Roguemouse: Algo Trading Governance

We realized that a lot of developers, researchers, or stakeholders in FinTech spend their time creating trading platforms that make hundreds of decisions per minute. When something goes wrong, the forensic record either doesn't exist, or isn't trustworthy. Roguemouse is an AI Operations Officer that sits *over* an algo platform, not in it.

When an anomaly is detected (unusual IV/RV ratio, position drift, model disagreement), Roguemouse runs a multi-agent debate. A Risk Officer voice running on Gemini examines the anomaly from a risk-management perspective. An Ops Engineer voice running on Vultr's Nemotron-3-Nano-Omni examines it from a systems-and-data perspective. A Synthesizer reconciles the two into a single governance recommendation: investigate, refuse to propose action, or escalate to a human operator.

Every step of the debate is written to a SHA-256 hash-chained audit log stored on Vultr Object Storage. The chain is anchored to a public genesis hash (sha256 of "roguemouse-audit-genesis-v1"), which means every governance decision is externally verifiable. Judges and auditors can click "Verify chain" on any run and recompute the hash chain client-side, so the audit does not depend on trusting our servers.

The pattern Roguemouse demonstrates is different from what competing projects are building. It is not a smarter trader with internal guardrails. It is an external governance layer that any existing trading system can integrate with. The hackathon demo runs on canonical fixture data. Production deployment would integrate with the firm's existing anomaly-detection pipeline (PagerDuty, Slack webhooks, or platform-specific alerts).

Note: the demo URL is substituted with the GitHub repository link.