What Backtest.ai is, and who it's for

What Backtest.ai does

The platform takes a strategy idea from a hypothesis to a backtest on minute-resolution options data, with point-in-time discipline and every load-bearing default in view.

Three surfaces carry the work:

• Signals. You write derived indicators in the Signal language, a Signal expression language that reads market data and returns a {timestamp, value} timeseries. Realized vol, an IV rank, an IV-minus-RV spread: each is a signal you can chart and analyze.
• Backtests. You run multi-leg options positions over history, with contract selection, entry and exit logic, adjustments, sizing, and margin all modeled explicitly.
• The AI assistant. You ask the AI research assistant questions grounded in the Signal language reference, with citations back to the source.

The data behind the accessors is minute NBBO. Symbol universe and per-symbol history live on one page, Data coverage, so the rest of the docs link there instead of repeating the list.

Every number the platform reports is reproducible, because the defaults that produced it sit next to it rather than three screens away.

Who it's for

The reader we write for is a self-directed, options-fluent trader, running the strategies this platform is built around: 0DTE, premium selling, volatility trades. You trade your own capital, and you decide whether to trust a number by looking at what produced it. You have been burned by curve-fit backtests and by tools that hide fills and look-ahead, so a result with no visible assumptions reads to you as a result with something to hide.

These docs assume you know options. Greeks, verticals, condors, calendars, IV rank versus IV percentile, term structure, skew: none of it gets re-explained here. They assume nothing about the Signal language, which is taught from the ground up for a reader who writes Excel formulas and a little Python.

We write trader-to-trader, not vendor-to-user.

What makes it different

The difference is a stance, and it shows up as mechanisms rather than adjectives. You can see what produced every number, because the platform refuses to hide the thing that shaped it.

Concretely:

• No result appears without its defaults. A displayed value states the assumptions behind it, or links to them.
• Rolling windows count observations, not calendar time. rolling_mean(x, 20) averages the last 20 observations of x. On a 1-minute series that is 20 minutes, not 20 days. To express 20 days, you sample to daily first.
• A daily value is invisible intraday. A daily bucket becomes readable only at its session-close label, so today's daily value is not available at 10:00am. That is what keeps mixed-frequency expressions look-ahead-safe.
• Fills price with the Backtest.ai fill model, at mid, or at bid_ask, and you choose. The fill model is the calibrated mode that estimates where between the midpoint and the touch an order actually fills, and paper and live trading always price with it. mid fills at the exact quoted midpoint, an optimistic bound; bid_ask pays the full spread at the touch, a pessimistic bound. Use the fill model to match paper and live, mid for the optimistic bound, bid_ask for the pessimistic one.
• The equity curve marks at mid regardless of fill mode. Position marks and margin always use mid, so your marked equity does not flatter itself when you fill at the spread.
• Forced liquidations cross the spread. A risk-driven liquidation pays the spread, the way a real one does.

For more on how fills and marks are modeled, see fills, marks and slippage and backtest assumptions and realism.

What it is not

Backtest.ai is analysis and automation software. It is not financial advice, it makes no performance guarantees, and it projects no returns. A backtest tells you what a set of rules would have done against historical data under stated assumptions, nothing more. What you do with that is yours.

Where to go next

• Core concepts and vocabulary defines the terms the rest of the docs lean on: Signal, accessor, spine, observation-count, position, book, leg.
• Quickstart: your first backtest walks a strategy from idea to a running result.