All facts. No manufacturing.
We poll multiple AI models about the same event over time and detect when narratives shift without new evidence. Track the story. Catch the drift.
How It Works
1
Track an Event
Add any news event. We query Claude, GPT-4o, Grok 3, and Gemini with the same prompt, grounded by real-time search results.
2
Detect Drift
Each poll cycle compares responses to previous ones using embedding similarity, measuring how the narrative changes over time.
3
Expose Anomalies
Ghost Pivots flag narratives that shift without new sources. Memory Holes catch facts silently dropped. Context Rejections detect when models ignore evidence they were given.
Features
Multi-Model Polling
4 AI models queried in parallel on every cycle for independent perspectives on the same event.
Search Grounding
Real-time Tavily search results fed to models so they can't hide behind training cutoffs.
Narrative Drift Detection
Embedding-based cosine similarity tracks how stories change between polls.
Context Rejection
Detects when models ignore facts they were explicitly provided, revealing selective reasoning.
Tracked Events
Loading events...