Seldon Vault — AI Geopolitical Forecast Engine
Seldon Vault is a free, public AI-powered geopolitical forecasting engine. It uses a multi-agent LLM architecture with 7 specialized analysts, an adversarial Skeptic with fact-checking, and a Seldon Arbiter to generate daily probabilistic forecasts.
How it works
The system collects news signals from RSS feeds, GDELT, ACLED, Reddit, Telegram, and other sources. Signals are processed through 7 AI analysts in parallel (geopolitician, economist, technologist, sociologist, climatologist, military analyst, cybersecurity analyst), subjected to adversarial Skeptic review with Tavily fact-checking, and synthesized by the Seldon Arbiter. Probabilities are updated via Bayesian inference every 6 hours. Accuracy is tracked using Brier Score.
Key features
Daily AI-generated geopolitical, economic, and technological forecasts
Multi-agent ensemble analysis (7 analysts + skeptic + arbiter)
Bayesian probability updates every 6 hours
Brier score accuracy tracking
Cascade narrative detection (causal chains between forecasts)
The Seldon Plan — monthly structural forecasts (1-10 year horizons)
Interactive world map with regional risk analysis
Bilingual support (English and Russian)
Public read-only REST API
Pages
API
Public REST API at /developers . Endpoints: GET /api/v1/forecasts, GET /api/v1/narratives, GET /api/v1/metrics, GET /api/v1/regions, GET /api/v1/events/stream (SSE). No authentication required.
More details: /llms.txt | /feed.xml (Atom feed) | /sitemap.xml
← Cascade Narratives > Zaporizhzhia Safety Crisis Forces IAEA Emergency Session Amid Reciprocal Strike Cycle ↑ Escalating active Military & Defense Geopolitics Environment europe global
The intensifying Russian-Ukrainian reciprocal strike campaign raises nuclear safety risks at Zaporizhzhia NPP, increasing pressure on the IAEA to convene an emergency session within two weeks.
// Cascade Logic Russian strike campaign + Ukrainian deep strikes → grid/cooling incidents around ZNPP → diplomatic and safety alarm → IAEA emergency session.