Start with a market question
A condition should identify context, trigger an entry or exclude a situation. Choose that purpose first, then a measure that expresses it.
Several similar indicators can create an impression of confirmation without supplying different information. Compare versions with and without a filter to investigate its contribution.
Understand condition families
| Family | Typical question | Limitation to examine |
|---|---|---|
| Trend | Is price following a defined direction? | Delay and sideways phases |
| Momentum | Is movement strengthening or weakening? | Extremes that persist |
| Volatility | Is the amplitude suitable for the hypothesis? | Regime changes |
| Volume and levels | What activity or area is observed? | Volume type and availability |
| Structure | Which level or movement is confirmed? | Pivot definition and confirmation time |
| Sessions | When are entries allowed? | Timezone and seasonal clock changes |
Review the parameters
Check the price source, period, threshold and unit. “Above” describes a state; “crosses above” describes a change of state. Substituting one for the other can substantially change the number of signals.
For several timeframes, define which horizon supplies context and which supplies the entry. Ask when information is considered confirmed. An observation still forming can change before its period ends.
Purpose: context filter, not a standalone entry trigger.
Measure: moving average on the selected horizon.
Buy condition: reference price is above that average.
Parameters: average period and horizon.
Decision required: completed or current candle observation.Define a structure or zone
Terms such as breakout, sweep, imbalance and structure change require a definition for your strategy. Describe the reference level, distance, confirmation and signal expiry.
A level selected visually after a move might not have been identifiable when the decision would have occurred. Check that the necessary information was available before the entry being studied.
Help define a level rejection unambiguously.
Explain level selection, what counts as contact, the required confirmation
and how long the signal remains valid. Include a counterexample to reject.
Do not present the definition as evidence of profitability.Check data requirements
A volume-dependent rule needs volume information suitable for its meaning. A rule sensitive to the exact sequence of prices may require more detail than a candle supplies. Review the test’s data information and warnings.
If the condition is unavailable, describe it in Flow rather than choosing a similarly named card that measures something else.
Evaluate a filter’s contribution
Keep data and costs comparable. Read trade count, net result, drawdown and the trades removed by the filter. Reducing losses can also remove much of the profit.
A smoother result with very few trades needs further checking. See Compare in AI Lab for experiment design and Understand metrics for interpreting the trade-offs.