How to Spread a Lie

Inspiration

Misinformation doesn’t just spread, it evolves. A headline can start grounded in reality and, as it moves from source to influencer to audience, gradually shift into something entirely different. We were interested in that transformation process: how small changes, repeated across a network, can compound into large distortions. This project explores how truth bends, stretches, and sometimes breaks as it travels.

What it does

How to Spread a Lie is a misinformation propagation simulator. Headlines originate from Sources, pass through Influencers, and reach Agents—mutating at every step through exaggeration, omission, framing shifts, and fabrication.

Instead of treating misinformation as a static object, the simulator models it as something dynamic. It tracks:

• How far a headline drifts from its original truth
• Who contributed to that distortion along the way
• Whether the most viral content is also the most distorted

It also allows you to test interventions, like introducing a fact-check mid-simulation, to see whether the system can recover—or if the damage is already done.

How we built it

We designed a network of interacting actors:

• Sources generate initial headlines with a baseline truth value
• Influencers reshape and amplify content based on bias, incentives, and reach
• Agents consume, interpret, and propagate information further

Each interaction applies transformation functions that introduce measurable “drift.” Over multiple cycles, the system logs how each actor modifies the headline, building a traceable path from truth to distortion.

The project is implemented in Python and Javascript, with modular components for actors, transformations, and simulation cycles, making it easy to experiment with different behaviors and scenarios.

Challenges we ran into

• Modeling abstract ideas like truth and bias required simplifying assumptions that still felt realistic
• Designing believable transformation rules without real-world data was difficult
• Quantifying “drift” in a consistent way was harder than expected
• Keeping the simulation expressive without making it overly complex

Accomplishments that we're proud of

• Built a working simulation that captures multi-step information cascades
• Created a system to quantify and track distortion at each hop
• Enabled meaningful comparisons between virality and accuracy
• Developed a flexible framework for testing interventions like fact-checking

What we learned

• Misinformation is rarely created in one step—it emerges gradually
• Small, seemingly harmless changes can compound into major distortions
• The structure of a network matters just as much as the content itself
• Defining and measuring abstract concepts forces you to think precisely about what you care about

What's next for How To Spread A Lie

• Use AI to evaluate how far truth drift from the truth more accurately
• Learn transformation patterns from real-world data instead of relying on intuition-based heuristics
• Expand the model to include richer network dynamics and feedback loops

Built With

• css
• flask
• html
• javascript
• lovable
• pytest
• python