1. Freeze the research specification
Before studying results, record the instrument, venue, timeframe, date range, initial virtual capital and strategy settings. State the entry and exit rules in terms that another person could reproduce. A change in a threshold or sizing assumption creates a different study, even if the strategy keeps the same name.
Keep a simple hypothesis, such as comparing one defined EMA/RSI research rule with a baseline on the same eligible data. This is an example of study design, not a recommendation for those indicators. Record unsuccessful variants too; keeping only the best-looking run obscures how many choices were tried.
2. Check what was known at each decision
A causal simulation uses only observations available at the historical decision time. In the AIONEX methodology, prior-candle signals and next-open simulated execution prevent a completed candle's later information from being used for an earlier decision. Indicator warm-up and missing-candle handling must be explicit.
Lookahead bias means future information has leaked into the past. TradingView's strategy documentation describes this problem in historical simulations. Ask whether higher-timeframe values were complete and whether the simulator assumed a path within a candle that the dataset cannot prove. A smooth equity curve cannot settle these questions.
3. Make costs and fills inspectable
Record fees, slippage, spread assumptions and position size in the correct units. If the modeled instrument requires additional costs or constraints, identify whether the study actually includes them. A spot study cannot stand in for a leveraged futures study with different margin, funding and liquidation behavior.
For a deliberately simplified hypothetical example, suppose the only modeled costs are a 0.10% fee and 0.05% slippage on each side. The combined entry-and-exit cost is approximately 0.30% before compounding and other costs. These invented inputs are not current venue prices. A small gross advantage could disappear under another cost scenario.
AIONEX distinguishes supplied cost assumptions from proof that those assumptions match real fills. Keep that distinction when comparing reports.
4. Separate tuning from evaluation
Arrange observations chronologically. Use an earlier training window for a bounded parameter comparison and a later untouched holdout window for evaluation. The holdout is data withheld from the tuning decision. If you inspect it and tune again, it no longer serves as an untouched test for that decision.
AIONEX's documented Auto-Tuner compares a bounded deterministic grid and reports the holdout separately. Read both windows and their trade counts. A selected winner can be a historical coincidence; neither tuning nor a holdout result establishes a calibrated runtime probability or validates every production output.
5. Read beyond total return
Review drawdown, trade count, turnover and the distribution of gains and losses alongside return. Drawdown describes a fall from a prior equity peak; a win rate describes the fraction of winning trades. A high win rate can coexist with a few large losses. Ratios can be unstable when history is short or observations are sparse.
Compare only reports with compatible units, dates, accounting and cost assumptions. AIONEX's independent research benchmark is not a production-equivalent replay. Historical benchmark output must not be presented as verified accuracy of the current production signal.
A backtest review checklist
A report is most useful when someone can explain what changed between runs. Preserve the specification and the unavailable fields with the result.
- Dataset: source, instrument, dates, timeframe, gaps and warm-up.
- Rules: exact settings, position sizing and historical decision timing.
- Execution: simulated fill convention, fees, slippage and excluded costs.
- Evaluation: chronological split, tuning variants and untouched holdout.
- Outcome: return, drawdown, trade count and limitations in both windows.
What to do with an incomplete result
If required data or execution assumptions are missing, treat the report as unavailable or limited research rather than repairing it with optimistic defaults. The AIONEX workspace uses virtual capital for eligible studies and does not place orders. Its interface is currently English, and plan and data availability apply. Keep the study as a research record, with uncertainty attached.