PolyBot

Overview

A Python trading bot platform for Polymarket BTC Up/Down markets, with active focus on 5-minute BTC lanes. It is structured as a platform with shared runtime, risk, execution, storage, observability, adapters, bots, and lanes. I deploy it on GCP Compute Engine VM instances for continuous operation.

CLI / UI

The product surface is a Python CLI rather than a web app. It supports run, report, dashboard, settle, and compound commands. The dashboard and reporting layer includes text reports and a live terminal dashboard for lane metrics and operational state.

Core System

A registry-driven runtime resolves bot and lane identifiers, builds a normalized run context, dispatches the lane module, and persists artifacts. BTC lanes pull BTC market data and Polymarket quotes, evaluate policy logic, and only execute when an order intent exists. The policy can return buy_yes, buy_no, or abstain with reason-code diagnostics for stale data, spread quality, expiry timing, market quality, and edge size. Live execution is fail-closed by default with readiness checks, lane allowlisting, kill switch logic, exposure caps, balance checks, and top-of-book drift checks.

Persistence

Uses file-based lane-scoped artifacts instead of a traditional database. Runs write signal logs, trade logs, portfolio state, latest stats, latest scans, run logs, compound state, and certification snapshots. Shared account-risk state is persisted on disk and locked so multiple BTC lanes cannot reserve the same allocation at the same time.

Integrations

Polymarket Gamma discovery, Polymarket CLOB market and execution flows, Binance BTC ticker, candle, and microstructure data, and Chainlink Data Streams as an additional BTC feed. The system also consolidates resolved-trade lessons into compound-learning metrics like drawdown, profit factor, Brier score, CLV, EV realization, and mistake recurrence.

Testing / Reliability

Unit tests cover CLI behavior, runtime dispatch, Polymarket discovery and execution, account-risk locking, dashboard rendering, live-risk checks, and compound learning. Reliability features include fail-closed live execution, shared capital reservation, public-search fallback in discovery, kill-switch support, readiness scorecards, and deterministic artifact generation.

Important

This is an engineering project, not financial advice. No claims of profits, win rate, or guaranteed edge.

Outcome

An executable multi-lane trading platform with paper and guarded-live paths, not a one-off script. It produces lane-scoped analytics and readiness artifacts for promotion decisions.