Prediction Engine

A single launcher that installs and runs three open-source prediction-market tools in parallel — with honest exit codes and full attribution.

What this is NOT: this binary contains no trading logic, no AI, and no backtesting engine of its own. It is a dependency-checking orchestrator for three upstream projects by other authors. Earlier versions of this repository vendored those projects under the same name; that was misleading and has been removed.

The three engines

Engine	What it does	Upstream (all credit to)
polyseer	Multi-agent AI market research: evidence grading, Bayesian probability aggregation, web UI	Polyseer by yorkeccak
papertrader	Polymarket paper trading for agents: MCP server, order-book execution, fee/slippage model	polymarket-paper-trader by agent-next
backtest	NautilusTrader-based strategy backtesting: book replay, Optuna optimisation	prediction-market-backtesting by evan-kolberg

Install

go install github.com/Archsec-Emman/Prediction-Engine@latest
# or build from source:
git clone https://github.com/Archsec-Emman/Prediction-Engine.git
cd Prediction-Engine && go build -o prediction-engine .

Zero dependencies beyond the Go standard library.

Usage

# see engines, dependency status (git/node/python/pip), install state, credits
prediction-engine status

# clone upstream repos into ~/.prediction-engine and run their installers
prediction-engine install all
prediction-engine install polyseer

# run one or all engines (output streamed live per engine, summary at end)
prediction-engine run papertrader
prediction-engine run polyseer            # starts the web UI dev server
prediction-engine run all -- --help       # args after -- are passed through

Exit code reflects reality: if any requested engine fails, run exits non-zero and the summary shows which one.

Design notes

• Nothing is vendored. Engines are cloned from their canonical repositories at install time, so you always run upstream's latest code with upstream's own license.
• Dependency checks before action. status reports exactly which tools (git, node, npm, python, pip) are missing before any install or run attempt.
• Windows-safe process handling. Batch-file shims (npm.cmd etc.) are invoked through raw command lines so paths containing spaces work.
• Tested. Unit tests cover dependency resolution, output streaming, exit-code propagation and the registry contract; CI runs them on Linux, Windows and macOS.

License

MIT — see LICENSE. The engines it launches remain under their own licenses (see each upstream repository).