Evaluating systems with paper trading and walk-forward windows

When a historical backtest is unavailable or set aside, remaining live evaluation paths include a live real-money test, paper trading, and later review of the platform report for the window after the rules were frozen.

A live real-money test runs the coded procedure with actual orders so fills and liquidity effects come from the market rather than from a simulator. Funding many unproven rule sets this way can become expensive.

Paper trading confirms signal timing

Paper trading runs the same entry, exit, and abstention rules against a simulated book to confirm that signals fire when intended, without committing capital.

It is useful for that timing check, not for treating simulated account growth as a performance verdict. Simulated equity growth does not show that the same procedure would produce comparable results with real capital, and it does not reproduce the emotional effects of money at risk.

A frozen window is not the development sample

Walk-forward analysis in this setting means finishing and freezing the coded rules, then later inspecting only the subsequent unseen period in the engine report.

Freezing a finished rule set and scoring only that later window keeps development and evaluation from being the same sample.

Engine fidelity before any window is trusted

The frozen-window check requires engine fidelity: a platform report and live execution must produce the same decisions and comparable fills for the same rules, plus realistic slippage assumptions. Some styles, including scalping procedures, can mislead the engine.

Simulation engines can admit fill or data assumptions that would not occur in live markets, including executions at locked limit-up or limit-down prices. An unrealistic fill is a simulated trade that could not have been completed under live market constraints, such as a short entered during a locked limit-down session.

Robustness testing stresses fill quality, slippage, and whether the engine path and the live path agree before any evaluation window is treated as informative. Those checks also increase how much a later historical backtest can be trusted.