Evolution Brain
Most strategies aren't good or bad — they're good somewhere and bad somewhere else. This finds which conditions carry yours, and which quietly bleed it.
Not one number for the strategy. A number for the strategy in each situation it trades.
Where the evidence says a setup doesn't pay, it stands down — and tells you why.
It narrows, tunes and reports. Retiring a strategy is your decision, never ours.
Whatever you build here is tested against real market conditions and graded out-of-sample, so you always know whether it's working — on your own strategy, not a marketing figure. How the proof engine grades.
We don't publish how conditions are grouped, how their edge is measured, or where the bar sits for standing down.
We publish what a module does and what it's for. We don't publish how it works. The method is the product — and anyone holding the method holds the product.
Evaluating this for a desk? Talk to sales and we'll go as deep as an NDA allows.
- 01A systematic sweep, not blind mutation
Your strategy is tested across the same option space you could set by hand — sessions, trend alignment, volatility regime, structure gates, exits — one lens at a time, then in combination.
- 02Chosen on one slice, confirmed on another
Winners are picked on a selection slice and must then hold on a sealed slice they never saw. A variant that only worked where it was chosen is reported and not shipped.
- 03Improvements arrive as challengers, never as replacements
A better variant is deployed as a paper copy running beside your original. It has to earn promotion from its own live record, and only you promote it.
- 04The width of the search travels with the result
"Best of 31 combinations, confirmed on 6 sealed trades" is attached to the delta. The same number invites the opposite decision depending on that context.
One strategy, three runs, as structure filters are added.
- No filters
- 222 trades−42.0%
- Pattern filter
- 35 trades−2.3%
- Session + pattern
- 17 trades+6.2%
The climb is real and so is the shrinking sample. The engine reports the improvement and the 222→17 collapse together, because a gain earned by discarding 92% of your evidence is a hypothesis, not a result.
Every one of these is a thing we could ship and choose not to. They are here because the limits are the part of a research tool you actually have to trust.
Improvements run as paper challengers; promotion is yours.
The coach optimises and reports. Retiring is the user's decision, always.
Trials tried and trades confirmed travel with every improvement.
One pipeline, one strategy shape end to end. What you backtest is byte-for-byte what papers and what exports — there is no re-implementation step where drift can hide.
The same learning system that runs our own research desk, pointed at your strategies. You choose which folders it may learn from. Improved copies must beat the original on forward paper trades before promotion is even offered — and a promotion that stops holding up is demoted back to testing by the evidence, never quietly kept.
≈ 0R over 87,000 sweeps across a decade. Unselected, the famous setup pays nothing.
Flat across 530,000 samples. We do not gate or size by volume anywhere — and we say so.
No edge over 1,401 graded trades on a decade of data.
A real out-of-sample edge, consistent across every yearly window we held out — fragile to tight stops, which the platform tells you rather than hiding.
Materially better than holding, measured across 286,000 fade situations.
These are our own studies, run on our own archive, and the negative results ship inside the product next to the positive ones. A platform that only ever finds edges is selling you something. Most ideas don't work; the value is knowing which — before money does the experiment for you.
Will the AI change my strategy without asking?
No. Improvements are deployed as paper copies running alongside your original, which is untouched. You promote one only if its own record convinces you.
How do I know an improvement is real and not curve-fitting?
Every winner is chosen on one time slice and confirmed on a sealed one it never saw, and the result carries how many combinations were tried and how many sealed trades confirmed it. When the lift sits inside what search luck alone produces over that many attempts, the page says so.
Draw your rules on a canvas, or just describe them in plain English. Either way you end up with a strategy the platform can test, trade and grade like any other.
Bring a strategy you already run. An MT5 expert or a Pine script comes in and gets held to the same standard as everything else here.
It reads the chart the way you do — structure, swings, ranges, compression, the shape of the candles — and it reads it as of that bar, never with hindsight.
Every losing trade is evidence. This reads all of them, finds what they had in common, proposes a change to the rule — and proves the change before it ships.
Every module has an opinion. This turns them into one call, with the reasoning attached — and it will only act on an edge that has actually proved out.
A backtest tells you what a strategy would have made. This tells you whether its decisions were actually any good — against history it was never shown.
It goes to paper the moment it earns it. Live stays off until you turn it on — and you're the only one who can.
A decade of minute-resolution history across thirty instruments — the thing that decides whether a backtest is evidence or an opinion with a chart attached.
Edges decay. A strategy exported six months ago is quietly rotting on someone's terminal, and nothing tells them. This does.
Bring a strategy you already trade.
Test it yourself — or bring your desk's questions to us.
No card required · sales is for desks and teams