A controlled refinement loop
Optimization is a research decision, not a button
The objective is not to chase the best-looking historical output. It is to make a clear comparison between named versions under documented conditions. The Strategic Edge AI can organize that sequence from evidence to a proposed next test; you independently review the rules, platform behavior, settings, and risk.
01 · Freeze the baseline
Preserve the version you are comparing against
Record the exact entry and exit rules, market and timeframe, data window, sizing, session settings, commissions, slippage, and any relevant platform assumptions. A result without its settings is not a reliable baseline.
02 · Diagnose before changing
Describe the observed weakness in context
Ask whether the behavior may come from a market condition, a concentration of trades, an implementation mismatch, or an assumption in the test. Do not treat a single headline metric as a diagnosis.
03 · State one hypothesis
Make one defined rule change at a time
Write the proposed change as a falsifiable hypothesis. Keep the rest of the strategy and the comparison plan fixed so a later difference has a clear starting explanation.
04 · Compare evidence, not a promise
Review the trade-offs under stated assumptions
Compare profit factor, maximum drawdown, win rate, trade count, and Sharpe or Sortino where relevant. These describe historical behavior for a fixed test; none predicts future results.
05 · Check robustness
Look beyond the same favorable slice of history
Review the version across relevant conditions such as volatility shifts, false breakouts, bullish, bearish, or sideways periods. Use holdout, forward, or simulation work where appropriate before drawing stronger conclusions.
06 · Decide the next experiment
Retain, reject, or investigate the change
Document why the change is retained, rejected, or needs another test. The next action should be a controlled experiment, not an automatic promotion to live use.
An auditable record
Keep a visible version log
A version log separates a deliberate experiment from a vague claim that a strategy was “improved.” The simple example below is a research record, not a performance result.
| Version | Purpose | Defined change | Decision rule |
|---|---|---|---|
| V1 | Baseline | No rule change; rules and test assumptions are recorded. | Keep as the comparison point. |
| V2 | One hypothesis | Add one explicit condition while all other recorded settings remain fixed. | Compare the stated evidence and trade-offs. |
| V3 | Follow-up only if needed | Test a new, documented hypothesis from the retained baseline. | Retain, reject, or investigate—never auto-promote. |
Do not reverse-engineer a result. If multiple rules, data choices, or test settings change at once, the comparison may not show what caused the observed difference.
From backtest to next test
Use the evidence boundary to improve the question
Keep code and rules connected
A documented specification makes it easier to verify that the intended single change appears in the NinjaScript you independently test—or the Pine Script when you deliberately choose the secondary TradingView workflow.
Accept trade-offs explicitly
A lower drawdown, different trade count, or changed metric can introduce a trade-off. Record it before deciding whether a version deserves another test.
Continue with the evidence you already have
Start with explicit rules and a documented test, then use the result to frame the next controlled question—not an automatic optimization or a performance promise.
Educational strategy-development content only. The Strategic Edge AI does not provide signals, brokerage, automated execution, or financial advice.