Skip to content

FAQ

Common technical questions. (These also help seed the in-app Ask the docs assistant — the ? button in the header.)

A self-hosted, single-user platform to research trading strategies honestly — ingest data, write or generate strategies, validate them through a strict anti-overfit gauntlet, and run survivors on paper and (gated) live through Interactive Brokers. See introduction.

No — not for research. You can ingest data, backtest, run the full gauntlet, and use the Studio with just a (free) Tiingo key. IB is only needed when you want to paper or live trade equities. Crypto paper trading uses a fully local simulator (no account needed). See paper trading.

Most settings are read at startup by both the API and the worker, so restart after changing backend/.env (scripts/stop.sh && scripts/start.sh). The Anthropic key is the exception — Studio/ assistant generation picks it up immediately. See configuration.

My backtest looks great — is the strategy good?

Section titled “My backtest looks great — is the strategy good?”

Almost certainly not yet. A positive backtest is the start. Run it through the gauntlet: does it beat random entries (rule test), survive out-of-sample (walk-forward), clear the deflated Sharpe gate, hold up at 2× costs, and beat buy-and-hold? Watch its readiness score on the Research tab.

How does the Studio run AI-generated code safely?

Section titled “How does the Studio run AI-generated code safely?”

Generated Python is validated against a static allow-list, executed with restricted builtins and a guarded import, and backtested in an isolated, resource-limited subprocess. It can’t go live until it clears the gauntlet and you approve it. See Studio and the security model.

Keys live in backend/.env on your machine and are sent only to the vendor you gave them to; data stays in the local catalog. The AI features call your own Anthropic key. Nothing phones home. See the security model.

Write a file under backend/ats/strategies/ following the contract (and the canonical STRATEGY_GUIDE), or describe it in the Studio. Either way it joins the picker and the gauntlet.

Yes — python -m ats.sweep screens the whole basket and runs the deep gauntlet on the survivors, printing a ready-first leaderboard. See command-line research.

Where’s the full API / config / metrics reference?

Section titled “Where’s the full API / config / metrics reference?”

REST API (generated from the live schema), the configuration reference, and the metrics reference.

Yes — they’re portable Markdown. See the rendering note in deployment.