SabrTrader Documentation

AI Strategy Builder: Build, Backtest and Deploy Strategies with AI

The SabrTrader AI Strategy Builder is a complete rule-based strategy engine with a natural-language AI layer on top of it. You describe a strategy in plain English — instrument, timeframe, logic and risk limit — and the builder pulls historical bars, drafts the entry, exit, sizing and risk rules, backtests them, reports in-sample and out-of-sample results, and proposes concrete refinements you can accept with a single reply. Every strategy it produces remains fully editable by hand, can be validated in the Strategy Analyzer and Optimizer, and can be deployed to a simulated or live account from Control Center > Strategies.

Overview: What the AI Strategy Builder Is

The AI Strategy Builder is the strategy development workspace in SabrTrader. It combines two things in one window: a conventional manual strategy builder (entry conditions, exit conditions, position sizing, risk filters) and an AI assistant that can write, backtest and iterate on those same rule blocks from a plain-language prompt.

A trading strategy (also called an algo) is a codified set of entry, exit, sizing and risk rules that can be tested against historical bars and executed automatically by the platform. The AI Strategy Builder's job is to let you produce, validate and deploy such a rule set without writing code.

The workflow arc

1. Prompt — you state the instrument, timeframe, logic family and a risk constraint in one sentence.
2. AI builds and backtests — the assistant pulls bars, drafts rules, runs an internal backtest and returns a first-draft report with in-sample and out-of-sample statistics.
3. Iterative refinement — you accept suggested next steps (for example a stop/target sweep or an EMA period sweep) or type your own objective. Each reply creates a new saved version.
4. Validation — you open the best version in the Strategy Analyzer, set a realistic date range, commissions and slippage, run a full backtest, then optimize on a risk-adjusted objective and review trades on the chart.
5. Deployment — you add the strategy in Control Center > Strategies, attach a connection and an account (a simulated account first), and press Start.

Scope and expectations

• The builder is a foundational component of SabrTrader, not an add-on. It sits alongside charting, order flow and the Trade Performance analytics and AI Coach.
• It is multi-asset. The same framework applies to futures, equities, ETFs, FX and crypto. The demonstration uses Tesla (TSLA) equity on a 5-minute chart precisely to show that the platform is not futures-only.
• The output of a first prompt is a starting point, not a finished algo. The value is the development loop — draft, measure, refine, revert, validate — not the first draft itself.
• You must supply your own AI provider account. SabrTrader does not resell model access. See AI Setup: Connecting an LLM Provider to SabrTrader.

Feature Reference: Components of the Strategy Builder

The following table lists every component covered on this page and what it does.

Component	Where it lives	What it does
AI provider & model selection	Preferences > AI	Selects which LLM company (Anthropic, OpenAI and other supported providers) and which model (for example Opus 5) powers the builder. Requires your own account/API key.
Saved strategy list	Left pane of the Strategy Builder	Lists every strategy you have created, with a Source column showing whether it originated from AI or was built manually.
Prompt / chat pane	Right pane of the Strategy Builder	Where you type prompts and read the AI's progress stream, draft reports and suggested next steps.
Starter prompt library	Prompt pane, before first message	Pre-written example prompts that demonstrate acceptable phrasing and give beginners a working starting point.
Manual rule blocks	Open any strategy from the list	Entry conditions, exit conditions, position sizing and risk filters, editable by hand on any strategy regardless of source.
Version history	Strategy record	Stores every iteration as a numbered version with its own metrics, so you can compare and revert.
Strategy Analyzer	Open in Strategy Analyzer	Full backtest configuration and report: symbol, connection, timeframe, session template, trading hours, date range, stop/target, commissions, slippage, equity curve, trade list and performance metrics.
Optimizer	Strategy Analyzer, lower panel	Sweeps the strategy's exposed parameters using one of two optimization algorithms against a chosen objective function.
Monte Carlo simulation	Strategy Analyzer, lower panel	Resamples or reshuffles the trade sequence to estimate the range of possible equity outcomes.
AI Backtest tab	Bottom tabs of the builder	One click inserts a prefilled backtest prompt and returns a structured good / weak / how-to-improve review with actionable choices.
AI Coach	Builder chat and performance tooling	Answers diagnostic questions such as "how do we smooth this out?" or "how do we reduce drawdown?" against the current results.
Show trades on chart	Strategy result panel	Opens the instrument chart with entry and exit markers for every backtested trade.
Control Center > Strategies	Control Center, rightmost tab	Hosts Add Strategy, parameter confirmation, connection and account assignment, and the Start/Stop controls for live or simulated execution.

Prerequisites: AI Provider, Model and Data Connection

Two connections are required before the builder is usable: an AI provider and a market data / broker connection.

1. Connect an AI provider

An AI provider is the company that hosts the language model and holds the billing relationship with you. An API key is the private credential that authenticates your requests and identifies the account they are billed to; it functions as a password and must never be shared or screenshotted.

1. Open Preferences > AI.
2. Select the provider. The demonstration uses Anthropic. OpenAI and the other listed providers are equally valid — the choice is a function of where you already hold an account and which models you prefer.
3. Select the model. The demonstration uses Opus 5, a high-capability reasoning model. Larger models generally produce better rule logic and better refinement suggestions; smaller models are cheaper and faster.
4. Paste your API key if the provider requires one.
5. Press Save. The setting is global: the same provider and model power the Strategy Builder, the AI Backtest tab and the other AI panels in SabrTrader.

Full credential setup, key handling and billing guidance are documented in AI Setup: Connecting an LLM Provider to SabrTrader.

2. Connect a data feed or broker

A broker connection supplies both historical bars for backtesting and order routing for live execution. The demonstration uses Charles Schwab (thinkorswim), which covers equities, ETFs, futures and options. See Connecting Your Schwab Account to SabrTrader.

• Without a data connection the AI cannot pull bars and the backtest will return no results.
• The connection you choose in the Strategy Analyzer determines which historical series the backtest reads and which venue a live strategy routes to.
• For futures-specific routing and prop-firm accounts see Connecting SabrTrader to Rithmic; for FX see Connecting an OANDA Forex Account; for crypto see Connecting Coinbase to SabrTrader.

Touring the Strategy Builder Window

The Strategy Builder opens as a two-pane window. Its appearance is deliberately plain; the functionality behind it is a full strategy development environment.

Left pane — saved strategies

• Every strategy you have created is listed here.
• A Source column marks each strategy as AI or Manual, so you always know whether the logic was generated or hand-built.
• Selecting a row loads that strategy's rule blocks and its chat/version history.

Right pane — prompt and chat

• This is where you type prompts and read the AI's output.
• Before your first message, a starter prompt library is displayed. These are pre-written examples that show acceptable phrasing and give you something to run immediately.
• Progress updates, draft reports, metric tables and suggested next steps all appear inline in this pane.

Manual building blocks

Opening any strategy reveals the manual editor. It contains the component types you would expect from a conventional strategy builder:

• Entry conditions — the rules that must be true for the strategy to open a position (indicator comparisons, price relationships, pattern conditions, time conditions).
• Exit conditions — the rules that close a position, including stop loss, profit target, trailing logic and signal-based exits.
• Position sizing — how many contracts or shares each signal trades: fixed size, currency-risk based, or volatility based.
• Risk filters — conditions that block trades entirely, such as session windows, maximum trades per day, volatility bounds and a maximum daily loss.

The important property: AI-generated strategies remain fully editable by hand. The AI writes into the same rule blocks you would use manually, so you can open an AI draft, change a single filter, and leave everything else untouched. Equally, you can build a strategy manually and then ask the AI to optimize it.

For a visual node-based alternative to rule blocks, see AlgoStudio Pro (Visual Strategy Designer).

Writing Your First Prompt

Prompt engineering for strategies is the practice of writing a natural-language specification precise enough that the AI builds the strategy you actually intended. A good strategy prompt contains four elements.

Element	Why it matters	Example
Instrument	Determines which bars are pulled and which tick size, contract value and session apply.	Tesla / TSLA, ES, EUR/USD, BTC/USD
Timeframe	Controls signal frequency, noise level and realistic trade count.	5-minute chart
Logic family	Tells the AI what kind of edge to encode: trend-following, mean reversion, breakout, pullback.	9 EMA pullback trend strategy
Risk constraint	Forces a hard risk rule into the first draft rather than bolting it on later.	Max daily drawdown is $500

The demonstration prompt

build a 9 EMA pullback trend strategy on a 5-minute Tesla chart. Max daily drawdown is $500

That prompt is deliberately broad, and broad prompts are a valid starting point — the AI fills the gaps with sensible defaults and then exposes them for refinement. The sentence encodes:

• Instrument: Tesla (TSLA) equity.
• Timeframe: 5-minute bars.
• Logic: buy pullbacks into the 9-period exponential moving average while the trend is intact.
• Risk: stop trading for the session once the account is $500 down.

Richer prompt templates

Once you understand the loop, tighten the specification. Each additional clause removes an assumption the AI would otherwise make for you:

• Long only. Only trade 09:45–15:30 ET. Skip the first 15 minutes after the open. (direction and session filter)
• Trade 100 shares per signal, maximum 3 trades per day. (position sizing and frequency cap)
• Require price above the 200 SMA on the 15-minute chart for longs. (higher-timeframe trend filter)
• Use a 1.5R profit target and an ATR-based stop at 1.5x the 14-period ATR. (risk/reward and volatility-scaled exits)
• Hold out the last two months as out-of-sample data. (validation split)
• Expose the EMA period, stop multiple and target multiple as optimizable parameters. (ensures the Optimizer has variables to work with)

Related prompting patterns for non-strategy objects are covered in Alerts: Conditional Builder and AI-Generated Alerts.

What the AI Does After You Hit Send

Sending a prompt starts a visible, multi-stage process. The chat pane streams status lines so you can see what the assistant is doing rather than waiting at a blank screen.

1. Pulling historical bars. The AI requests the instrument's price history from the active connection for the timeframe you specified.
2. Analysing. It examines the data series, the volatility regime and the session structure to choose reasonable default parameters.
3. Drafting rules. It writes entry conditions, exit conditions, sizing and risk filters into the strategy's rule blocks.
4. Running an internal backtest. It executes those rules over the historical bars using the platform's own backtesting engine, not an estimate.
5. Producing the first-draft report. It returns the rule summary plus performance metrics.

What the first-draft report contains

• A plain-language description of the rules it built, so you can confirm it understood the prompt.
• In-sample statistics — results on the data used to build and tune the strategy.
• Out-of-sample statistics — results on data held back and not used for tuning. This is the honest number.
• Headline metrics — net P&L, profit factor, trade count, win rate, maximum drawdown.
• A "next steps" block — a short list of concrete refinements the AI recommends based on what its own backtest revealed, such as sweeping the stop/target grid or testing alternative EMA periods.

Treat the next-steps block as a menu, not instructions. Each suggestion is one hypothesis about why the current version underperforms.

Iterative Refinement and Suggested Next Steps

Refinement is conversational. You either click a suggested next step or type your own instruction; the AI edits the strategy, re-runs the backtest and reports the change. Every reply is effectively an edit that produces a new saved version.

Demonstrated refinements

Instruction	What the AI does	Demonstrated result
sweep the stop/target grids	Tests a matrix of stop-loss and profit-target combinations together and reports which pair performs best on in-sample and out-of-sample data.	Improvement of +$705 in out-of-sample net.
sweep the EMA periods	Re-runs the strategy across a range of EMA lengths to find whether a different period produces a more robust pullback reference.	Evaluated; kept only if out-of-sample results improved.

Free-form objectives

You are not limited to the suggested actions. Any instruction expressed in plain language is applied logically to the strategy. Useful examples:

• optimize for a 50/50 blend of Sharpe ratio and net profit — a blended objective, which balances raw profit against return consistency.
• add a filter that blocks trades in the first 15 minutes of the session
• reduce the maximum drawdown even if net profit falls
• run a three-way sweep across EMA period, stop and target
• show me the worst five trades and tell me what they have in common

Why one change at a time matters

Each sweep adds degrees of freedom. Sweeping stop, target, EMA period, session window and trade cap in a single step almost guarantees curve fitting — tuning the strategy so tightly to historical noise that it fails on new data. Change one block, measure the out-of-sample effect, then decide whether to keep it.

Version History and Reverting

Strategy versioning means every iteration is saved as a numbered snapshot of the strategy's complete logic together with the metrics it produced. Nothing is overwritten.

What you can do with versions

• Open any version to inspect the exact rules it contained.
• Compare in-sample and out-of-sample results across versions to see which change actually helped.
• Revert by instruction — typing revert to version two restores that version exactly, rules and parameters included, and the AI confirms the restore with the version's metrics.
• Continue iterating from the reverted version, so a bad branch costs you nothing.

When to revert

Revert as soon as a change degrades out-of-sample performance. The common failure mode is stacking: a sweep hurts the out-of-sample number, so the trader adds another filter to compensate, then another. Each layer is fitted to the previous layer's damage and the strategy becomes brittle. Reverting to the last version that was clean on unseen data is almost always the correct response.

Keeping notes

Record why each version was abandoned — "v4: EMA sweep improved in-sample +$1,200 but out-of-sample went negative, reverted to v2". Version numbers alone do not preserve reasoning, and six versions later you will not remember.

Strategy Analyzer: Settings Reference

The Strategy Analyzer is the full backtest and optimization workspace. The AI's internal backtests are fast approximations of your intent; the Strategy Analyzer is where you control every input explicitly and produce the result you will actually judge the strategy on. You can open the current version in the Analyzer at any point in the conversation.

Field-by-field

Setting	Meaning	Guidance
Symbol	The instrument to test.	Carried from the strategy (TSLA in the demonstration). Confirm it matches the symbol the rules were designed for.
Connection / data source	Which feed supplies the historical bars.	Schwab (thinkorswim) in the demonstration. Different feeds can produce slightly different bars and therefore different results.
Timeframe / interval	Bar size the strategy runs on.	5-minute in the demonstration. Must match the timeframe in the prompt or the logic changes meaning.
Session template / trading hours	The time window in which the strategy may generate and manage trades.	A regular-hours template excludes pre-market and after-hours bars. An overnight/ETH template includes them. Mismatched templates are the most common cause of "no trades".
Backtest date range	Start and end dates of the historical test.	Roughly six months in the demonstration. Longer windows covering multiple volatility regimes are more informative.
Stop	Protective exit distance.	Carried from the current strategy version.
Target	Profit-booking exit distance.	Carried from the current strategy version.
Commissions	Per-trade broker and exchange fees deducted from results.	Leaving the field at 0 means "use the commission settings configured in Preferences" rather than "charge nothing". Verify those preference values are realistic for your broker.
Slippage (ticks)	The modelled difference between expected and actual fill price.	1 to 2 ticks is a reasonable default for liquid instruments. Use more for thin instruments or stop-market exits.
Run Backtest	Executes the test.	Approximately 10 seconds for a six-month 5-minute test in the demonstration.

Reading the output

• Equity curve — cumulative strategy P&L plotted over time. Shape matters more than the end point: a curve that loses for three months and then recovers is a different animal from one that grinds up steadily.
• Net result — total profit or loss after commissions and slippage. In the demonstration the strategy ends profitable but opens with a poor stretch, which is explicitly flagged as the thing to work on.
• Trade list — every simulated trade with entry time, exit time, direction, size and P&L.
• Performance metrics — profit factor, win rate, average win and loss, maximum drawdown, trade count and risk-adjusted statistics.

Optimizer, Objective Functions and Monte Carlo

Optimization is the search for parameter settings that maximize a chosen objective function. The Strategy Analyzer's optimizer automates that search.

Variables

The optimizer can only sweep the parameters the strategy actually exposes. In the demonstration, only stop and target are exposed, so those are the only two variables available. If you want to optimize an EMA period, a filter threshold or a trade cap, the strategy must expose it — ask the AI to expose the EMA period as an optimizable parameter and it will rewrite the rule block accordingly.

Algorithms

Two optimization algorithms are selectable:

• Exhaustive grid search — tests every combination in the defined ranges. Complete and reproducible, but the run time grows multiplicatively with each added variable.
• Guided / heuristic search — explores the parameter space intelligently and converges on strong regions without testing every cell. Much faster with three or more variables, at the cost of not guaranteeing the global best.

Objective functions

Objective	Definition	When to use it
Net P&L	Total profit or loss after commissions and slippage.	Rarely alone. It rewards a single outlier trade and ignores the path taken.
Profit factor	Gross profit divided by gross loss. Above 1.0 means the strategy made money.	Good for comparing efficiency across parameter sets with similar trade counts.
Sharpe ratio	Excess return divided by the standard deviation of returns — return per unit of volatility.	The default choice when the goal is a smoother equity curve and a smaller drawdown.
Blended objective	A weighted combination, for example 50% Sharpe ratio and 50% net profit.	When you want consistency without collapsing to a tiny, over-filtered strategy that barely trades.

In the demonstration the recommendation is to switch from net P&L / profit factor to Sharpe ratio and re-run the sweep specifically to attack the drawdown visible in the equity curve.

Monte Carlo simulation

Monte Carlo simulation randomly reshuffles or resamples the sequence of trades many times to estimate the range of equity outcomes the same edge could plausibly have produced. It answers "how unlucky could this strategy's ordering have been?" rather than "what happened?".

• It is available in the Strategy Analyzer alongside the optimizer.
• It is explicitly premature with too few trades. The demonstration declines to run it for exactly this reason.
• Practical guidance: at least 100 trades before Monte Carlo output is worth reading, and preferably 200+. Below roughly 30 trades, every statistic on the report — including profit factor and win rate — is noise.

Walk-forward validation

Walk-forward validation repeatedly optimizes on one window of history and tests on the next unseen window, rolling forward through the data. It is the strongest available defence against curve fitting because every result is measured on data the optimizer never saw. Where a formal walk-forward run is not configured, approximate it manually: optimize on the first four months, test the untouched final two months, and only accept the parameters if the held-out window holds up.

AI Backtest Tab and AI Coach

Two AI surfaces sit alongside the main prompt pane.

AI Backtest tab

The AI Backtest tab is launched from the tabs along the bottom of the builder. Clicking it drops a prefilled prompt into the chat — for example backtest my Tesla 9 EMA pullback trade on the 5-minute — and runs the test. The output is deliberately structured rather than a wall of numbers:

• What's good — the metrics and behaviours that are working, with the figures that support them.
• What's weak — drawdown depth, losing clusters, poor trade distribution, thin trade counts, regime dependence.
• How to improve — concrete, executable options rather than generic advice. In the demonstration these include revert to version two and run a three-way sweep.

Each option can be executed by replying with it. Replying revert to version two restores that version exactly and reports its in-sample and out-of-sample numbers back to you.

AI Coach

The AI Coach is the diagnostic conversational layer. You ask it questions about the current results in plain language and it answers against the actual backtest data:

• how do we smooth this out?
• how do we reduce the drawdown?
• why does the equity curve start badly?
• is this strategy dependent on one month of results?
• should I optimize for Sharpe instead of net profit here?

The same coaching concept applied to your own discretionary trade history — rather than a backtest — is documented in Trade Performance Analytics and the AI Coach and Automated Trade Journal & AI Coach.

Reviewing Trades on the Chart

Show trades on chart opens the instrument's price chart and plots trade markers — an entry marker and an exit marker for every trade the backtest generated. Statistics tell you what happened in aggregate; the chart tells you where and why.

What to look for

• Clustered losses. Several losing entries within the same hour or the same two days usually indicates a regime the strategy cannot handle, not random bad luck.
• Bad sessions and times of day. If the losers concentrate in the first fifteen minutes after the open or in the midday lull, a time-of-day risk filter removes them without touching the core logic.
• News bars. Large single-bar gaps or spikes that trigger entries point to a need for a volatility or gap filter.
• Counter-trend entries. Pullback entries taken while the higher timeframe is falling are a signal to add a higher-timeframe trend filter — see Syncing Higher Time Frames for Better Entries.
• Exits that consistently cut winners short or stops that sit just inside normal noise, both of which argue for re-running the stop/target grid.

Chart review is the main source of new filter ideas. Each observation becomes the next prompt: add a filter that blocks entries in the first 15 minutes after the open and re-test.

Deploying and Monitoring a Running Strategy

Once a version survives validation, it is deployed from the Control Center.

Add Strategy dialog

Open Control Center > Strategies (the rightmost tab) and click Add Strategy. The dialog asks for:

Field	Purpose	Notes
Name	A label for this running instance.	"Tesla test 88" in the demonstration. Use a name that encodes instrument, version and intent so multiple instances remain distinguishable.
Strategy	Which saved strategy to run.	Select the exact version you validated — for example the 9 EMA pullback trend 5-minute strategy.
Connection	Which broker or feed supplies data and routes orders.	Schwab in the demonstration. The connection must be live for the strategy to receive bars.
Parameters	Stop, target, EMA period and any other exposed inputs.	These default to the last-used values from the builder. Confirm them rather than assuming; they can be tweaked here.
Exchange	The venue for the instrument.	Prefilled from the strategy. Confirm it is correct for your account.
Account	Which trading account receives the orders.	Always select a simulated account first. This is the only field that separates a test from real money.

Starting the strategy

Clicking Start applies the strategy to the chart. After a moment the strategy label appears on the chart — "Tesla test 88" in the demonstration — confirming it is live. From that point the strategy runs on its own: when its entry conditions are met it places the order, manages the position per its exit rules, and stops trading for the day if its daily risk filter is breached.

Monitoring and stopping

• The strategy keeps running until you stop it. Stop or disable it from the same Control Center > Strategies tab.
• Changing parameters requires stopping the instance, editing and restarting; a running strategy should never have its logic changed mid-position.
• Verify the account label on the chart and in the dialog before every session. A sim-tested strategy accidentally attached to a funded account is the single most expensive mistake available in this workflow.
• Account-level protection independent of the strategy's own rules is configured in Risk Master: Profiles, Limits and Daily Locks. Use it as a second layer: the strategy's max daily drawdown governs the strategy, Risk Master governs the account.
• Order-management templates for discretionary or hybrid use are covered in ATM Strategies: ATR Stops, Brackets & Auto Trailing.

Multi-asset deployment

The demonstration uses Tesla, but the framework is asset-agnostic. The same prompt-build-validate-deploy loop applies to futures, equities, ETFs, FX and crypto, limited only by what your connected broker supports. Note that tick values, session templates, commission structures and realistic slippage differ by asset class and must be re-checked when porting a strategy from one instrument to another.

Trading Concepts Explained

Strategy logic

• Trading strategy (algo) — a codified set of entry, exit, sizing and risk rules that can be backtested and executed automatically. A rule set is not a strategy until all four parts exist.
• Exponential Moving Average (EMA) — a moving average that weights recent prices more heavily than older ones, so it reacts faster than a simple moving average of the same length. A 9-period EMA on a 5-minute chart tracks price closely and acts as dynamic support in an uptrend.
• 9 EMA pullback trend strategy — a trend-following approach that buys pullbacks into the 9-period EMA while the prevailing trend is intact, on the assumption that the EMA marks a short-term value area within a trend. The strategy requires a trend definition, a pullback definition and an invalidation point.
• Position sizing — the rule determining how many contracts or shares each signal trades. Fixed size is simplest; currency-risk sizing (risk a fixed dollar amount divided by stop distance) keeps risk constant as volatility changes.
• Risk filters — conditions that block trades regardless of signal quality: session windows, volatility bounds, maximum trades per day, and daily loss limits.

Risk measurement

• Drawdown — the peak-to-trough decline in account equity over a period. It measures the pain the strategy inflicts, not the profit it makes.
• Max daily drawdown — a hard currency limit on how much a strategy may lose in a single session before it stops trading. In the demonstration prompt this is $500.
• Stop loss — a protective exit that caps the loss on a position.
• Profit target — a predefined exit level where the strategy books profit.
• Stop/target grid — a matrix of stop and target combinations tested together, used to find a risk/reward pair that is robust rather than a single lucky cell.

Testing and validation

• Backtest — running a strategy's rules over historical bars to estimate how it would have performed.
• In-sample vs out-of-sample — in-sample data is used to build and tune the strategy; out-of-sample data is held back and used only to validate it on bars the tuning never saw. Only out-of-sample results carry evidential weight.
• Parameter sweep (grid search) — systematically testing many combinations of parameter values to see which performs best.
• Optimization — searching the parameter space for settings that maximize a chosen objective function.
• Curve fitting / overfitting — tuning a strategy so tightly to historical data that it fails on new data. Symptoms: excellent in-sample results, poor out-of-sample results, and a performance surface where neighbouring parameter values perform badly.
• Walk-forward validation — repeatedly optimizing on one window and testing on the next, rolling forward through history, to check that the edge persists in time rather than existing only in one period.
• Monte Carlo simulation — randomly reshuffling or resampling trade sequences to estimate the distribution of possible equity outcomes, including the worst plausible drawdown.

Performance metrics

• Net P&L — total profit or loss after commissions and slippage over the test period.
• Profit factor — gross profit divided by gross loss. Above 1.0 means the strategy made money; 1.5+ is generally considered usable for intraday systems once costs are modelled.
• Sharpe ratio — risk-adjusted return, measured as excess return divided by the standard deviation of returns. Higher means more return per unit of volatility.
• Objective function — the single metric the optimizer maximizes. Choosing it is a strategic decision, not a technicality.
• Equity curve — a chart of cumulative strategy P&L over time, used to judge consistency. Slope, smoothness and the depth and duration of its declines all matter.

Execution realism

• Commissions — per-trade broker and exchange fees. A high-frequency strategy that is profitable before commissions is often unprofitable after them.
• Slippage — the difference between expected and actual fill price, typically modelled in ticks per trade. Market and stop orders slip more than limit orders; thin instruments slip more than liquid ones.
• Trading session template / trading hours — the time window in which the strategy may generate and manage trades. Regular-hours and overnight sessions have different liquidity, volatility and spread characteristics.
• Sim (simulated) account — a paper-trading account used to validate automated execution without risking capital.
• Automated execution — letting the platform place, manage and exit orders when the strategy's conditions are met, without manual intervention.
• Broker connection / data feed — the link to a broker such as Schwab (thinkorswim) that supplies bars and routes orders.

Tutorial: From Prompt to a Running Strategy

1. Set the AI provider and model. Open Preferences > AI, choose a provider (Anthropic in the demonstration), choose a model (Opus 5), enter your API key if required, and press Save.
2. Confirm a data connection. Make sure the broker or data feed for your instrument is connected — Schwab (thinkorswim) in the demonstration. Without bars, nothing downstream works.
3. Open the Strategy Builder. Review the left pane of saved strategies and the Source column. Optionally open an existing strategy to see the manual entry, exit, sizing and risk-filter blocks.
4. Enter your prompt. Type a specification containing instrument, timeframe, logic family and a risk constraint. Example: build a 9 EMA pullback trend strategy on a 5-minute Tesla chart. Max daily drawdown is $500. Send it.
5. Watch the build. The status stream shows bars being pulled, analysis, rule drafting and the internal backtest. A first draft typically returns in well under a minute.
6. Review the first draft. Read the rule description to confirm the AI understood you. Note the in-sample and out-of-sample net, profit factor, trade count and maximum drawdown. Read the suggested next steps.
7. Run one suggested sweep. Reply with sweep the stop/target grids. Record the out-of-sample change. In the demonstration this produced +$705 out-of-sample net.
8. Run the next sweep. Reply with sweep the EMA periods. Again, judge it on the out-of-sample number only.
9. Revert to the best version. If a sweep degraded unseen-data performance, reply revert to version two (or whichever version was strongest). The AI restores it exactly and re-reports its metrics.
10. Open in the Strategy Analyzer. Confirm symbol, connection, timeframe (5-minute), session template and trading hours. Set a backtest date range of at least six months. Confirm the stop and target carried from the current version. Leave commissions at 0 to inherit your Preferences commission settings, and set slippage to 1–2 ticks.
11. Run the backtest. Expect roughly 10 seconds for a six-month 5-minute test. Review the equity curve, net result, trade list and performance metrics. Pay particular attention to the early part of the curve, not just the final value.
12. Optimize on a risk-adjusted objective. In the optimizer, select the available variables (stop and target in the demonstration), choose an algorithm, set the objective to Sharpe ratio or a 50/50 Sharpe/net-profit blend, and run the sweep to attack drawdown.
13. Run the AI Backtest tab. Click the AI Backtest tab to drop in a prefilled prompt. Read the what's good / what's weak / how to improve breakdown and execute one recommended action.
14. Ask the AI Coach a diagnostic question. For example how do we reduce drawdown? or how do we smooth this out?.
15. Show trades on chart. Open the chart with entry and exit markers and look for clustered losses, bad times of day and news-driven entries. Turn each observation into one new filter prompt, then re-validate.
16. Deploy. Go to Control Center > Strategies > Add Strategy. Enter a name ("Tesla test 88"), select the strategy, select the Schwab connection, confirm the parameters and exchange, and attach a simulated account.
17. Start it. Click Start. The strategy is applied to the chart and its label appears there. It now runs until stopped and will execute automatically when its conditions are met.
18. Monitor in sim. Let it run for weeks in simulation. Compare live sim fills against backtest expectations before considering any funded account.

Best Practices

• Always include a risk constraint in the prompt. A max daily drawdown or per-trade risk figure forces the AI to build risk management into the first draft rather than treating it as an afterthought.
• Keep out-of-sample data untouched. If you sweep parameters until the out-of-sample number looks good, that data is no longer out-of-sample — it has become part of the tuning set. Hold back a window you genuinely do not optimize against.
• Change one thing per iteration. One sweep, one filter, one parameter. If you change three things and results improve, you do not know which change helped or whether two of them cancelled out.
• Prefer risk-adjusted objectives. Optimize for Sharpe ratio, profit factor or a blended objective rather than raw net profit. Net-profit optimization reliably selects parameter sets carried by one or two outlier trades.
• Model realistic commissions and slippage. Set the commission schedule in Preferences to match your actual broker, and use at least 1–2 ticks of slippage. A strategy that only works at zero cost does not work.
• Require a meaningful trade count. Treat fewer than 30 trades as unreadable, 30–100 as indicative, and 100+ as the minimum for statistical confidence.
• Run Monte Carlo only once you have enough trades. Below about 100 trades the simulated distribution simply reproduces the noise in your small sample.
• Use the chart, not just the statistics. Show trades on chart on every serious candidate. Filter ideas come from looking at losers in context.
• Validate across regimes. A six-month window may contain only one volatility environment. Test a period containing both a trending and a choppy phase if the data is available.
• Paper trade in sim for weeks. Backtest fills are idealized. Simulated live execution exposes latency, partial fills, session-edge behaviour and data-gap handling.
• Keep notes on why you reverted. Version numbers do not store reasoning. A one-line note per discarded version prevents you from re-running the same failed experiment.
• Layer account-level protection. Configure Risk Master limits independently of the strategy's own daily loss rule so that a logic bug cannot exceed your account risk budget.
• Rehearse on recorded data. Market Replay lets you watch a strategy's triggers bar by bar in original sequence, which exposes behaviour that aggregate statistics hide.

Common Mistakes

Mistake	Why it fails	What to do instead
Accepting the first draft as a finished algo	The first draft is a structured starting point produced from a one-sentence specification. It has not been validated across regimes or stress-tested.	Treat it as version 1 of many. Run at least one sweep, one Analyzer backtest with realistic costs and one chart review.
Sweeping everything at once	Sweeping stop, target, EMA period, session and trade cap simultaneously produces a parameter set fitted to historical noise. It is textbook curve fitting.	Sweep one dimension per iteration and check the out-of-sample effect before moving on.
Unrealistic commissions and slippage	Zero slippage and zero commission can turn a losing intraday strategy into a profitable-looking one. The edge disappears on the first live fill.	Configure true broker commissions in Preferences and model 1–2 ticks of slippage minimum.
Judging a strategy on a short window	A six-week backtest may cover a single market regime. Results will not generalise.	Use at least six months, and prefer a window containing both trending and range-bound conditions.
Running an untested strategy on a live account	The account field in Add Strategy is the only separation between simulation and real money. A mis-selection risks capital on unvalidated logic.	Always select a sim account first and verify the account label before pressing Start.
Vague prompts	A prompt without symbol, timeframe or risk limit forces the AI to guess, and the resulting strategy may not be the one you intended to test.	Always specify instrument, timeframe, logic family and a risk constraint.
Optimizing only for net profit	Net-profit optimization selects for outliers and tolerates enormous drawdowns.	Optimize for Sharpe ratio, profit factor or a blended objective.
Ignoring an ugly start to the equity curve	A curve that loses heavily for months and then recovers has a real probability of being live during the losing stretch. The positive end point is survivorship of a specific start date.	Judge the curve's worst stretch and its duration, not just the final net result.
Running Monte Carlo on 20 trades	Resampling a tiny sample produces a wide, meaningless distribution that can be misread as confidence.	Build trade count first — longer window, more instruments, or looser entry conditions — then simulate.
Stacking fixes instead of reverting	Each compensating filter is fitted to the damage of the previous change, producing a fragile chain of dependencies.	Revert to the last version that was clean out-of-sample and branch again from there.

Plans, Trial and Availability

• Free download with a 14-day full-premium trial. The initial download is a fully functioning copy with every premium feature enabled for two weeks — roughly 10 trading days. No credit card and no payment are required.
• Free tier. After the trial, the free plan remains usable and includes 200+ standard indicators. See Free Indicators (Included in Every Plan).
• AI Strategy Builder sits in the advanced tiers. Additional premium tools are listed under Premium Indicators (Included in Paid Plans).
• Ultimate plan adds the higher-end automation tools: AlgoStudio Pro (visual strategy designer) and Auto Strategy Builder (AI strategy generation). The Ultimate plan is a lifetime licence.
• You must supply your own AI account. Model access is billed by your chosen provider (Anthropic, OpenAI or another supported provider), not by SabrTrader. The key is entered in Preferences > AI.

Frequently Asked Questions

Do I need my own AI account to use the SabrTrader Strategy Builder?

Yes. SabrTrader does not resell model access. You must hold an account with an AI provider such as Anthropic or OpenAI, obtain an API key from that provider, and enter it in Preferences > AI. Usage is billed by the provider based on the volume of tokens the builder consumes. Without a configured and funded provider account, the AI pane will not return results, although the manual strategy builder, Strategy Analyzer and Optimizer still work.

Which AI providers and models are supported?

Multiple providers are selectable in Preferences > AI, including Anthropic and OpenAI. The demonstration uses Anthropic with the Opus 5 model. All of the listed providers are viable; the choice is mainly a function of where you already hold an account and which models you prefer. Higher-capability reasoning models generally produce better rule logic and more useful refinement suggestions; smaller models are faster and cheaper per request.

Can I edit an AI-generated strategy by hand?

Yes, completely. The AI writes into the same rule blocks the manual builder uses — entry conditions, exit conditions, position sizing and risk filters. Open any AI-generated strategy from the left pane and every block is editable. You can also go the other way: build a strategy manually and then ask the AI to sweep its parameters or diagnose its weaknesses. The Source column only records where the strategy originated; it does not restrict editing.

How do I revert to an earlier version of a strategy?

Every refinement you make is saved as a numbered version with its own metrics. To restore one, type the instruction in the chat — for example revert to version two. The AI restores that version exactly, including its rules and parameters, and reports its in-sample and out-of-sample results back to you. You can then continue iterating from the restored version. Reverting is the correct response whenever a change degrades out-of-sample performance.

What does 'out-of-sample net' mean in the builder output?

Out-of-sample net is the net profit or loss the strategy produced on the portion of historical data that was deliberately held back from the building and tuning process. In-sample data is what the strategy was fitted to; out-of-sample data is unseen. Because any strategy can be made to look good on data it was tuned against, the out-of-sample figure is the one that carries evidential weight. In the demonstration, a stop/target sweep improved the out-of-sample net by +$705.

How long does a backtest take to run?

In the Strategy Analyzer, a roughly six-month backtest on 5-minute bars completed in about 10 seconds in the demonstration. Run time scales with the number of bars (longer date ranges and smaller intervals take longer) and with the complexity of the rule set. Optimizer sweeps take substantially longer because they run one backtest per parameter combination — an exhaustive grid of three variables with ten values each is 1,000 backtests.

What slippage and commission values should I use in a backtest?

Use 1 to 2 ticks of slippage for liquid instruments, and more for thin instruments or strategies that exit with stop-market orders. For commissions, leaving the field at 0 in the Strategy Analyzer means "use the commission settings configured in Preferences" — it does not mean trading is free. Set those preference values to match your broker's actual per-side costs. Underestimating either number is the most common way a backtest overstates a strategy's edge.

When should I run a Monte Carlo simulation?

Only once the strategy has produced enough trades for the resampling to be meaningful. The practical threshold is at least 100 trades, preferably 200 or more. Below about 30 trades, every statistic including profit factor and win rate is noise, and Monte Carlo will simply reproduce that noise as a wide distribution that can be misread as insight. In the demonstration the simulation is explicitly skipped as premature for exactly this reason.

What's the difference between the AI Backtest tab and the Strategy Analyzer?

The AI Backtest tab is a conversational review: one click inserts a prefilled prompt, the AI runs the test and returns a structured narrative organised as what's good, what's weak and how to improve, with executable actions such as reverting a version or running a sweep. The Strategy Analyzer is the explicit configuration workspace: you set symbol, connection, timeframe, session template, date range, stop, target, commissions and slippage yourself, then read the equity curve, trade list and full metric set. Use the AI Backtest tab for fast diagnosis and the Strategy Analyzer for the authoritative result you make decisions on.

How do I optimize for Sharpe ratio instead of net profit?

Two ways. In the Strategy Analyzer optimizer, select the variables to sweep, choose an optimization algorithm, and set the objective to Sharpe ratio before running the sweep. Alternatively, type the instruction directly into the AI chat — for example optimize this for a 50/50 blend of Sharpe ratio and net profit — and the AI applies that as the objective function. Sharpe-based optimization is the standard approach when the goal is a smoother equity curve and a smaller drawdown rather than a larger headline profit.

Which instruments can I automate — is it futures only?

No. SabrTrader is a multi-asset platform and the Strategy Builder framework applies to futures, equities, ETFs, FX and crypto. The demonstration deliberately uses Tesla equity to make this point. The practical limit is what your connected broker supports: Schwab (thinkorswim) covers equities, ETFs, futures and options; Rithmic covers futures; OANDA covers FX; Coinbase covers crypto. Remember that tick values, session templates, commission structures and realistic slippage all differ by asset class and must be re-checked when moving a strategy between instruments.

How do I run a strategy on a sim account before going live?

In Control Center > Strategies > Add Strategy, complete the name, strategy, connection, parameters and exchange fields, then set the Account field to a simulated account. Press Start. The strategy applies to the chart and executes automatically, but all fills are simulated and no capital is at risk. The account field is the only thing separating a test from a live deployment, so verify it before every start. Let the strategy run in sim for weeks and compare its simulated fills against the backtest's expectations before considering a funded account.

Why does my strategy show no trades in the backtest?

The usual causes, in order of frequency: the session template or trading hours exclude the period when signals would fire; the date range is too short or falls in a period without qualifying conditions; the symbol or connection is wrong so no bars were loaded; or the rules are too restrictive — too many stacked filters can make the entry condition practically unreachable. Widen the date range first, then relax the session filter, then ask the AI why does this strategy produce no trades? — it will inspect the rule set and identify the blocking condition.

Is the AI Strategy Builder available on the free plan?

The AI Strategy Builder sits in SabrTrader's advanced tiers, not the free tier. However, every new download includes a 14-day full-premium trial with all premium features enabled, so you can use the builder immediately without payment. After the trial the free plan remains usable and includes 200+ standard indicators. The highest-end automation tools — AlgoStudio Pro and Auto Strategy Builder — are part of the Ultimate plan, which is a lifetime licence.

What is included in the 14-day free trial?

The trial is a fully functioning copy of SabrTrader with all premium features enabled for 14 calendar days, which works out to roughly 10 trading days. No credit card and no payment are required. It includes the AI Strategy Builder, the Strategy Analyzer and Optimizer, premium indicators and the order-flow tooling. You still need your own AI provider account for the AI features, and your own broker or data connection for live bars.

How do I stop a running strategy?

Return to Control Center > Strategies, select the running instance by name, and stop or disable it. The strategy continues to run until you do this. Stopping the strategy halts new signal generation; how existing open positions are handled depends on the strategy's exit configuration, so confirm the position is flat afterwards. Never edit a running strategy's parameters mid-position — stop it, edit, then restart.

Why do results differ between the AI's internal backtest and the Strategy Analyzer?

Because the two runs use different inputs. The AI's internal backtest uses the range, session and cost assumptions it chose; the Strategy Analyzer uses whatever you configured. The usual culprits are a different date range, a different session template, and different commission and slippage values. Align all three and the results converge. If they still differ, check that the Analyzer is pointed at the same connection and the same strategy version.

What makes a strategy overfitted, and how do I detect it?

A strategy is overfitted (curve fitted) when its parameters are tuned so tightly to historical data that they capture noise rather than a repeatable edge. Three detection signals: (1) a large gap between strong in-sample and weak out-of-sample results; (2) a fragile parameter surface, where a small change to the stop or EMA period collapses performance; and (3) an unusually large number of parameters or filters relative to the trade count. The defences are sweeping one variable at a time, preferring risk-adjusted objectives, holding out a genuine validation window, and using walk-forward validation.

Can the builder expose more variables for the optimizer?

Yes. The optimizer can only sweep the parameters the strategy actually exposes — in the demonstration, only stop and target. To make more variables optimizable, ask the AI to parameterise them: expose the EMA period and the trend filter length as optimizable parameters. It will rewrite the rule blocks so those inputs appear in the optimizer's variable list. Be disciplined about this: every additional variable multiplies the parameter space and increases overfitting risk.

Does the strategy keep running if I close the Strategy Builder window?

Yes. Once a strategy has been started from Control Center > Strategies, it runs as a platform-level process attached to the selected connection and account, independent of whether the Strategy Builder window is open. The strategy label remains visible on the chart. It continues until explicitly stopped, so confirm its state at the end of each session rather than assuming closing a window disabled it.

Troubleshooting

The AI does not respond when I send a prompt

Check three things in order:

1. Preferences > AI — confirm a provider and model are selected and that you pressed Save. An unsaved selection behaves as if nothing is configured.
2. API key — confirm the key is present, has not been revoked, and was pasted without leading or trailing whitespace.
3. Provider account — confirm the account is funded or has available credit and has not hit a rate limit. Most provider errors surface as a silent non-response rather than an explicit message.

Full credential handling is covered in AI Setup: Connecting an LLM Provider to SabrTrader.

The backtest returns zero trades

• Session template / trading hours — a regular-hours template will exclude every signal a pre-market or overnight strategy generates. Switch templates and re-run.
• Date range — too short a window, or a window without the conditions the strategy requires. Extend to six months or more.
• Symbol and connection — verify the symbol is correct for the selected connection and that bars actually loaded.
• Over-restrictive rules — stacked filters can make the entry condition unreachable. Remove filters one at a time to find the blocking condition, or ask the AI why does this strategy produce no trades?

Data is not pulling — the AI cannot load bars

Verify the broker or data feed connection is live in the Control Center connections area. A disconnected or expired session prevents historical bar requests. Reconnect, confirm the status indicator, then re-send the prompt. Also confirm your data entitlement covers the asset class and depth you are requesting — some plans supply delayed data or omit certain exchanges. See Connecting Your Schwab Account to SabrTrader.

The strategy will not start from Control Center

• Account — confirm an account is selected in the Add Strategy dialog. A blank account field blocks startup.
• Connection status — the chosen connection must be connected, not merely configured.
• Market hours — some instruments will not accept a strategy attachment outside their trading session. Retry during the instrument's session.
• Parameters — confirm every required parameter has a value; parameters default to the last-used values from the builder but can be blank if the strategy was edited.

The optimizer shows no variables to sweep

The strategy exposes no optimizable parameters. The optimizer can only sweep inputs the strategy explicitly declares. Return to the builder and ask the AI to parameterise them — for example expose the EMA period, stop and target as optimizable parameters. Re-open the Strategy Analyzer afterwards and the variables will appear in the optimizer's list.

Results differ between the AI backtest and the Strategy Analyzer

The two runs are using different inputs. Align them one by one:

1. Date range — set the Analyzer to the same start and end dates the AI used.
2. Session template and trading hours — a different session window changes which bars are eligible.
3. Commissions and slippage — the Analyzer applies your Preferences commission schedule and the slippage ticks you entered.
4. Strategy version — confirm both are pointed at the same version; the chat may have advanced to a newer one.

Out-of-sample results collapse after an optimization sweep

This is the classic overfitting signature. Do not add compensating filters. Instead, revert to the last version that was clean on out-of-sample data (revert to version two), then re-approach the problem with fewer variables, a wider parameter step size, and a risk-adjusted objective such as Sharpe ratio rather than net profit. If the strong result exists only at one specific parameter cell and neighbouring cells perform badly, treat that cell as noise, not an edge.

The equity curve is profitable overall but opens with a long losing stretch

This means the final net result depends on your start date, and a live deployment during that stretch would have been painful or fatal. Investigate with Show trades on chart: look at where the early losses occurred, which session and which market regime. Common fixes are a higher-timeframe trend filter, a volatility floor or ceiling, or a time-of-day filter that removes the losing window. Then re-test and check whether the improvement survives out-of-sample.

Live or sim fills do not match the backtest

Backtest fills are idealized. Divergence usually comes from slippage that exceeds your modelled ticks, latency between signal and order, partial fills on larger size, or thin liquidity at the entry price. Increase the modelled slippage until the backtest approximates observed sim results, then re-evaluate whether the strategy still has an edge at realistic costs. Rehearsing on recorded tick data via Market Replay exposes this early.

The AI built a different strategy from the one I described

Read the plain-language rule summary in the first-draft report — it states exactly what was built. Where it diverges from your intent, reply with a correction rather than starting over: you used a crossover entry; I want entries only on a pullback that touches the 9 EMA and closes back above it. Each correction creates a new version, so nothing is lost. For future prompts, add the missing specificity (direction, session, sizing, exact entry trigger) up front.

Glossary

9 EMA pullback trend strategy

A trend-following approach that buys pullbacks into a 9-period exponential moving average while the prevailing trend is intact, treating the EMA as a short-term value area within the trend.

AI Coach

The conversational diagnostic layer that answers plain-language questions about current results, such as how to smooth the equity curve or reduce drawdown, using the actual backtest data.

AI Strategy Builder

The SabrTrader workspace that combines a manual rule-based strategy editor with a natural-language AI layer that drafts, backtests and iterates on strategies from plain-English prompts.

Automated execution

Letting the platform place, manage and exit orders automatically whenever the strategy's conditions are met, without manual intervention.

Backtest

Running a strategy's rules over historical bars to estimate how it would have performed, producing metrics such as net P&L, profit factor, win rate, drawdown and trade count.

Broker connection

The link to a broker such as Schwab (thinkorswim) that supplies market data and historical bars and routes orders for execution, and exposes account balances and positions.

Commissions

Per-trade broker and exchange fees that reduce net performance and must be modelled for a realistic backtest. In the Strategy Analyzer a value of 0 means 'use the commission settings from Preferences', not 'charge nothing'.

Curve fitting / overfitting

Tuning a strategy so tightly to historical data that it captures noise rather than a repeatable edge and subsequently fails on new data. Detected by a large in-sample versus out-of-sample gap and by fragile parameter surfaces.

Drawdown

The peak-to-trough decline in account equity over a given period, measuring the depth of loss a strategy inflicts rather than the profit it produces.

Equity curve

A chart of cumulative strategy profit and loss over time or trade sequence, used to judge consistency. Its slope, smoothness and the depth and duration of its declines matter more than its final value.

Exponential Moving Average (EMA)

A moving average that weights recent prices more heavily than older prices, so it responds faster than a simple moving average of the same period. Used as a trend reference and as dynamic support or resistance.

In-sample vs out-of-sample

In-sample data is the historical data used to build and tune a strategy; out-of-sample data is held back and used only to validate the strategy on bars the tuning process never saw. Only out-of-sample results carry evidential weight.

LLM provider / API key

An external AI account (Anthropic, OpenAI or another supported provider) connected in Preferences > AI that powers the builder. The API key is a private credential that authenticates requests and identifies the billing account; it functions as a password.

Max daily drawdown

A hard currency limit on how much a strategy may lose in a single session before it stops trading for the day.

Monte Carlo simulation

Randomly reshuffling or resampling a strategy's trade sequence many times to estimate the range of possible equity outcomes, including the worst plausible drawdown. Requires a large trade sample — at least 100 trades — to be meaningful.

Multi-asset trading

Applying the same strategy framework across futures, equities, ETFs, FX and crypto, limited only by what the connected broker supports.

Net P&L

Total profit or loss after commissions and slippage over the test period.

Objective function

The single metric an optimizer maximizes, such as net profit, profit factor, Sharpe ratio or a weighted blend of several metrics.

Optimization

Searching a strategy's parameter space for the settings that maximize a chosen objective function, using an exhaustive grid or a guided search algorithm.

Parameter sweep (grid search)

Systematically testing many combinations of parameter values to see which performs best against a chosen metric.

Position sizing

The rule that determines how many contracts or shares each signal trades. Common forms are fixed size, fixed currency risk divided by stop distance, and volatility-scaled sizing.

Profit factor

Gross profit divided by gross loss. A value above 1.0 means the strategy made money over the test period; values around 1.5 or higher are generally considered usable for intraday systems once costs are modelled.

Profit target

A predefined exit level at which the strategy books profit on an open position.

Prompt engineering for strategies

Writing a natural-language specification precise enough — covering instrument, timeframe, logic family and risk constraint — that the AI builds the strategy you actually intended.

Risk filters

Conditions that block a strategy from trading regardless of signal quality, such as session time windows, volatility bounds, maximum trades per day and daily loss limits.

Sharpe ratio

A risk-adjusted return measure calculated as excess return divided by the standard deviation of returns. Higher values indicate more return per unit of volatility and generally a smoother equity curve.

Sim (simulated) account

A paper-trading account used to validate automated execution, order routing and strategy behaviour without risking capital.

Slippage

The difference between the expected fill price and the actual fill price, typically modelled in ticks per trade. One to two ticks is a reasonable default for liquid instruments.

Stop loss

A protective exit order that caps the loss on an open position at a predefined distance or price.

Stop/target grid

A matrix of stop-loss and profit-target combinations tested together in a single sweep to identify a risk/reward pair that performs robustly rather than in one isolated cell.

Strategy Analyzer

The SabrTrader workspace for explicit backtest configuration and optimization, covering symbol, connection, timeframe, session template, date range, stop, target, commissions and slippage, and producing an equity curve, trade list and full performance report.

Strategy versioning

Saved numbered snapshots of each strategy iteration, including its rules and metrics, so iterations can be compared and any earlier version restored exactly.

Trade markers on chart

Visual entry and exit markers plotted on the price chart for every trade a backtest generated, used to inspect losses in context and generate new filter ideas.

Trading session template

A saved definition of the time window in which a strategy is allowed to generate and manage trades, such as regular trading hours or extended/overnight hours.

Trading strategy (algo)

A codified set of entry, exit, sizing and risk rules that can be backtested against historical data and executed automatically by the platform.

Walk-forward validation

Repeatedly optimizing on one window of history and testing on the following unseen window, rolling forward through the data, to check that an edge persists in time rather than existing only in one period.

Related Documentation