I've always been fascinated by how science can turn a simple question into something worth exploring. Take a pendulum, for example. If you replace its weight with something heavier, will it swing more slowly? It sounds like a simple question, but people often have different answers.

That got me thinking about Twitch. Millions of people watch streamers play games, build things, and share their interests. Viewers can chat and react, but what if they could take part in a scientific experiment? What if they could make predictions, explain their reasoning, and see whether the evidence supports what they believed?

That's where the idea for Proof of Wonder came from.

I wanted to build something that makes science feel less like watching someone explain an answer and more like discovering that answer together.

What it does

Proof of Wonder turns a Twitch livestream into an interactive scientific investigation.

Before an experiment begins, viewers can submit predictions directly through Twitch chat. They can also explain why they think something will happen.

The streamer then conducts the experiment, and Proof of Wonder collects and analyzes the observations. Instead of simply declaring a winner, the application examines the measurements, checks the quality of the evidence, and explains what the results actually support.

At the end, a feature I call Discovery Theatre brings everything together. It presents the experimental findings alongside the community's predictions, giving everyone a chance to see how their expectations compare with the evidence.

The first experiment investigates whether changing a pendulum's mass affects its period of oscillation.

One thing I particularly wanted to get right was uncertainty. Sometimes an experiment doesn't provide enough reliable evidence to reach a conclusion. Proof of Wonder is designed to acknowledge that rather than force an answer.

How I built it

I built Proof of Wonder with the help of OpenAI Codex, using it throughout development to explore implementation approaches, write and improve code, investigate bugs, and test the application.

The system connects several components that need to work together.

I used Twitch authentication and EventSub to connect the application to Twitch chat, allowing viewers to submit predictions using commands such as !predict increase, !predict same, and !predict decrease.

For the experiment itself, I built a video-based measurement workflow that tracks a visible marker on a pendulum. The application uses video timestamps and the marker's movement to estimate the oscillation period.

The backend checks the observations and calculates the results before they can be accepted as experimental evidence. I also added quality checks to identify problems such as missing observations, irregular timing, and insufficient measurements.

I built an experiment workstation where the streamer can inspect the motion, review graphs, compare trials, and examine the evidence.

Finally, I created Discovery Theatre, which presents the results through a cinematic interface designed for use in OBS.

I also developed synthetic test fixtures so I could repeatedly test the measurement and analysis workflow under controlled conditions. These are clearly identified as simulated observations rather than real physical measurements.

Challenges I faced

One of the biggest challenges was making several complicated systems work together while keeping the experience simple for the viewer.

Connecting Twitch chat was one part of that. I had to think about how to identify viewers correctly, prevent duplicate predictions from being counted incorrectly, and ensure that votes submitted after a round closes don't affect the results.

Another challenge was scientific reliability.

It's relatively easy to display a graph or calculate a number. It's much harder to make sure that number is meaningful. Video frames can be irregular, tracking can fail, and measurements can be noisy. I had to design the system to recognize these problems instead of treating every observation as reliable evidence.

I also spent considerable time thinking about presentation. Scientific results can be difficult to communicate, especially during a livestream where people expect things to move quickly. I wanted the final reveal to be engaging without sacrificing scientific accuracy.

Working with Codex taught me another important lesson: generating code is only one part of software engineering. You still have to understand what the code is doing, test its assumptions, identify failures, and make sure the different components work together.

What I learned

This project changed how I think about both science and software development.

I learned a lot about Twitch integrations, real-time interactions, video-based measurements, backend validation, and designing interfaces that communicate complex information clearly.

But my biggest lesson was about the difference between an answer and evidence.

A system can produce a convincing result and still be wrong. That's why I wanted Proof of Wonder to preserve the observations, show how conclusions were reached, and admit when the evidence is inconclusive.

I also learned how important it is to build around the people who will actually use a product. The technical work matters, but the real question is whether a streamer and their community will find the experience enjoyable and meaningful.

What's next

Proof of Wonder is still a prototype, and there is more I want to explore.

The next steps include further validating its measurements against physical experiments, improving the camera-tracking experience, and making it easier for streamers to set up and run investigations.

Eventually, I'd love to expand beyond the pendulum experiment and support other kinds of scientific demonstrations.

I can imagine a science teacher using it to make a lesson more interactive, a creator using it to explore everyday questions with their audience, or students joining a livestream from different parts of the world and investigating the same problem together.

I don't know exactly how far Proof of Wonder will go, but I know why I wanted to build it.

Because science becomes more interesting when we don't just watch discoveries happen, but get to be part of them.

Built With

Share this project:

Updates

Submission history