Open source
Polymarket Trading Bot
A modular prediction market signal pipeline for Polymarket US sports markets. Phases 0–2 active: ingestion, normalization, and a model engine that scores edge, liquidity, timing, and position sizing. Execution gated by design.
76
Unit tests passing across contract, ingestion, model, and risk
0
External dependencies in core model engine (pure stdlib)
Gated
Phase 3 execution disabled by default — validate first
The challenge
What needed to change
Prediction market trading requires combining sportsbook odds, CLOB depth, timing signals, and risk state into a single coherent model — and doing it in a way where the signal logic is auditable and the execution step is safely gated until validated.
The approach
How I built the system
Four-phase Docker pipeline: risk bridge → ingestion → model/decision engine → execution (disabled). The core model engine is pure Python stdlib — no external ML dependencies. Phase 2 emits decision packets with execution_eligible: false. All secrets and account data are gitignored; the engine operates on normalized data only.
The outcome
What exists now
- → Full signal pipeline from market discovery to sized order intents.
- → Auditable model: every signal attribution logged per decision packet.
- → Paper-trade-first design: inspect output for days before considering live use.