Core concepts and vocabulary
The terms the rest of these docs lean on, defined once and grouped by the pipeline stage where each one first shows up.
Every result on the platform inherits a fixed set of defaults. Rolling windows are observation-count on the current spine, never calendar days, and mixed-frequency values follow look-ahead-safe visibility: a daily value appears at its session-close label, not intraday. For the quote basis behind IV and the price adjustments applied to bar.*, see implied volatility and stock price adjustments.
Signal language terms
These belong to the Signal language, the expression language you write signals in. Definitions here are one line each; for depth see the Reference glossary.
Signal. A {timestamp, value} timeseries carrying an asset_classes tag; the value column is Float64 and nullable.
scalar. A plain number, never wrapped in a Signal. A program that ends on a scalar raises.
accessor. A named market-data object you read values from: bar, option, time (see data accessors).
spine. The valid timestamps for a Signal. The accessor spine runs 09:31 through the session close on NYSE trading days: 390 observations on a normal day, 210 on a 13:00 half-day.
union spine. The merged set of timestamps two Signals align on when you combine them. The result carries every timestamp present in either operand.
backward as-of fill. How each operand lands on the union spine: at every union timestamp, each side carries its most recent value forward. Timestamps before a Signal's first observation are null. See the execution model.
observation-count window. Rolling and lag windows count observations on the current spine, not calendar days. rolling_mean(x, 30) is 30 observations, which on a 1-minute series is 30 minutes. Sample to daily first for "30 days."
no-lookahead. Windows include the current observation and past observations only, never future ones.
null vs NaN. null is missing data (no row, or a value before the first observation); NaN is an invalid computation such as log(0). NaN takes precedence over null.
sampling vs bucketing. sampling selects a point-in-time row; bucketing aggregates a window into one value. They are mutually exclusive on a single accessor call.
dte, log-moneyness, and leg slugs
dte. Calendar days to expiration; the first positional argument of every option.* method.
log-moneyness. ln(K/forward), log-moneyness centered on the implied forward, where 0 is at-the-money-forward. It is the moneyness column and the moneyness argument.
leg / strategy slug. A string you choose to name one option-contract role in a position, e.g. short_put. The slug tracks that role across rolls, so a windowed trigger on leg("short_put") continues after the contract is replaced. Canonical slugs: short_put, long_put, short_call, long_call.
Backtest terms
These belong to backtests, the layer where a strategy runs over history.
backtest. Running a strategy over historical data.
position. The structure under evaluation, read with position().
book. Whole-account state, read with book(): cash, net liquidation value, margin, and P&L.
leg. One contract role in a position, addressed by its slug: leg("short_put").
lots / size. size is the signed per-lot contract quantity (negative is short). lots is the position's whole-lot count.
fill mode. mid, bid_ask, or the Backtest.ai fill model, the calibrated mode that paper and live trading price with. A required choice with no default.
results / summary. The three result frames (an equity curve, a margin curve, and a trade ledger) plus the end-of-run summary stats (see the Metrics glossary).
close reason. Why a position closed, one of: exit_signal, expiration, soft_trim, forced_liquidation, equity_halt, adjustment_full_close.
Pipeline-stage terms
The pipeline runs Signal → Model → Backtest. Every workflow ends in a backtest; the earlier stages are optional.
Signal. The stage where you write and inspect a signal. EDA lives inside the Signal tab, not as a separate stage.
Model. A fitted mapping from signal features to a target, trained with walk-forward validation.
Return Dataset. A named historical record of one trade structure's outcomes, used as the EDA benchmark.
Benchmark Context. What a signal is read against, plus how the comparison observations are built.
Backtest. The terminal stage, where the position runs over history and produces results.
Beyond a single backtest
These extend one backtest into an ongoing strategy. See Portfolios, Accounts & Live Trading.
Portfolio. Combine multiple backtests under percentage allocations into one set of portfolio-level results.
Account. A wrapper over a Portfolio that runs it on live data, paper or live. The live-trading feature is an automation.