ENGINE WHITE
PAPER.
SIGMA's architecture, quantitative methodology and risk management — how a strategy is validated before it risks a single dollar of real capital.
QUANTITATIVE DISCIPLINE,
NOT DISCRETION.
SIGMA is a multi-asset quantitative trading engine built and operated by a small group of professional traders alongside a team of AI agents covering research, execution, risk and infrastructure. It is not a signal bot or an indicator: it is a complete system for generating, validating and executing strategies, running on real capital since June 17, 2026.
Most retail traders don't lose for lack of good ideas — they lose for lack of process. SIGMA replaces discretion with a systematic, auditable process with explicit gates before any capital is put at risk.
The underlying mission is self-custody: every profit cycle is designed to eventually withdraw to cold storage. Not your keys, not your coins.
THE LANGUAGE OF THE MARKET
SIGMA is not an indicator or a moving-average cross that buys or sells. It's a protocol — layers that, together, read the market as a language: no signal is an isolated event, each one only makes sense within its context. With the same standard of honesty as the rest of this document, each layer is described by what already runs today on real capital and what is still roadmap.
THE UNIVERSAL RULES
The strategy generation and validation pipeline — Bayesian search and genetic programming, subject to the same robustness gate without exception. The most mature of the four layers: runs 24/7 across all five engines.
THE GRAMMAR OF CROSSES
A 0-100 rule filter that scores each signal against its context before entry: market regime, correlation clusters between assets, and trades already open. Already gates live entries, exposed via the API as sigma_gamma_score.
READING THE CURRENT STATE
Each asset's regime, computed in real time, determines which strategies are allowed to trade at any given moment. Visible today on the public HUD and in the per-engine readiness breakdown.
THE SCORE THAT SYNTHESISES THE EVIDENCE
A real per-model score: backtest CAGR, Calmar, frequency, win rate and profit factor, with the win rate adjusted by Bayesian credibility using real live evidence — plus statistical confidence and recency. It's the number that decides what runs live, exposed via the API as sigma_score.
We are not presenting a finished protocol. We are presenting how much of this vision is already code running on real capital, and how much is the road toward it.
WHY THE AVERAGE TRADER FAILS
SELECTION BIAS
Testing hundreds of variants and keeping whichever performed best in the past (in-sample) overstates its real out-of-sample edge.
NO OOS VALIDATION
Without walk-forward testing there is no way to tell a real statistical edge apart from a coincidence of the period tested.
DISCRETIONARY RISK
Without hard rules for position size and maximum drawdown, a bad streak turns into a blown account.
FIVE ENGINES, ONE SAME STANDARD.
SIGMA runs five engines in parallel, each with its own asset universe, optimisation pipeline and risk gates. Every asset/timeframe/direction combination is treated as an independent slot, with variants generated by the two search engines that run 24/7 across the entire universe (section 4) — none gets a shortcut to production without first passing the validation pipeline (section 5). Engines 3 through 5 trade tokenised Binance perpetuals that track the real underlying asset — they don't require a traditional broker to execute. A regime gate, based on a weekly reference EMA200, determines whether the market context favours trend, reversion, or no active strategy at all.
CRYPTO
USDT-M perpetuals. A deliberately closed universe — more assets isn't more edge, it's more noise and more surface for failure.
COMMODITIES
Gold, silver, WTI crude, copper, natural gas and platinum. Same validation pipeline, different market regime — a real diversifier, not just a nominal one.
US EQUITIES
Via tokenised Binance perpetuals that track the real underlying. Same validation gate as crypto and commodities. CVX excluded from production due to ticker collision (the perp tracks Convex Finance, not Chevron).
INDICES
Via tokenised Binance perpetuals. Shares its live capital budget with Engine 3 so it doesn't compete for liquidity against crypto/commodities.
INTERNATIONAL
Country ETFs — Japan, Taiwan, Korea — via tokenised Binance perpetuals. Geographic diversification under the same validation standard.
HOW A STRATEGY IS BORN.
No strategy in SIGMA is hand-written by a human. Two search engines run 24/7 in parallel, generating and evaluating thousands of variants across the entire asset universe — neither gets a shortcut to production without first passing the same validation pipeline (section 5).
BAYESIAN SEARCH
Explores the parameter space in a directed way — each prior trial informs the next, instead of random brute force. The primary tuning engine for entries and exits across all five trading engines.
GENETIC ENGINE
Treats each strategy as an organism: its DNA is rules — moving-average crosses, RSI, MACD, higher-timeframe trend, minimum volatility, trading hours. It generates thousands of variants, crosses them with each other, mutates a fraction at random, and repeats the process over hundreds of generations. The fittest survives, not the strongest.
The goal is not to find "the perfect strategy" once — it's to sustain an ecosystem that keeps generating, crossing and discarding variants as the market changes.
THE ROBUSTNESS GATE
No strategy reaches production just for having a good backtest result. Before becoming a slot "champion" candidate, it must survive six filters:
WALK-FORWARD
Out-of-sample validation, not just the optimisation period.
SELECTION BIAS
The edge of the best variant among N tested is adjusted downward — part of that result is an artefact of the search itself.
INDEPENDENT BETS
PCA over the set of active strategies to avoid a false sense of diversification.
ADVERSARIAL PERTURBATION
Simulated slippage more aggressive than the real one, to confirm the edge doesn't depend on perfect execution.
SANITY CEILINGS
Metrics outside a physically reasonable range are automatically discarded as a sign of a bug, not good luck.
FRACTIONAL KELLY
Position sizing by fractional Kelly, halved while a strategy hasn't yet accumulated enough live history.
Long and short can coexist as independent champions of the same slot — instead of forcing a single winner per asset and timeframe.
FOUR LAYERS OF CONTAINMENT
CIRCUIT BREAKER
Automatically cuts trading after consecutive losses or drawdown exceeding the defined threshold.
CROSS-ENGINE CORRELATION
Prevents stacking the same directional risk disguised as diversification across engines (US tech and the Nasdaq-100 via ETF count as the same bet, not two independent ones).
FUNDING-RATE GATE
In perpetuals, persistently unfavourable funding penalises or suspends the strategy.
HRP
Hierarchical Risk Parity — capital is distributed across active strategies weighted by risk, not by conviction.
Risk is never adjusted to hit a target or meet a deadline. Process discipline outweighs the calendar.
VERIFIABLE, NOT PROJECTED.
Paper trading — Paper trading — the system does not move to trading real money until it meets, all at once: a minimum of 30 closed trades, a minimum of 21 days of observation, win rate ≥55% and profit factor ≥1.5. The engine cleared all four gates before receiving the green light.
Real capital — Real capital — active since June 17, 2026. This is, deliberately, an early stage — days of real trading, not years — and is presented as such, without projecting that performance forward. Account and capital details are private.
Live PnL figures are not published here because this document is static and those figures change every day. The source of truth is always the public HUD — never a number frozen on a page.
WHERE TO FOLLOW THE ENGINE.
SIGMA's public presence — news, HUD and community — lives across these channels.
The full roadmap, with measurable milestones and their current status, lives on a dedicated page and is updated more often than this document.
FOLLOW THE ENGINE.
NOT THE LUCK.
Every trade SIGMA executes went through the same robustness gate described in this document — validation gates before a single dollar, not discretionary conviction. Follow the process live on the public HUD and across the community channels.
- —Futures and derivatives trading carries a real risk of total loss of invested capital.
- —Backtest and paper trading results do not guarantee future results on real capital.
- —SIGMA's real-capital track record is recent (days, not years, at the time this document was published) and must be evaluated with that context in mind.
- —This document is informational and does not constitute financial advice, an investment offer, or a guarantee of profitability.