Signal-Driven#132

Multi-Agent Debate Convergence Signal

For each market, stage a structured AI debate: one agent argues Yes, one argues No, and a judge agent updates its estimate over several rounds. The signal is how fast and how firmly the judge settles toward one side. Markets where the debate converges strongly against the current price get bigger bets; wishy-washy debates are skipped. The edge is that disciplined back-and-forth reasoning exposes prices that are falsely confident.

What you need to run it

Orchestration for multi-round LLM debate per market
Judge-agent calibration against resolved outcomes
Confidence-to-size mapping + risk overlay
Compute budget for repeated debate runs

Where this applies

Markets on Polymarket where multi-agent debate convergence signal is the natural play:

Will China launch a military invasion of Taiwan before the end of 2027?
Will the favorite win Best Picture at the 2027 Academy Awards?
Will the US enter a recession before the end of 2027?

Capabilities this demands

Model / quantCustom code / APIPatienceRisk management

At a glance

CategorySignal-Driven

Requirements4

CapabilitiesModel / quant, Custom code / API, Patience, Risk management

VenuePolymarket (CLOB, Polygon)

Build it

Related signal-driven strategies

This is documentation, not advice. Poly Research & Robotics publishes how these strategies work because the method should be checkable — not as a recommendation to trade them. See the full strategy database (147 strategies) or the data resources directory.