The generator

Compose: invent a batch of strategies

Compose is where a strategy is born. You never write code: you pick a direction, set a few ranges (or let the AI set them), and one click invents a whole batch of fresh strategies, backtests every one on real market data, and ranks the winners as they land.

The Compose screen: choose archetypes, set the generator's dials, then generate and backtest a whole batch in one click.

Choose a direction

Pick from 34 archetypes across 6 families — Trend-following, Mean-reversion, Breakout / Volatility, Regime / Quant, Structure / Flow, and Sizing / Stateful. Tap a single archetype, tick Use a group to draw from a whole family (up to 6), or Use all for a random mix across everything.

The 34 archetypes across six colour-coded families — All 34 archetypes, colour-coded by family, so a batch's character is visible before you generate it.

Set the dials, by hand or with AI

The Generation Settings block has an AI / Manual toggle. In Manual you move the sliders yourself. In AIyou type what you’re after (e.g. “aggressive scalping with wide targets”) and one LLM call maps your words onto every dial, animating the sliders to its picks. Every dial is a range, not a single value, so the generator samples across it and the batch spans fast and slow ideas rather than fifty clones.

The generation dials — Each dial is a range the generator samples across. Defaults shown: Lookback 8–60, Reward:Risk 1.5–3.0R, Stop 1.5–4.0× ATR, Risk 1–2%.

Lookbackbars · 8–60: How many bars indicators look back over. Shorter = twitchier and reactive; longer = smoother and slower.
Reward : RiskR · 1.5–3.0: Take-profit distance as a multiple of the stop. 2.0R = target twice the stop distance; higher R needs a lower win rate to break even.
Stop × ATR× · 1.5–4.0: Stop distance as a multiple of Average True Range, so stops scale with each instrument's volatility instead of a fixed pip count.
Risk / trade% · 1.0–2.0: Percentage of the $10,000 account risked to the stop on each trade; position size is derived from this and the ATR stop.
Diversityspread of ideas: How far the generator wanders from the centre of each range. Low = tightly clustered variants; high = a wide spread of distinct ideas.
Idea pooltemplates: How many underlying strategy templates the batch is drawn from before variation: a bigger pool means more structural variety.
Confirmations1–4: How many entry conditions must agree before a trade fires. More confirmations = fewer, higher-conviction entries.

Shape and generate the batch

How many10 / 25 / 50: Batch size: how many distinct strategies to invent and backtest in this run.
Varyindicators · exits · entry · risk: Which parts the generator may change between siblings. Untick a dimension to hold it fixed across the whole batch.
Exit styleR:R · Trailing · Breakeven · Time · Signal: The exit mechanics allowed. Leave all off for “any exit”, or restrict to, say, trailing stops only.
Symbol & timeframes: The instrument and one or more timeframes, on real 2020–2026 data. 25 instruments, six timeframes. Default H1 + H4 (best for bots).
Seedre-roll · lock: A random seed drives the batch. Re-roll for a fresh set, or lock the seed to reproduce the exact same batch every time.
Presets: Save the whole Compose configuration under a name to reuse (Save / Load / Delete). Settings are also remembered per device automatically.

The winners land as ranked cards

Hit generate and the batch backtests live, filling in as each strategy finishes. Every result is a card (name, symbol, trade count, return and Sharpe) badged STRONG or WEAK so the winners are obvious. From a card you can Test (a fresh backtest), Load it into the Builder, or Save it to your library. With Automatic Validation on, the strong ones start validating across every year on the spot.

The generated strategies results grid — A finished batch: ten fresh strategies, each backtested and ranked. The top STRONG one is already validating across 2020–2026.

Not just one lucky year

Per-year validation: did it survive every year?

Generating is only half of it. One backtest can get lucky, so RoboticEA re-runs each strategy on every year of its instrument as separate, independent, replayable backtests, and shows a year-by-year grid of return and drawdown with a positive/negative tally. With Automatic Validation on, it does this to the batch's winners the moment they land.

Automatic Validation: hands-off proof, right in the batch

Turn Automatic Validation on in Compose, set the return threshold a strategy must clear to qualify (and, optionally, a limit to top N winners so only the best few validate), and every qualifying strategy in a batch is validated across every year the moment it lands. No extra clicks. When it finishes, the year-by-year grid appears right on the result card:

A generated card with automatic validation finished — Auto-validation done on a freshly-generated card: a +21.6% strategy that only cleared 2 of 7 years (worst DD -22%). The honest picture, without you lifting a finger.

The Automatic Validation setting — The setting, in Compose: “validate when return ≥ N%”, optionally limited to the top few winners per batch.

The full validated grid

Whether it ran automatically or you fired it by hand, the result is the same year-by-year grid: the mean %/yr, a “N of 7 profitable” tally, the range and worst drawdown, and a green/red chip per year. Click any year to replay it.

A validated strategy card with its per-year grid — A validated strategy in full: mean %/yr, profitable tally, worst drawdown, and every year's return — the losing years shown plainly.

Validate: Runs the strategy on every data year of its instrument and shows how each year performed. Click any year to watch that year's replay.
Quick Validate: On the saved-runs page, validate a strategy across all years in one tap without opening it.
Automatic Validationdefault off · threshold 25%: Turn it on in Compose and any strategy that clears your return threshold is validated across every year automatically, so survivors arrive already proven.

Saved strategies with Quick Validate — Your saved strategies. Quick-Validate any that haven't been tested across every year, then open a single year to replay it.

Read & edit

The Builder: open any strategy and change anything

Every strategy, whether the AI invented it or you pulled it from the library, is a readable schematic, not a black box. The Builder shows the indicators, the entry, the stop, the target and the sizing as blocks you can edit by hand, and it's where you set exactly how a backtest runs.

The strategy as blocks

The schematic draws the whole strategy as connected blocks — indicators feed the entry, which opens a long or short, governed by a stop, a target and an optional signal-exit. It’s a live map of what the strategy actually does.

A strategy schematic — Indicators → entry → long / short → stop · target · signal-exit. Every block is editable.

ModeCustom spec · Strategy library: Edit a fully custom spec, or drop in one of the ready-made library strategies as a starting point.
Indicatorsid · type · period · source: Each indicator is an editable block: rename its id, change its type (EMA, ATR, ADX, …), drag its period, and choose its price source.
Symbol / timeframe / year: What the backtest runs on. Year picks a single calendar year; Advanced / Custom Range opens an arbitrary date window.
Cost presetReal Spread (default): The friction model applied to the run (see §8). Real Spread is the honest default.
Risk per trade% of equity: The strategy's own risk, used on every run of it. Stored on the strategy, never silently overridden by account settings.
Validation status: A live check that the spec is valid before you run it: a green SPEC VALID badge means it will backtest and export cleanly.

From here, Run Backtest executes the strategy, or Refine Variations opens the refiner (§6) to breed and battery-test dozens of variants.

Prove it

The backtest: every number the engine owns

A purpose-built C++ engine runs each strategy from a fixed $10,000 account over real market data and reports the full metric set. Nothing is estimated after the fact — spread, commission and swap are charged inside the simulation, so the headline return is already net of friction.

A backtest result — The result header: a plain-English verdict and the return, with Sharpe, max drawdown and profit factor right beside it. Never a headline number on its own.

The Metrics panel reports the complete picture, all anchored to the same $10,000 start:

The metrics strip — Total return, CAGR, Sharpe, Sortino, Calmar, max drawdown, profit factor, win rate, expectancy, trades, and the real costs: commission, swap and average spread.

Total Return / CAGR: The net percentage gain over the window, and its compound annual growth rate.
Sharpe / Sortino / Calmar: Risk-adjusted return versus total volatility (Sharpe), downside-only volatility (Sortino), and worst drawdown (Calmar).
Max Drawdown: The deepest peak-to-trough equity fall: the loss you'd have had to sit through. Shown next to the return, always.
Profit Factor: Gross profit ÷ gross loss. Above 1.0 is profitable; the further above, the more cushion.
Win Rate / Expectancy / Trades: The share of winning trades, the average dollar outcome per trade, and how many round-trips: context for whether the numbers are meaningful.
Commission / Swap / Avg Spread: The friction actually charged: broker commission, overnight swap, and the mean spread paid per trade. Real costs, itemised.

The curve and the calendar

The account equity curve, with the worst peak-to-trough drawdown window shaded and an underwater strip beneath it.

Month-by-month returns — A month-by-month returns grid: the losing months in plain sight, not averaged away.

Watch it trade

Trade replay: bar by bar, on the real chart

A strategy-tester-style replay of the actual run: real OHLC candles with entry and exit markers and SL/TP lines, a synced equity curve building trade by trade, and live stats that update as you scrub.

Replay transport controls — The transport: play / pause, single-bar stepping, speeds 1×–8×, and a timeline marked with each trade's win / loss.

Playback1× · 2× · 4× · 8×: Play / pause, speed control, and single-bar stepping. Fully keyboard-driven (Space, arrows, Home / End).
Jump-tobiggest win · biggest loss · drawdown start: Skip straight to the moments that matter — the best and worst trades, or exactly where the worst drawdown began.
Live read-outs: Time, equity, return, drawdown, open position and realised P&L update as the replay runs: the behaviour, not just the summary.

Replay jumped to the drawdown — Jump-to-drawdown lands you at the exact bar the worst losing streak began, and shows you why.

Improve without breaking

The variation refiner: breed and battery-test

Take a proven strategy and spin up perturbed copies of it, then backtest every one against the same battery as the baseline. A variation is promoted only if it beats your objective AND keeps a robustness invariant — a “bigger headline” that breaks the invariant is flagged FRAGILE, not promoted.

The Refine Variations panel — Refine Variations: choose how many copies, how hard to perturb, which parts to change, the battery years, and the invariant every variant must keep.

Variations (N)1–50: How many perturbed copies of the baseline to breed and test.
Perturbation ±%5–60%: How hard to nudge each chosen parameter away from the baseline value.
Objectivemaximize: The metric a variation must beat to be considered: Total Return, Sharpe, Calmar or Max Drawdown.
Perturb which paramsindicators · exit · entry thresholds · risk sizing: Which parts are allowed to change. Untick a part to hold it fixed (risk sizing is off by default).
Battery years: The set of years every variation is tested over, so nothing is judged on a single window.
Robustness invariant: A per-period rule every test window must satisfy. The default is a positive return every year. Break it and the variant is rejected as fragile.
Auto-validate winners: Optionally validate every variation that clears your threshold across all years, with the baseline shown alongside for a year-by-year compare.

Start from proven ground

Preset library, saved strategies & leaderboard

You don't have to start from a blank prompt. Browse a large library of ready-made named presets plus tunable native strategy types, each with a “what's inside” schematic; everything you keep lands in your saved strategies, and every run you make lands on a personal leaderboard.

The preset library — Around 20,000 named presets plus 74 tunable native strategy types. Each shows its schematic; open one straight into the Builder.

The leaderboard — Every backtest you run is recorded and ranked, filterable by symbol, timeframe, year, cost preset and risk.

Preset library: Ready-made presets and native strategy types, each showing the indicators and structure inside before you run it.
Saved strategies: Everything you keep: each card carries its schematic, run count and validation state, with Open / History / Refine / MT5 Export / Delete and bulk select.
Leaderboard: Your own results, sortable and filterable, in a ranking view or a year-by-year matrix.

Honest by default

Realism, costs & trading defaults

The engine defaults to the honest case, and the knobs that change it are explicit. By default strategies trade with their natural behaviour and every run charges real friction. You opt in to idealised or bounded variants, never by accident.

Cost model

Every run carries a cost preset that decides the friction the engine charges. Real Spread is the default; the others exist for raw-account testing or frictionless debugging.

Real Spreaddefault: Recorded per-bar spread from real market data plus broker commission, with overnight swap always modelled. Friction lives inside the engine, not a bolted-on estimate. Required for gold & indices.
Flat 1 pipretail: A flat one-pip spread and no commission, a simple realistic-retail approximation.
Raw SpreadECN: A raw ~0.2-pip spread plus a per-lot commission, modelling a raw / ECN account.
Zerodebug: Frictionless: no spread, no commission. For sanity-checking a strategy's structure only, never for a realistic result.

Backtest realism

Leverage capUnlimited · 1:30 · 1:100 · 1:500 · Custom: By default positions size to their natural behaviour (unlimited leverage). Set a cap to test a margin-bounded version. A margin model then applies.
Margin stop-outwhen a cap is set: The equity level at which the broker force-closes positions once a leverage cap is active (default 5%).
Ruin flooralways on: The one rule you can't turn off: an account can lose at most 100% — it can never drop below $0. A guardrail, not a setting.

Trading defaults

Two account-level protections applied to every strategy: both at backtest time and, crucially, exported into the MetaTrader 5 EA as inputs so live behaviour matches.

Flatten before weekenddefault on: Closes any open position before the weekend. Avoids the 3× Wednesday swap, weekend gap risk and overnight financing. Exported as InpFlattenBeforeWeekend.
Avoid daily rolloverdefault on · 21:00–22:00 UTC: Skips new entries and exits during the toxic broker rollover hour where spreads blow out and stops get speared; deferred exits fire at the real post-window price. Exported as InpAvoidRollover.
Risk per trade: Deliberately NOT an account setting: risk is owned by the strategy (§3), so the exact value that produced a result is the value that runs and exports.

Ship it

Export to MetaTrader 5

When a strategy has earned it, one click writes a complete MetaTrader 5 expert advisor — the real .mq5 source, with a preview of the full code and an install guide. Drop it into your Experts folder, compile it once in MetaEditor, and it trades your exact strategy on your own broker and account.

Export & download. Click Export to MetaTrader on any completed run. RoboticEA generates the .mq5 source, runs it through a server-side compile gate (it’s never handed to you if it fails) and downloads a named file such as RoboticEA_…_XAUUSD_H1.mq5.
Put it in the Experts folder. In MetaTrader 5, open File ▸ Open Data Folder, then go into MQL5 ▸ Experts and drop the .mq5 there. The folder is usually C:\Users\<You>\AppData\Roaming\MetaQuotes\Terminal\<ID>\MQL5\Experts.
Compile it yourself.MetaTrader won’t compile a freshly-dropped file on its own. It only picks it up after you compile it (or restart the terminal). Open MetaEditor (in MT5: Tools ▸ MetaQuotes Language Editor, or press F4), open your RoboticEA_….mq5 from the Experts folder, and press Compile (toolbar button, or F7). Wait for 0 errors, 0 warnings.
Attach it. Back in MetaTrader 5, right-click the Navigator ▸ Refresh, then drag the Expert Advisor onto a chart of the same symbol and timeframe (or load it in the Strategy Tester) and enable AutoTrading.

The export actions on a run — One click on Export to MetaTrader. Included on the free tier.

Before it downloads, Preview source & install guideshows the full generated code with the compile-gate badge, so you can read exactly what you’re about to run before it ever touches a live account. The file downloads under a descriptive name, e.g. RoboticEA_TurboElderPowerSessionQuick_XAUUSD_H1.mq5.

The expert advisor running in MetaTrader 5 — The exported expert advisor running in the MetaTrader 5 terminal on a live chart.

Honest note: the MetaTrader tester won’t reproduce the engine’s numbers exactly. A different broker’s feed, spread, commission, swap and fill model all differ. Single-position strategies reproduce closely; stateful grid / martingale designs can diverge more. The export is for running the strategy, not re-deriving the backtest.

What it costs

Plans & tokens

Every plan unlocks the full engine — all 34 archetypes, 25 instruments, backtesting, per-year validation and MetaTrader 5 export. Work is metered in tokens, so pricing tracks what you actually run: only new backtests cost tokens; rerunning a saved strategy or watching a replay is always free.

Free tokens250 one-time: Every new account gets a one-time grant of 250 tokens, no card required, enough to compose, backtest, validate and export your first strategies.
What costs a token: One token per new backtest and one per validation-year: a Compose batch of 50 is 50 tokens; a variation battery is one per variation. A backtest that's already been computed (rerun, replay, history) is free.
STARTER20,000 tokens / month: A paid tier with a monthly token allotment. Billed monthly or yearly; see Pricing for the current rates.
PRO50,000 tokens / month: A paid tier with a monthly token allotment. Billed monthly or yearly; see Pricing for the current rates.
UNLIMITEDUnlimited tokens / month: A paid tier with a monthly token allotment. Billed monthly or yearly; see Pricing for the current rates.

See full pricing →