Does it actually work?

Short answer

Vektor is built and has been run end to end on real market data. It has not yet executed in a live account.

The eleven-step workflow, from universe screening to funded, trade-ready positions, was run on SGX, LSE and NASDAQ mandates in one session on 21 May 2026, with every action logged.

On the reference SGX mandate, 98.94% of available capital was deployed after board-lot rounding.

No capital moves without human approval.

What has not been shown is live performance. There is no live track record, and the reference results are backtests.

What we can show

• Eight exchanges configured. New markets take minutes, not projects. Different market. Same workflow. Same evidence.
• SGX: SGD 469,948 of SGD 475,000 available capital deployed (98.94%), with 100-share board lots and five active positions. NASDAQ: USD 9,497.04 of USD 9,500 deployed (99.97%), with single-share lots and four active positions. The platform applies the correct lot-size rule per exchange with no developer involvement and no deployment.
• The optimisation is a deterministic max-Sharpe solve, not sampled. Signals are chosen per instrument by grid search across SMA, EMA, MACD, Bollinger Bands, RSI and the Stochastic Oscillator, ranked by Sharpe, then validated by an XGBoost model. The two layers draw on different evidence classes.
• The portfolio manager reviews the strategy and the allocation before orders are generated. Every user action is timestamped, attributed and immutable, and every order is traceable from generation to execution.
• Rebalancing, working today: drift monitoring, 5% threshold detection, corrective-order calculation, minimum-trade filtering and dry-run review.

What this means

Working means built and run end to end on real market data. Live means running in a client account: not yet.

A large backtest Sharpe ratio is where the questions begin, not where they end.

Belief can’t be tested. Evidence can.

What remains unproven

• Live execution: Interactive Brokers connectivity is validated on paper trading. Real-money execution begins with the first Proof Partner’s first production period.
• There is no live track record. The reference output’s expected return and Sharpe ratio are backtest figures, and the document says they should be discounted. The runs used a public data source, not a professional data feed, which production requires.
• The NASDAQ and LSE runs were small mandates, USD 10,000 and GBP 10,000. The LSE result, 92.02%, is lower, explained by mandate size and concentrated weights.
• Signals: bounded ranges, Sharpe-based selection and daily re-optimisation reduce overfitting risk materially but do not eliminate it. Re-optimisation gives adaptiveness, not formal out-of-sample validation, and a three-year window is not guaranteed to be stable. The news sentiment layer is designed, not operational.
• Transaction cost modelling is not embedded in the optimisation engine.
• Not built: calendar-based and corridor rebalancing, unattended scheduled execution and tax-lot-aware order generation. Valuation is mark-to-market for listed equities only, on weighted-average cost basis.