Systematic Trading
How NOVAQ turns research into diversified, controlled execution. This page explains principles and capabilities, while trading rules stay private.
A diversified book
On 3 November 2026 NOVAQ ran 161 trading legs: 157 in the main platform's registry and four native strategies in the trading platform. Research shows about 12 effective exposures inside the book, and risk is organised around them.
| Family | Principle | Backtest profile, 2020–2026 |
|---|---|---|
| Opening-range and breakout | Continuation after a defined range gives way | Sharpe 1.71, return to drawdown 19.2 |
| Mean reversion | Return to value after rejection or stretch | Sharpe 2.25, return to drawdown 14.0 |
| Momentum | Continuation after a change in short-term pressure | Sharpe 1.47, return to drawdown 12.2 |
| Trend | Holding a directional view over a longer horizon | Sharpe 1.13, return to drawdown 8.5 |
| Fade and retest | Rejection and reclaim of reference levels | Part of the wider book |
| Native ultra-short strategies | Very short holding periods, run inside the trading platform | Not included in this table |
| Neural research books | Model-based strategies in simulation | Research only, no figures published |
Family figures are equal-weight historical tests of selected strategies after costs, and each family was positive in all 7 calendar years (2026 partial). Read them with the notes on the Evidence page. Mean-reversion live populations differ from the backtest.
Daily correlation between strategy families
Portfolio construction
The number of independent bets matters more than the number of strategies. NOVAQ groups books by what actually drives them, sets risk by bet and tests combinations on shared calendar days. Combining families with non-overlapping drawdowns lifted return to drawdown above that of the best single family in the backtest (25.5 against 20.0).
Risk control
| Control | Purpose |
|---|---|
| Safety gate | Blocks conflicting exposure before an order goes out |
| Kill switches | Stop new risk at account or system level |
| Crisis and dead-man protection | React to abnormal conditions or loss of supervision |
| Daily limits | Stop trading after a defined loss |
| News blackout | Halts new risk around scheduled major events |
| End-of-day flat | Closes positions before the session ends |
| Pre-trade budget checks | Limit exposure per account |
| Governor scaling | Reduces risk for strategies losing statistical support |
| Agent gate | Keeps AI agents inside the same limits |
| Governed commands | Record deployment changes through the platform as audited actions |
Native platform strategies
Some strategies hold positions too briefly for a polling engine. NOVAQ runs them natively in the trading platform from a C# core. Code parity with the research study was checked across 109,583 15-second bars, with all sampled decisions and trades matching. This shows that the code matches the study. It is not evidence of profit, and live results have lagged the study.
Three environments
| Environment | What it shows | What it cannot show |
|---|---|---|
| Backtest | How rules would have behaved historically | Real fills |
| Simulation | Whether the live system generates the expected signals | The full reality of broker fills |
| Live | What actually happens | Statistical certainty over a short period |
At the last check in late October 2026, the strategy book in the main platform traded on simulation accounts, prop firm simulation accounts and, for a short period, a live NinjaTrader account.
Strategy results on simulation accounts
NOVAQ strategy trades on 27 simulation accounts from 10 July to 3 November 2026. Entries are trades that were actually executed, and exits follow the strategy rules. For the 10.5% of trades that ended outside the strategy rules, the rule-based exit was computed from second-level data.
| Trades | Win rate | Profit factor | Payoff | Per trade | Result | Max DD |
|---|---|---|---|---|---|---|
| 7,380 | 42.2% | 1.30 | 1.76 | +$44 | +$323,003 | −$105,805 |
| Month | Result | Trades | Win rate |
|---|---|---|---|
| July | +$12,438 | 881 | 42% |
| August | +$13,961 | 1,357 | 39% |
| September | +$42,203 | 2,043 | 42% |
| October | +$175,033 | 2,878 | 44% |
| November (to 3 Nov) | +$79,368 | 221 | 49% |
This is simulation, not real capital. Simulation fills orders more favourably than the market, and computed exits assume a fill exactly at the stop or target level. Position size is set by the system and differs between accounts. October delivered most of the result.
Fidelity and reconciliation
The same logic runs in backtest and live wherever possible, supported by three checks.
- Bit audits compare signals and order fields from deployed code with the original study.
- Paired-trade audits compare forward trades with their backtest counterparts. No strategy family showed a robust gap on 2,272 pairs, mostly simulation. This is a technical tracking check.
- Broker truth rebuilds missing fill prices from broker account snapshots. It repairs only unambiguous cases and refuses the rest with a recorded reason.