PolyMonitor: Open-Source Intelligence Infrastructure for Prediction Markets
We introduce PolyMonitor, an open-source live intelligence workspace for Polymarket — aggregating market data, on-chain flow, oracle activity, and AI-generated forecasts into a unified dashboard, paired with a 10-node multi-agent forecasting pipeline.
We introduce PolyMonitor (polymonitor.club), an open-source intelligence workspace for Polymarket participants — the companion platform to our paper Unlocking the Forecasting Economy: A Suite of Datasets for the Full Lifecycle of Prediction Market.
What PolyMonitor Does
Prediction market research today is fragmented: prices, on-chain trade flow, oracle activity, order-book depth, and macro context live in separate tools. PolyMonitor unifies them into a single live workspace, covering the full lifecycle of a prediction market from listing through settlement and archive.
The platform is organized around five core modules:
| Module | What it tracks |
|---|---|
| Market & Lifecycle | Active markets, outcome probabilities, price movement, trade history |
| Flow & Liquidity | OrderFilled tape, whale tracking, order-book depth, bid/ask snapshots |
| Oracle Watch | UMA oracle events, resolution timelines, settlement risk signals |
| Intelligence & News | AI-generated market briefs, macro panels (CPI, Fed), sports odds, weather markets |
| Distribution | Telegram publishing layer for topic-based channel updates |
Unlike wrappers around public Polymarket APIs, PolyMonitor maintains independent infrastructure — a custom market catalog, token catalog, chain indexing system, runtime snapshots, and panel registry — enabling comprehensive inspection across the full market lifecycle.
Polymarket Agent
The intelligence layer is powered by the Polymarket Agent, a structured multi-agent forecasting pipeline built on the Forecast Intelligence Graph. Rather than free-form conversation, the agent produces auditable, market-level claims grounded in prices, fills, related markets, oracle state, and external catalysts.
The pipeline executes as a directed state machine with 10 sequential nodes:
The nodes divide into deterministic stages and LLM specialist stages:
Deterministic stages (rule-based, fast):
- Evidence Builder — constructs structured evidence from market prices, fills, and external context
- Related Markets — links sibling markets and cross-market correlations
- Quant Forecaster — computes price drift and microstructure indicators
- Reflexion Memory — loads prior forecast episodes to detect drift and anchoring
LLM specialist stages (model-backed):
- Microstructure Agent — analyzes trading mechanics, liquidity depth, and order flow patterns
- Catalyst Agent — identifies external event triggers and timing signals
- Resolution Agent — examines settlement criteria and oracle signals
Synthesis stages:
- Calibration Agent — aggregates confidence scores across all data sources
- Skeptic Agent — adversarially critiques weak evidence and overconfidence
- Panel Writer — generates the final dashboard output
Each run returns a dashboard payload, a quant snapshot, and a replayable event trace with per-node outputs, latency metrics, and token usage — making every forecast fully auditable.
Connection to the Forecasting Economy Dataset
PolyMonitor is the operational platform behind our Unlocking the Forecasting Economy dataset suite, which provides structured data covering every stage of the prediction market lifecycle: market creation, trading dynamics, oracle resolution, and settlement. The dataset spans thousands of Polymarket markets with linked on-chain and off-chain signals, enabling end-to-end forecasting research across the full arc from listing to close.
Links
- Website: polymonitor.club
- Agent Docs: polymonitor.club/docs/polymarket-agent
- GitHub: virusLuke3/polymonitor
- Paper: arXiv:2604.20421
