# Deception Check

- **Event:** [Meta Global AI Developer Hackathon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon)
- **When:** Oct 26, 12:00 PM – Nov 4, 9:00 PM (EST)
- **Where:** Online
- **Team:** [Nicholas Carroll](https://cerebralvalley.ai/u/DeceptionCheck)
- **GitHub:** https://github.com/DeceptionCheck/pump-station-zero-community
- **Demo video:** https://youtu.be/c8zRYXETCEo
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4445

Pump Station Zero: One Click Stops the Pump

Pump Station Zero turns an abstract cybersecurity incident into an experience you can investigate. Visitors enter an isolated simulated water-pump station, perform actions through a working industrial interface, and see the resulting Modbus activity become evidence. They can then ask questions, review an AI-assisted interpretation, and retrieve a report connecting their actions to the recorded events.

Our goal is to help junior analysts move from “something happened” to “here is what the evidence supports.” The experience combines an interactive operational scenario, evidence timelines, illustrated research, model-assisted analysis, and an editable voice-input workflow. AI helps interpret the evidence; it does not control the station or silently execute instructions.

Meta products contributed both to building the project and to its capabilities. Meta Spark 1.3 produced application versions represented in this submission, including the hosted Muse v1 and Muse v2 experiences and one DecoyBench version. We tested Llama locally on an NVIDIA RTX 5060 Ti, then used an RTX 4060 Ti for the local demo system. The public judging demo runs Llama on a Runpod RTX 4090, alongside Meta Spark and Meta Voice integrations. We also tested Glimmer on the RTX 5060 Ti for DecoyBench; Glimmer is not enabled in the public demo.

The submission includes the main Pump Station Zero experience, an experimental Muse v2 visitor journey, a preserved read-only Muse v1 baseline, and DecoyBench research and benchmark repositories. The report documents verified results alongside missing measurements and controlled-restart limitations.

This is a research and training environment using simulated infrastructure—not a connection to a real water system. Its central idea is simple: make security evidence tangible, then make investigating it accessible.

Links for judges
- Project website: https://meta.deceptioncheck.ai/
- Interactive Pump Station Zero demo: https://meta.deceptioncheck.ai/demo/
- Muse v2 experimental visitor journey: https://meta.deceptioncheck.ai/visit
- Muse v1 read-only historical baseline: https://meta.deceptioncheck.ai/muse-v1/
- Illustrated research report: https://meta.deceptioncheck.ai/research/DecoyBench-Report.html
- Downloadable report PDF: https://meta.deceptioncheck.ai/research/DecoyBench-Report.pdf
- YouTube demonstration: https://youtu.be/c8zRYXETCEo
- Community source release: https://github.com/DeceptionCheck/pump-station-zero-community
- DecoyBench: https://github.com/DeceptionCheck/decoybench
- DecoyBench alternative A: https://github.com/DeceptionCheck/decoybench-alt-a
- DecoyBench alternative B: https://github.com/DeceptionCheck/decoybench-alt-b

## More from Meta Global AI Developer Hackathon

- [DnA-Arkitecs](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4442)
- [NovelSpell Remix](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4443)
- [Super Ninja](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4444)
- [Anville](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4446)
- [Doc_Coder](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4447)
- [dotbion](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4448)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4445. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt