New Build page
Standard for quick picks, Explore for every setting.
Free for every Nebula user
The Edge Update
A new Build page, the Lean engine and risk you can see before you run.
- 1,000+compute hours
- 2.5M+systems tested
- 5.7M+price bars
- 36,000+systems checked on unseen prices
The testing behind this update, 7 to 25 October 2026.
What changes
Lean engine
Standard now builds with the newer Lean engine.
Risk you can see
Books start at their tested risk.
Gold book
Gold systems built and run together as one book.
Mission Control, new view
Systems and books in one view.
No broker needed to start
Build and test before you link MetaTrader.
What 1,000+ hours of testing looks at
The markets, the maths and the checks behind the update. The method stays ours.
5.7M+
Price bars on file
12 markets, four timeframes, M15 to H4. Every bar is a real broker price, with the spread it had at the time.
120x
Markets are not bell curves
On EURUSD, hourly moves past 4 standard deviations happen 120 times more often than a bell curve allows. Gold: 130 times. A test that assumes normal markets underrates the bad days.
The maths
Each hourly move is a log return r = ln(close / previous close), scaled to z = (r − mean) / sd. A normal curve puts 0.0063% of hours beyond |z| = 4. The data puts 0.762% there. Excess kurtosis: 15.9 on EURUSD, 24.2 on gold, where a bell curve scores 0. 100,530 EURUSD hours, 2010 to 2026.
3.5x
The market keeps hours
EURUSD moves 3.5 times more in its busiest hour than its quietest. Systems are tested across every session, so a quiet hour cannot pass for skill.
5σ
Test millions, and luck looks like genius
Try 2.5 million random systems and the best one looks like a 5 sigma result by chance alone. That is why the best looking system proves nothing, and why every system we keep is checked again on prices it never saw.
The maths
For N independent tries with no real skill, the best score lands near the 1 − 1/N point of the normal curve: Φ⁻¹(1 − 1/N). One search of 1,200 tries gives about 3.1σ. 2.5 million gives about 4.9σ. The bar a real result must clear rises with every system tested.
The yardstick
What luck looks like, and why we never trust it
1,000 systems with no edge at all, trading real EURUSD prices: a coin flip for direction, a fixed stop, the real spread. The best of them look brilliant on the prices they were built on. Then they meet prices they have never seen.
- top 20 on the prices they were built on
- the same 20 on the next 200 unseen trades
- luck band after 200 trades
Luck does not repeat. That is why no Nebula system is judged on the prices it was built on, and why every system kept is checked again on prices it never saw.
The maths
Each trade result is a number in R (1R = the stop). With mean μ and spread σ per trade, n trades add up to about nμ, with a spread of σ√n. Here μ = -0.006R (mostly the spread, about 0.015R a trade) and σ = 1.34R, so after 200 trades the middle 95% of pure luck runs from about -38R to +36R: the dashed band. Pick the top 20 of 1,000 and you have picked the luckiest 2%, not the best. Their next 200 trades are fresh draws from the same coin, so they fall back to about nμ. A real edge is the opposite: it keeps showing up on prices it has not seen.
Built onUnseenTop 20 by luck
How Nebula tells skill from luck
- Search wide.2.5M+ candidate systems tried. Most are luck, and the search knows it.
- Judge on unseen prices.36,247 kept systems re-run on prices they never saw: 19.0M trades.
- Beat random, not zero.Every result is compared with random systems of the same size on the same prices, so luck has a score to beat.
- Searched systems beat random ones on prices neither had seen.
- On both of our big test pools, the new Lean engine did better per trade than our old default.
How every test splits time
- Build. The search only ever sees this part.
- Check. Systems must hold up on prices they were not built on.
- Unseen. A final look, used once, after choices are locked.
- Held back. A sealed stretch nobody looks at until the rules are written down first.