Multiverse (synthetic robustness)
Generates thousands of markets that never existed and re-runs your algorithm in each one, to know if it has a real edge or just memorized the past.
The Multiverse is the toughest robustness filter in TradingNote. Your backtest only knows one price history: the one that happened. The Multiverse fabricates thousands of new histories —synthetic markets that never existed but are statistically identical to the real one— and re-runs your full algorithm in each. It answers the only question that truly matters: does your algorithm win because it has an edge, or because it conformed exactly to a past that won't repeat?
Where it lives: the Multiverse runs on a saved algorithm, from its page in «My Algorithms», or on the candidates of an exploration from the Algorithm Tester. It's not a separate menu: it's a validation tab of the algorithm.
How it works, step by step
- 1
Start from your algorithm and its real market
The Multiverse measures the statistical properties of the asset you trained on: its volatility, its jumps, its tendency to cluster in streaks.
- 2
Generates thousands of synthetic markets
It creates new price series that preserve those properties but with a different sequence. They're indistinguishable from a real market for any statistical test, but your algorithm never saw them.
- 3
Re-runs the algorithm in each universe
It runs your full system —rules, costs and sizing— on each synthetic market, and records whether it was profitable.
- 4
Gives you a distribution, not a number
The result is in how many of those thousands of universes your algorithm kept winning. That proportion is the robustness metric.
How to read the result
| % of profitable universes | Interpretation |
|---|---|
| High (the vast majority) | The edge doesn't depend on the exact history. Candidate for real money (alongside the other filters). |
| Medium | There's something, but fragile. Treat with caution and require it to pass Walk-Forward and DSR comfortably. |
| Low (only wins on the original history) | It memorized. Discard it without regret, however pretty its backtest was. |
Real data: in many markets, the Multiverse discards over 95% of high-Sharpe algorithms. It's not that the filter is too harsh — it's that 95% of those high Sharpes were noise disguised as signal. A filter that discards nothing isn't filtering.
A concrete example
Suppose two algorithms on EURUSD, both with a Sharpe of 2.1 in the backtest. Identical on paper. You run them through the Multiverse:
| Algorithm A | Algorithm B | |
|---|---|---|
| Backtest Sharpe | 2.1 | 2.1 |
| % profitable universes | 88% | 31% |
| Verdict | Robust edge — on to the next filter | Overfitting — discard |
The same Sharpe hid two opposite realities. Without the Multiverse, you'd have traded B with the same confidence as A — and B would have collapsed the moment the market changed its face.
Multiverse vs Monte Carlo: don't confuse them
Generates new MARKETS and re-runs the whole algorithm. It validates the EDGE: does it work in markets other than the one that trained it? It's the filter to decide if it deserves real money.
Reshuffles and varies the TRADES the algorithm already made. It validates the RISK: what drawdown should I be willing to take? It has its own page: see «Monte Carlo».
The Multiverse is one of the four angles of robustness, not the only one. Always combine it with the Deflated Sharpe Ratio (discounts sampling luck), Walk-Forward (data it didn't see) and Monte Carlo (a bad trade sequence). An algorithm that passes all four is the closest you'll get to a genuine edge.