# Lacuna

- **Event:** [Built with Opus 4.7: a Claude Code hackathon](https://cerebralvalley.ai/e/built-with-4-7-hackathon)
- **When:** Apr 21 at 12:00 PM – Apr 27 at 2:00 AM (EDT)
- **Where:** Online
- **Team:** [JangKeun Kim](https://cerebralvalley.ai/u/jangkeunkim)
- **GitHub:** https://github.com/jang1563/lacuna-falsification
- **Demo video:** https://youtu.be/eB-gREA4zGI
- **Gallery:** https://cerebralvalley.ai/e/built-with-4-7-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/built-with-4-7-hackathon/hackathon/gallery/240

Lacuna is an AI-for-science system built to say no.

Most discovery tools are optimized to generate more hypotheses. Lacuna focuses on the harder scientific step: rejecting weak ones before they become convincing stories.

Built with Claude Code, Opus 4.7, and Claude Managed Agents, Lacuna proposes candidate biological laws, then sends them through a fixed five-test Python gate. The AI cannot change the rules after seeing the result. Failed candidates are saved as useful scientific evidence, not treated as wasted work.

On a public kidney cancer dataset with 505 patients, Lacuna rejected 194 of 203 candidate laws. One simple survivor, TOP2A − EPAS1, matched a known kidney-cancer growth program. That was the point: this is not presented as a new biological discovery, but as a methodology proof that the system can rediscover known truth under strict rules.

That survivor also passed an independent survival test in a kidney cancer clinical-trial dataset. A later three-gene extension proposed by the system failed the same gate.

The broader goal is a repeatable scientific discipline: propose, test, reject, remember, and only then interpret.

---

Markdown version of https://cerebralvalley.ai/e/built-with-4-7-hackathon/hackathon/gallery/240. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt