Trading Chaos
A local-first paper-trading research platform with explicit safety gates, replayable experiments, and a backtest to walk-forward promotion loop.
Problem
Strategy experimentation gets risky and untrustworthy when it lives in ad hoc scripts. I wanted a system that made research reproducible and paper-safe by default.
Approach
I built a Python research and operator loop around SQLite, run manifests, risk gates, paper sessions, experiment artifacts, and guarded parameter promotion.
Outcome
Trading Chaos is an active private research platform. Crypto is the primary path, with equities secondary and prediction-market work extracted into a separate project.
Highlights
- Strategies emit signals; a central execution boundary owns portfolio mutation and safety checks.
- Risk controls include kill switch, daily loss cap, cooldowns, drawdown pause, max positions, and stale-data protection.
- Experiments produce inspectable reports, manifests, journals, and registry entries.
Safety Boundary
The system is paper-safe and fail-closed by default. Strategy code does not directly mutate portfolio state; execution flows through centralized checks before anything becomes a simulated trade.
That design makes it harder for research code to accidentally become live-trading-adjacent behavior.
Research Workflow
The loop is intentionally boring: screen, backtest, tune, walk forward, promote guarded parameters, and inspect artifacts. SQLite is the local source of truth, and reports are written so runs can be compared after the fact.
Operator Tooling
The platform includes health checks, session monitoring, local API/dashboard surfaces, JSONL event streams, alerts, and Docker-deployable scheduling. Those pieces make it feel more like a system than a notebook.