I'm building a belief database for AI agents. Here's a prototype — do you have real datasets to test with?

Hey everyone, I'm working on **Verus** — an open-source belief database designed for AI agents. The core idea: A regular DB stores facts: `user.city = 'Lisbon'` Verus stores **claims** — multiple assertions about the same thing from different sources, each with confidence scores, validity periods, and conflict detection. It helps the agent decide **what to believe**. Example: 4 sources say 4 different things about a user's location (CRM says NYC, user says Lisbon, IP geolocation says London, calendar says Tokyo). Verus tracks all four with their confidence and validity, detects contradictions, and exposes this to the agent via policy-driven resolution. The prototype shows: * Conflict graph — nodes are claims, edges are contradictions between them * Confidence scoring with visual decay over time * Source filtering (CRM, user input, API, documents, agent inference) * Timeline of claims with validity windows **The stack:** Rust core (binary storage, single file, zero dependencies) + MCP server (works with Claude Code, Cursor, VS Code, Windsurf) + Web UI for visualization. **What I'm looking for:** I've built this with synthetic data (24 claims, 9 conflicts across 5 sources). Before I go deeper, I want to validate with real-world scenarios: 1. **Do you have datasets** with conflicting information from multiple sources that I could use for testing? 2. **What's your current approach** to handling conflicting data in your agents? Hardcoded rules? LLM prompting? Something else? 3. **What types of conflicts** do you see most often — contradictions, stale data, source reliability issues? Any feedback on the concept or the prototype is also very welcome. Thanks!