index unit. This is a hand-written proxy formula, not an observed social outcome. Export preserves the selected topic, weights, rounded metrics, top three scores and threshold-written recommendation; its research attribution metadata does not validate those synthetic results.

Super generates helpful tools and automates fact-checking across the internet proactively. If you enjoyed this tool, build your own with Super and share it with a friend.