the goal stays on screen · finished phases fold away · the machine does the work and shows its proof
One path, top to bottom, in five phases: Set up → Discover → Configure → Prove → Ship. You never have to go looking for a setting — each step shows you what it is for, does the heavy work itself, and locks when you confirm. Finished phases collapse so the page stays short, and the bar at the top always says what you are building and why this step matters. Works on any market — forex, indices, stocks, crypto. Nothing is faked: if a step can't produce a genuine edge, it says so.
Every search in the Builder auto-saves that pair's top strategies here. Pick a pair + up to 3 strategies — the AI checks if they work together (low correlation = they cover each other's weak spots), merges them into one, then steps through timeframes & weightings until it finds the best combined setup, and shows you 2-week / 3-month / 12-month profit projections with success rates so you can see if it holds up. No experience needed — press the one button and the AI does it all.
select up to 3 above
⬆
Improve — upload an EA / script, evolve it, prove the gain
Upload a strategy EA Forge made — .eaforge.json, or a generated .mq5 / .pine (they carry the strategy inside them). EA Forge reads it, backtests it, then evolves it to beat itself and shows you exactly how much better it got. Then re-export the improved .mq5 / .pine / .json and it's saved to your library so every build makes the next one smarter. Note: a compiled .ex5 is machine code — it can't be read back. Upload its matching .mq5 instead. Non-EA-Forge scripts can't be parsed into the engine.
no file loaded
Expert manual toolkit — hand-pick indicators, rules, risk and run the analysis tools yourself.
Fastest path — do just these two
1 · GET DATA Real prices load automatically. Press GET DATA for fresh/other pairs. Never synthetic.
2 · GENERATE Answer 5 questions → one press = a complete, working strategy + code instantly. You could stop here.
Make it bulletproof (optional — but this is the real edge)
3 · AUTO-EVOLVE Breeds hundreds of variants, optimizes the winner, stress-tests it → PASS/FAIL on unseen data.
4 · RESEARCH FUNNEL Only keeps strategies that survive other timeframes, slippage & wiggled settings.
5 · BEST-TF + SCAN PAIRS Finds which timeframe & which of the 58 instruments has the strongest edge.
6 · CONTINUOUS SEARCH Runs until you stop, filling a ranked databank of the best-ever (profit + reliability + low DD).
8 · PROP CHALLENGE + PORTFOLIO Checks if it passes your prop firm's rules; bundle uncorrelated strategies into a portfolio.
The last step
9 · EXPORT— Apply the winner, then Download code (top bar): .mq5 → MetaTrader 5, .pine → TradingView. The generated code is the bottom card. Always demo-test before live.
✨
AI Build — guided setup
tap to open · build in 5 steps
Just want profitable EAs? Pick a pair + style, press one button.
Generates & validates hundreds of strategies, then hands you a ranked shortlist of the best that stayed profitable on unseen data. Pick one → Optimize & stress-test puts it through worst-case survival (walk-forward, Monte-Carlo, extra slippage) before you trust it. Honest: weak candidates are labelled, and if nothing survives it says so — a fake "winner" only loses money live.
i
Multi-symbol portfolio search (StrategyQuant style). ON = every candidate must stay profitable across a whole basket of related pairs at once (e.g. NAS100 + SPX500 + US30 for indices), not just the one you picked. A strategy that works on 4 correlated markets is far more likely to be a real edge than one curve-fit to a single chart. Slower, but the survivors are much more robust. OFF = single-pair search.
Bundled pairs work offline. ⬇ Max data pulls the deepest Yahoo history (~2y H1). Want 10+ years like StrategyQuant? Download free QuantDataManager → export Dukascopy CSV → upload CSV below. More data = more trustworthy results.
✓ Applied — refine it (StrategyQuant-style pipeline):
1 · Evolve — breed variations of this strategy & keep the best on unseen data.
2 · Auto-tune — after evolving, let the databank's results set a smarter accept bar. (⚙ Auto-tune buttons in the Continuous Search & Research cards)
3 · Research project — run the robustness gauntlet (OOS · other TFs · slippage · wiggled params).
4 · Best timeframe — find which TF has the real edge on this pair.
Prefer to guide it? Answer each step in order — the next unlocks once you fill the one before. Tap any ⓘ to learn what a choice means.
Pulls fresh Yahoo data (locally) or uses bundled real data. This app never uses synthetic prices.
i
Optional. If you have an idea, type it plainly — "buy gold dips when RSI oversold, London only, tight stop" — and it biases the build (adds those indicators/session/mean-reversion-vs-trend). Leave it blank and the machine finds a profitable strategy for you by testing thousands. You don't need an idea.
1
Broker
i
Broker = where your trades actually run. It sets the real trading costs (spread + commission) used in the backtest, so results aren't fantasy. A regular broker is your own account; a "prop" one is a funding firm with loss limits. Pick the one you'll trade with.
Where you'll trade. Sets realistic spread + commission costs.
2
Pair / symbol
i
Symbol = what you trade. XAUUSD is gold vs US dollar (volatile, trends well). EURUSD is the euro vs dollar (tight cost, ranges a lot). Indices like US30/NAS100 track stock markets. BTCUSD is Bitcoin (very volatile). Each has its own typical spread + behaviour.
The instrument the strategy trades.
3
Trading style
i
Style decides how often you trade and how long you hold. It auto-picks the indicators, entry rules, stop size and targets that suit that approach. Tap each option's ⓘ below to compare.
How you trade — sets indicators, stop, target and exits for you.
Scalp
i
Scalp — lots of very short trades (minutes). Buys quick oversold dips / sells overbought spikes, expecting a snap back. Tight stop, small 1.2R target, part-closes fast. Best on fast timeframes (M1–M5).
Intraday
i
Intraday — a few trades a day, all closed before the day ends. Follows the day's direction using a moving average + MACD momentum. Medium stop, 1.5R target. Good on M15–H1.
Swing
i
Swing — holds trades for days, riding bigger trends. Uses a fast/slow moving-average stack + ADX (trend-strength) to confirm. Wide 2× stop, 2.5R target. Best on H4–D1.
Breakout
i
Breakout — jumps in when price bursts out of a range with strong momentum (high ADX + MACD). Aims to catch the start of a fast move. Scales out at 1.5R and 3R with break-even.
4
Timeframe
i
Timeframe = how much time each price bar covers. M5 = 5-minute bars (fast, for scalping). H1 = 1-hour. D1 = 1 day (slow, for swing). Smaller bars = more trades but more noise; larger bars = fewer, cleaner signals. Match it to your style.
The size of each price bar. Match to your style (Scalp→small, Swing→large).
5
Account type
i
Account type decides whether a loss-limit guard is added. Choose the one matching how you'll run this strategy.
Your own broker account, or a funded prop-firm account with rules.
Broker
i
Broker — your own money. No forced daily loss limit; you manage your own risk. No drawdown guard is added.
Prop firm
i
Prop firm — the firm funds a bigger account but fails you if you lose too much in a day or overall. Picking this adds an automatic guard that halts trading before you breach the daily / max drawdown limit.
i
Generate builds the whole strategy from your 5 answers: it picks indicators + entry rules, sets stop-loss / target / exits, applies the broker's costs, and — for a prop account — switches on the drawdown guard. Then it writes ready-to-use MT5 and TradingView code below.
Complete step 1 to begin.
i
Auto-evolve & stress-test takes your generated strategy as a starting point and lets a genetic algorithm breed hundreds of variations, keeping the best on data it hasn't seen. Then it stress-tests the winner: split-sample, extra slippage, and wiggled settings, plus a Monte-Carlo reshuffle to estimate worst-case risk. You get a PASS/FAIL verdict — PASS means the edge held up, not just curve-fit luck.
Optional: harden the strategy beyond the template.
1
Describe your strategy
AI suggest
// suggestions appear here — click Apply to wire them into the build
2
Setup
basics
3
Indicators
signals
4
Entry logic
rules
ALL = every checked condition must be true. ANY = one is enough.
5
Risk & exits
money
ATM: OFF
Overrides the single Target above. Close part of the position at each R level; optionally move stop to break-even after the first tranche fills.
Exit Long when
Exit Short when
Add custom exit conditions in Card 4 → Build your own condition → Side = Exit Long / Exit Short.
!
Generated code is a starting template, not a validated edge. Backtest and forward-test on demo before any live capital. Prop rules (daily/max DD) are your responsibility.
6
Gates & guards
prop-safe
Markov gate: OFF
EA reads file each new bar. Contains BULL / BEAR / SIDEWAYS (or LONG/SHORT). Only trades in-bias direction; SIDEWAYS = no trade. Fail-open if file missing. Pine has no file I/O → exposes a manual regime input instead.
Funded phase: no profit target — the EA banks at most Consistency% of the withdrawal target each day (so no single day breaches the firm's consistency rule), then stops when total profit reaches the Withdrawal target, ready to pay out. Consistency auto-fills from the firm when known; edit to match your plan.
Used by the challenge check: it converts each trade's R into % of account (1R = your risk-per-trade %), then tests if the strategy reaches the target before breaching drawdown. Shown after Run backtest / Auto-evolve.
MQL5: tracks equity vs daily anchor + initial balance, halts when either limit hit. Pine: same via strategy.equity. Static (absolute) DD off initial balance — matches FundedNext Stellar / FTMO. Trailing-DD firms differ.
8
Strategy Evolver — genetic
mutate + evolve → OOS fitness
Set the options below, then press Start — it breeds hundreds of variants and keeps the best on unseen data.
data: —
Filters use in-sample metrics (the funnel). Cross-checks run once on the final databank and badge each survivor. Trades/month derives from the builder timeframe (Card 2).
// run evolution to populate
base TF: —
Runs the databank survivors through a robustness gauntlet, like an SQX project (Build → OOS → M30 → H4 → Slippage → Parameters). Resamples your loaded bars up to coarser timeframes (can't go finer than your data). Keeps only strategies that stay profitable on every enabled stage.
i
The robustness gauntlet. A strategy must stay profitable at every enabled stage or it's dropped. OOS split = tests two separate unseen windows. M30/H4/D1 = re-runs it on coarser timeframes (resampled up from your data) — a real edge survives a timeframe change. Slippage = re-tests with extra cost added. Parameters = nudges the settings ±and checks it doesn't collapse (not curve-fit). Toggle stages on/off by clicking.
OOS split (2 windows)
M30
H4
D1
Slippage
Parameters
i
Slippage bump = extra points of cost added on top of spread + commission during the Slippage stage, simulating worse fills than the backtest assumes (fast markets, wide spreads). A strategy whose edge vanishes under +5 pts was living on unrealistically perfect fills. Raise it to stress-test harder; a robust edge survives realistic slippage.
i
The pass bar for each stage. A strategy stays in the funnel only if, on every enabled stage, its profit factor ≥ this AND it takes at least this many trades. 1.1 = must stay at least marginally profitable everywhere; raise to 1.2–1.3 to demand a clearer edge on unseen conditions. Min trades guards against a stage passing on a lucky handful. Use ⚙ Auto-tune to set these from your databank's own results.
i
Which timeframe actually has the edge? It evolves a fresh, independent strategy on every achievable timeframe (your base plus coarser ones resampled up — M15→M30→M45→H4→D1) and ranks them by out-of-sample fitness. The winner is the TF where this instrument trades cleanest. Can't go finer than your data — load M1/M5 bars to scan the whole ladder; from M15 data you can only reach M15 and coarser.
i
Population size per timeframe — how many strategies compete each generation. Bigger = explores more of the search space (more thorough) but slower. 30 is a light default for a quick ladder scan; 60–100 for serious rigor.
i
Generations — how many breed-and-select rounds per timeframe. More = the population refines further toward the best strategy, but slower. 15 is quick; 30–40 for depth. ⚙ Auto scales both from how much data you loaded.
Evolves an independent population on each achievable timeframe (base + coarser via up-resampling), then ranks by best-strategy out-of-sample fitness. Answers "which TF has the real edge on this instrument". Uses a lighter GA per TF — bump pop/gens for rigor. Feed fine data (M1/M5) to scan the full ladder.
★
Databank + Continuous Search
every tool feeds this
data: —
Evolves round after round forever, keeping every strategy that beats the bar in a ranked databank. The longer it runs, the higher the bar (Deflated Sharpe deflates as more are tried) — so it hunts genuinely better strategies, not lucky ones. Leave it running; come back to a ranked wall of the best it found.
i
Optimize for = what the Score column ranks by, so the databank sorts best-first for your priority. Balanced = profit + reliability + low drawdown together. Max profit = ranks purely by net R (highest earner at #1). Max reliability = favours smooth, stable equity (high Stability/DSR). Min drawdown = favours the shallowest losing streaks. Changing it instantly re-ranks the whole list.
i
The accept gate — a strategy must clear BOTH to enter the databank. Min PF (profit factor) = gross win ÷ gross loss on the out-of-sample half; 1.1 lets marginal edges in, 1.3+ keeps only clearly profitable ones. Min trades = reject strategies with too few trades to trust (a 3-trade fluke isn't an edge); 15 is a sane floor, raise to 30+ for statistical confidence. Tighter = fewer but better survivors.
i
Max DD (R) = reject any strategy whose worst peak-to-trough drawdown exceeds this many R (risk units). Lower = only strategies that never dig a deep hole. Keep top = how many best-ever strategies the databank holds; when full, the weakest is dropped as better ones arrive. 25 keeps the list short and readable; raise it to hoard more candidates.
— run a while first, then let the results set a smarter accept bar.
i
Always sorted best-first by the Score column, which reflects your Optimize for goal — so #1 (★) is the top pick for that priority. Note: on Balanced, a strategy with slightly lower net R can outrank a higher earner if it's far steadier (better Stability/DSR/DD). Want pure profit order? Set Optimize for → Max profit and the highest net R jumps to #1. Re-sorts live as new strategies arrive.
// press Start — fills as it searches
9
Workflow — automation pipeline
20 task blocks
Click a block to append it to the pipeline. Each block carries its own parameters. Runner executes feasible steps top-to-bottom; flow blocks (Go To Task, Wait, Stop & Start) control the loop. Pipeline saves into your JSON export.
i
Reference — pulled & baked in
web-sourced Aug 2026
Daily / Max drawdown, profit target, and DD type per firm. Typical current values — verify with your firm & edit in Data Manager. Selecting a firm as Account type applies these automatically.
strategy.exit(trail_points, trail_offset) — both required, in ticks. Generator emits slDist/syminfo.mintick when ATR trail on.
London 07–16 · New York 12–21 · overlap 12–16
MT5: save .mq5 → MQL5/Experts/ → F7 compile → drag to chart. TradingView: Pine Editor → paste .pine → Add to chart. Markov file: your desk writes regime to the named file in MT5 Common/Files.
▾ Advanced settings — optional; the AI Build wizard already does all this. Expand a section, then tap its ⓘ to learn what it does.
7
Backtest Lab
walk-forward · monte carlo
Paste CSV
Upload CSV
Bundled real
Bundled real market data ships with the app — the default. Press GET DATA for fresh Yahoo prices (local), or Upload your own MT5 CSV. Never synthetic.
Deducted from every trade in R. MT5 CSVs auto-fill spread from the SPREAD column.
Load a CSV to see its date range. Synthetic data is dated up to now.
If a proxy fails it auto-tries several fallbacks, then a direct call. "Update ALL" refreshes every loaded/scanned instrument in one click.
Pulls live OHLC from Yahoo. Works when you run this app as a local .html file — inside claude.ai the sandbox blocks outside requests (you'll get a message). The proxy adds the CORS headers browsers need; swap it if one goes down.
Evolves a champion on every loaded instrument (upload multiple CSVs, or use + market) and ranks them by out-of-sample edge — the AI picks the best pair for you. Apply loads that pair + its strategy.
i
Ruin threshold = how deep a loss counts as a "blow-up" in the Monte-Carlo test. R = risk-per-trade units (1R = the amount one trade risks). −15 means: if the reshuffled equity ever drops 15R below the start, that run is a ruin. The app then reports what % of runs hit it = your risk of a catastrophic drawdown. More negative = you tolerate a deeper hole before calling it ruin.
i
Execution-stress test. Each Monte-Carlo path randomly throws away this % of trades — simulating missed entries (bad fills, slow broker, you were away). A real edge survives missing a chunk of its trades; a fragile one that leans on a few lucky trades falls apart. 0 = off (unchanged). Try 10–20% to see if the edge is broad or brittle. Does not change the strategy — only the test.
i
Execution-stress test. Subtracts this many R-multiples of extra cost from every trade in each Monte-Carlo path — worse spread/slippage than the backtest assumed. 0.05 = every trade does 0.05R worse. If a small extra cost erases the edge, it was never real. 0 = off. Does not change the strategy — only the test.
Base backtest stays in R-multiples (the edge). This layer prices that edge in dollars on a compounding account, so drawdown and risk-of-ruin read in real money — the number that matches a live/prop statement.
SL
TP-R
period
Rolling walk-forward: grid-searches params on each in-sample window, tests the winner on the next unseen window, stitches out-of-sample trades. Ranks by OOS expectancy — the honest number. Apply-best writes params back & regenerates code.
!
V
Data Vault — deep history, auto-kept
persistent · weekly · gap-checked
Downloads the deepest reliable history per pair and stores it on this PC (persists between sessions). Weekly it adds only the new bars — no re-downloading. Every series is gap-checked (no holes). Sources: Yahoo (TRUE daily back 20–56 years · hourly ≈ 2y) · Binance (deep crypto). Built vault feeds 🎯 automatically. For 15y+ hourly FX/gold, Yahoo can't — use QuantDataManager → Dukascopy CSV → upload (see the 🎯 note).
Empty — press Build. In the DESKTOP app it pulls via Node (no CORS, reliable). Inside claude.ai the browser blocks it.
10
Data Manager — broker profiles
costs · specs · prop
Applying sets the backtest spread + commission cost model, the builder symbol, and — for prop firms — the drawdown guard preset. Specs are typical ECN values baked in (CSP blocks live broker fetch); edit any cell to match your real fills.
11
Multi-symbol portfolio
evolve · correlation-cull · bundle
markets: —
Evolves a champion strategy on each loaded symbol (add via Backtest Lab + market, or load multiple CSVs), then greedily keeps the highest-OOS-fitness champions whose per-bar P&L stays below the correlation limit. Combined curve shows the diversification benefit (portfolio DD < sum of individual DDs).
12
Strategy Library
bulk · combos × timeframes
base TF: —
Creates many different strategies — random indicator combinations across the timeframes you pick — backtests each on unseen data, and ranks them. Apply any to load it. Higher counts + "optimize each" take longer.
M5
M15
M30
H1
H4
D1
Only timeframes coarser-or-equal to your loaded data can be built (finer needs finer data).
Grid-searches stop / target / period for the picked strategy on that timeframe, then reports the improvement, the best timeframe overall, the best days to trade this pair, and the strongest session.
↓
Generated code — the last step
MT5 + Pine
This is the final output. Everything above feeds it — it updates live as your strategy changes. Copy it, or use Download code (top bar) to save the .mq5 / .pine file.
MQL5 (.mq5)
Pine v6 (.pine)
EA Forge · builds MQL5 Expert Advisors + Pine Script v6 for TradingView
code regenerates live as you edit · export/import JSON to save a build · nothing leaves your machine