# Team Statistics

- **Event:** [Built with Opus 4.6: a Claude Code hackathon](https://cerebralvalley.ai/e/claude-code-hackathon)
- **When:** Feb 10 at 12:00 PM – Feb 17 at 10:00 AM (EST)
- **Where:** Location TBA
- **Team:** [mian abdullah](https://cerebralvalley.ai/u/mian)
- **GitHub:** https://github.com/mian-abd/prediction-market-analysis
- **Demo video:** https://youtu.be/W_PEd88-Y10
- **Gallery:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery/59

I built PredictFlow because I kept staring at Polymarket thinking that price looks wrong, but I had no way to prove it or decide how much to risk.
PredictFlow pulls live data from Polymarket and Kalshi, runs markets through a three model ensemble, calculates the edge after fees, and automatically paper trades with guardrails. Users stay in control without babysitting.
I trained models on 4,341 resolved markets using a strict temporal split. While building I found leakage in volume features with a 0.85 correlation to outcomes. I learned and removed them, and documented it. The cleaned system still beats baseline by 21.5%, especially in the 20 to 80 percent range where trades happen.
Claude Opus 4.6 returns structured probability, drivers, and risks. Requests are SHA-256 cached, so repeats cost zero. Under the hood are PostgreSQL, many APIs, and a loop that collects, predicts, trades, and scores itself.
I started with a signal page. I went above and beyond and built a small quant desk.

---

Markdown version of https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery/59. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt