Four AI agents (Bull, Bear, Black Swan, Ostrich) debate across the Johari Window matrix (Known/Unknown × Proactive/Ignored). After 3 rounds with Arbitrator coaching, a final verdict is reached. First application of Johari Window to AI geopolitical intelligence.
Every MAD debate generates short (7d), medium (30d), and long (180d) horizon predictions. After verify dates, actual SPY market movement is compared. Correct predictions increase source trust weights via EMA. First automated geopolitical prediction validation framework.
Three independent LLM runs at different temperatures (0.1, 0.3, 0.7) generate 95% confidence intervals for sentiment scores. Uses t-distribution (t=4.303 for n=3). Adds statistical rigor to qualitative AI analysis. First CI framework for LLM-based geopolitical sentiment.
Full geopolitical intelligence platform running 4 autonomous pipelines (gni_pipeline, gni_mad, gni_heartbeat, gni_adaptive) entirely on free tiers. Groq 100K tokens/day + Supabase 500MB + Vercel + GitHub Actions. Zero cost, unlimited operation. Challenges assumption that AI research requires paid infrastructure.