Introduction to Geomacro
Onchain geopolitical risk intelligence, settled in USDC, on Arc. Autonomous market-generation pipeline with no human in the loop.
What is Geomacro and why did we build it?
Most prediction market platforms wait for a human to notice the news, write a proposal and get it approved before a market exists. By the time a human-curated platform lists a market, a meaningful part of the uncertainty has usually already resolved.
Geomacro removes that gatekeeper. A scheduled pipeline reads live news, an LLM scores each story for severity and relevance, two AI agents (Hawk and Dove) argue opposite outcomes to give stakers context, and a market opens automatically on Arc Testnet. Everything settles onchain in USDC. No custodian, no order book, no admin approval step.
How a market actually gets created
Seven scheduled GitHub Actions jobs cover the full lifecycle, each doing one job: ingest news, scan for high-severity events and open a market, post the AI verdict at the resolution point, close the dispute window, keep every market's lifecycle stage in sync (self-looping every 15 minutes), reconcile onchain stake events (every 30 minutes), and a two-tier WARN/CRITICAL anomaly monitor watching the rest. No human approval step in any of them.
Before you stake, the AI Duel view shows the strongest argument for escalation (Hawk) and the strongest argument for de-escalation (Dove) on that specific market, generated from the same story rather than a generic template.
After the resolution window, an LLM re-reads the original story and the follow-up coverage to judge which side aged better. Winners split the losing side's pool proportionally to their stake. The newer AgentArenaV2 contract (deployed to testnet, not yet the default in production) adds a 5-member AI jury that can overturn a contested AI verdict on a 4-of-5 supermajority.