Glossary

Higher is better; zero means the benchmark explains the whole return. Orbital takes the mean daily strategy excess return, subtracts beta times the mean daily benchmark excess return, and multiplies by 252. Excess is above a risk-free rate of 4.25% a year, and the benchmark's returns are total return.

Positions held open at the end are marked at each close and included.

Example: beta 0.5, a strategy daily excess return of 0.06% and a benchmark one of 0.04%. Alpha is (0.06 minus 0.5 times 0.04) times 252, about 10% a year.

Results are hypothetical. A run is queued, then running, and ends completed, failed or cancelled.

Decisions are taken at a bar's close and fill at the next bar, so a strategy never sees the bar it trades on. Positions still open at the end are held and marked at the last close.

A workspace may have four runs in flight and 200 in a rolling 24 hours. Each run records the version and settings it tested, so one configuration run over two periods is still one configuration.

Alpha and beta are measured against it, and the equity chart draws it dashed beside the strategy. Compare rebases every strategy and the benchmark to 100 on the first day they all have data.

The benchmark is part of a run's configuration, so the same strategy against two benchmarks is two configurations.

Neither direction is better; it describes exposure. Orbital divides the covariance of daily strategy returns with the benchmark's total-return daily returns by the benchmark's variance, over the days both have. Positions held open at the end are marked at each close and included.

Example: a strategy that moves about 0.5% in the same direction on days the benchmark moves 1% has a beta near 0.5.

Higher is better; compare it with the benchmark's over the same period. Orbital computes it as final equity divided by starting capital, raised to the power of 365.25 divided by calendar days, minus 1, times 100. Positions held open at the end count at the last close.

Example: $100,000 grows to $121,000 over two years. CAGR is 10% a year.

A short run compounds a small gain into a large yearly figure, so the figure means little without the run's period.

An entry larger than the budget left, or than Max per trade, is refused, not trimmed. Trade's Capital allocated tile sums every running deployment's budget, and each deployment's P&L comes from its own fills since it started.

A custom rate is capped at 1,000 bps or $1.00 per share. A larger figure is almost always a percent typed into a basis-point field, or a per-order fee typed per share. Regulatory pass-throughs on sales are not modelled.

Net P&L is after commission, and the report shows commission paid, split long and short.

Interactive Brokers Pro Tiered cannot be simulated yet: its schedule has a per-order minimum and a cap, and the engine's cost model is a single rate. A run under it is refused with that reason.

Frictionless charges nothing on purpose; a result under it is an upper bound on what any real broker would leave you.

The report states the profile and, in one sentence, what was modelled and what was not.

Where backtest pricing is on, a backtest's estimate states its credits before it runs, and the run's row records what was charged. Billing and Usage under Settings show what was spent and by what.

Higher is better; a negative figure means the search alone explains the result. Orbital follows Bailey and López de Prado. The hurdle is the expected maximum Sharpe of N trials, sized by how much the strategy's own trial Sharpes scatter. One configuration is not a search, so the figure needs at least two.

When the full daily series is on record, the probability that the true Sharpe clears the hurdle is shown too.

Example: a raw Sharpe of 1.5 after 10 configurations whose Sharpes scatter by 0.5. The hurdle is about 0.79, so the deflated Sharpe is about 0.7.

A paper deployment runs within four limits (budget, max per trade, open positions, daily-loss pause) under a trading permission of Deny, Ask, Auto or Always-approve. It is running, paused, needs attention, stopping or stopped; stopping is final, sells nothing, and leaves what it holds to its own exits.

Signal is an exit rule firing or a Program's exit call. Stop, Target, Trailing and Time are the Exit Rules node's stop loss, profit target, trailing stop and time stop. Margin call is a forced liquidation after a maintenance breach. Delisted closes a symbol at its last available bar. Account blown is a wipe-out that ends trading.

Left universe is a screened universe's name leaving the screen; the position closes at the next session's open. Unpriced is a name with no print for ten sessions in a row, closed at its last print. Left the tape is a name that moved from the exchange to OTC, closed at its last listed print.

Every run holds its open positions at the end, so those rows read Open rather than End of run.

Higher is better; a negative figure means the average trade lost money. Positions still open at the end are not closed trades and are left out.

Example: 100 closed trades with a combined net P&L of $20,000. Expectancy is $200 a trade.

A run can instead fill at the next session's close, or rest a limit order. Fills are whole: an order the account cannot cover is rejected, never trimmed. A stop that gaps through fills at the worse of its trigger and the bar's open; a limit fills at the better. A trailing stop updates at each close and can fire from the next bar.

A limit order rests for the session or a number of bars. One that never fills expires and is counted, so no fills reads differently from no signals.

Start one from a completed run. The forward part begins at the first session that opens after you click, and a strategy has one forward test at a time.

After each session settles, Orbital re-submits the frozen settings with only the end date moved. It is one continuous run, so later edits never touch the backtest part. Only a correction to the price history, such as a restated split, can shift it.

Each update is billed and counted like a run of its window. The test stops from the ⋯ menu. It also stops itself when the window outgrows one run, or when an update is refused in a way no later session can lift.

Nothing is sent to a broker. Paper trading is a Trade deployment.

Smaller is better. Orbital measures it per trade at the execution feed's resolution. For a long it is the lowest low since entry against the entry price; for a short, the highest high. The trade list draws it as the against side of the Against · for bar.

A position still open at the end has an MAE through its last bar.

Example: a long entered at $100 trades as low as $94 before exiting at $103. MAE is 6%.

Larger is better, and a large MFE on a trade that closed for little points at an exit that gives gains back. Orbital measures it per trade at the execution feed's resolution. For a long it is the highest high since entry against the entry price; for a short, the lowest low. The trade list draws it as the for side of the Against · for bar.

Example: a long entered at $100 reaches $110 before exiting at $103. MFE is 10%.

The first version you save becomes main. For a Rules or Program strategy, main starts at the first version whose backtest completes. After that it moves only by Make main, from the editor, Backtesting or Trade, or by an agent granted the Make main permission.

A Rules or Program version needs a completed backtest before it can be made main or deployed. A running deployment is not switched in place: stop it, then deploy the other version.

Shorts pay a 3% a year borrow fee, and a maintenance breach liquidates at the next open. Buying power is equity divided by the initial requirement, less what is already deployed; an order beyond it is rejected and counted. Short strategies need a margin account.

The borrow fee accrues on calendar days, so a short held over a weekend pays three days. If equity reaches zero the run liquidates everything, stops trading and is marked account blown, and ratio metrics are suppressed rather than shown.

Shallower is better; it is shown as a negative percentage with its peak and trough dates. Orbital marks equity at every bar close of the execution feed and takes the lowest value of equity divided by its running peak, minus 1. On a daily feed it is not an intrabar figure.

Positions held open at the end are marked at each close, so a loss on them deepens the drawdown.

Example: equity peaks at $120,000 and later falls to $102,000 before recovering. Max drawdown is -15%.

Positive is a profit. Orbital computes it as final equity minus starting capital. Positions still open at the end are held and marked at the last close, so their open P&L is included. The report shows that part as Of which open.

Example: $100,000 of starting capital ends at $108,500. Net P&L is +$8,500, of which $1,200 may still be open.

Every run holds its open positions at the end rather than closing them, which is what lets a forward test's longer run repeat the shorter one. The report shows the number of open trades and this figure under Net profit as Of which open.

Closed-trade statistics, such as win rate, profit factor and expectancy, leave open positions out.

Example: net P&L is +$8,500 and two positions are still open at +$1,200 together. Open P&L is +$1,200.

Live accounts are never connected, and a deployment to a live account is refused. A Rules or Program version needs a completed backtest before it can be deployed, and it trades on paper under Always-approve, within the limits you set.

A Program can instead pick a size inside a range. Risk percent needs a stop loss; a trailing stop does not count. Orders the account cannot cover are rejected whole, never quietly shrunk.

A run can also cap one position and gross exposure as percentages of equity. Quantities are whole shares.

Higher is better; below 1 the closed trades lost money overall, and by convention 1.5 is solid and 2 strong. Orbital sums every closed trade's positive gross P&L and divides by the sum of the negative ones, as a positive figure. Gross P&L is the price move alone, before slippage, commission and borrow. A trade can win on that gross leg and lose after fees, so the profit factor and the win rate can disagree. With no gross loss to divide by, the figure is not shown. Positions still open at the end are not closed trades and are left out.

Example: winners total $30,000 and losers total $15,000. The profit factor is 2.0.

A Program starts as no_std Rust; the authoring guide says to delete the no_std and trap_on_panic lines when the program needs std's float methods. It links only the orbital_strategy crate besides Rust's own core and std, against host interface version 4. Source is at most 256 KiB.

The Program node names three things. The decision bar is daily by default, or an intraday bar. The history the program reads is up to 20,000 bars of that bar. The instruction budget per bar is 10,000,000 unless the node sets another whole number.

Build compiles the source and runs it over three synthetic tapes as a check. The compiled module is named by the SHA-256 of its bytes, and a run refuses a Program whose source changed without a rebuild.

Apps such as Backtesting are turned on per project, and a connected assistant is bound to one project until you connect again. A project is archived, never deleted, and the default project cannot be archived.

Orbital computes it as net P&L divided by starting capital, times 100. It includes positions held open at the end, marked at the last close.

Example: starting capital of $100,000 and a net P&L of $12,000 give a return of +12%.

The benchmark's return over the same days sits beside it on the strategy page. CAGR states the same gain as a yearly rate.

Together they describe a strategy without code. An Indicator (a moving average, RSI, ATR, volume or return) feeds a rule. A rule fires when its first value crosses or compares against a fixed value, a price or another indicator. Exit Rules add a signal exit, a stop loss, a profit target, a trailing stop or a time stop. Position Sizing sets how big each position is.

Indicators read daily bars unless you pick an intraday bar, which moves the run onto the minute tape. The Regime Filter always reads daily closes and gates every entry.

A daily strategy decides once per session, at the close. An intraday strategy decides on each completed bar inside the session, and a 1-minute Program is called about 391 times. Extended hours are not simulated, and the Session Filter node pins trading to regular hours.

Backtests read settled sessions, and a forward test adds a session after it settles. Friday settles on Saturday, so Monday morning still reads through Friday. A Tuesday update after 11:30 ET covers Monday.

A forward test's live view shows today provisionally until the settled curve takes over.

Higher is better; by convention 1 is good and 2 very good. Orbital computes it on daily close-to-close equity returns: the mean daily excess return divided by its standard deviation, times √252. The risk-free rate is 4.25% a year, so each day subtracts 4.25% divided by 252.

Positions held open at the end are marked at each close, so they are in the daily returns.

Example: daily returns beat the risk-free rate by 0.1% on average with a 1% standard deviation. The Sharpe ratio is 0.1 divided by 1, times 15.87, about 1.6.

Orbital shows the deflated Sharpe ratio with the number of configurations tried beside it. They appear in the strategies list's Full columns, Run properties and get_backtest.

Slippage is explicit on every run and shown in the strategy's settings line, for example 0.05% slippage. The report lists its total cost beside commission and borrow.

It is a stand-in, not a measurement. Orbital reads bars, not quotes, so the bid-ask spread and market impact are not modelled, and the report says so.

Higher is better, and it is usually above the same run's Sharpe ratio. Orbital computes it on daily close-to-close equity returns above a 4.25% a year risk-free rate. The mean daily excess return is divided by the downside deviation, taken over negative excess days only, then multiplied by √252.

Positions held open at the end are marked at each close and included.

Example: a mean daily excess return of 0.1% and a downside deviation of 0.7% give 0.1 divided by 0.7, times 15.87, about 2.3.

There are three kinds. Rules strategies are built from rule cards in the workflow editor. Program strategies are Rust code Orbital compiles. Signals strategies replay a dated table of buy and sell rows you supply.

A saved workflow that qualifies is enrolled on save. Backtesting lists every strategy with its state, Draft, Tested or Paper, its latest result and its trend against the benchmark.

Orbital's bars are 1d, 3h, 2h, 1h, 30m, 15m, 6m, 5m, 3m, 2m and 1m. The Indicator card's menu offers daily, hourly, 3-minute, 2-minute and 1-minute; the Program node's Decision bar offers them all.

A decision is taken at the bar's close and fills at the next bar. Daily decides once per session; a 1-minute Program decides about 391 times a session, so the run takes longer. Any intraday bar moves the run onto the minute tape, which the estimate prices differently.

A configuration is a version plus its run settings, with the period left out. Running one configuration over another period adds a run but not a trial. Forward test updates never count. A configuration whose runs all failed still counts as tried.

The deflated Sharpe ratio and this count appear together in the strategies list's Full columns, Run properties and get_backtest. The strategy page's Sharpe tile and statistics show the raw Sharpe alone.

The versions card also shows runs per version and runs across the whole strategy, which are different numbers.

Example: a strategy tried at 10, 14 and 20-day lookbacks, each over two periods, has six runs and a trial count of 3.

Before a run starts, a symbol listed too recently to be warm at the first session is dropped and named in the result. The run is refused instead when fewer than two symbols would remain, or more than half would go. That would be a different strategy, or the wrong period.

A rule strategy reads its own bars for every symbol; no other data node is needed.

A version is made on purpose: Save version in the editor, or a backtest of code that changed. Autosave never makes one.

Backtesting also records each run's configuration: the version plus the universe, bars, capital, margin, costs, fill policy and benchmark. The period is left out, so one configuration can be tested over several periods and compared. A changed setting is a new configuration with its own trial count.

For rule cards the warm-up is the deepest window any indicator reads, plus one bar the engine adds for the cross operators. A 200-day moving average reads 200 bars, and RSI, ATR and an N-bar return read one more than their period. The engine takes the deepest of those and adds one. It pads that into calendar days, about 1.55 times the bars plus 10 days, before the period starts.

For a Program it is the node's History the program reads, in bars of the decision bar, up to 20,000. The program is not called on these bars.

A symbol listed too recently to be warm on the first day joins the run at its own first warm session, rather than trading from the start. The run's warning names each such symbol and the date it is warm from. A run in which no symbol is ever warm is refused.

Higher is not always better. A low win rate with large winners can beat a high one with small winners. The profit factor and expectancy say how big the wins and losses were. Orbital computes winners divided by closed trades, times 100. Positions still open at the end are not counted.

Example: 55 of 100 closed trades made money. The win rate is 55%.

Members, billing, plugins, app installs and API keys belong to the workspace; strategies, workflows, dashboards, watchlists and agents belong to a project inside it. Switch workspaces from the avatar menu.