Best Score and Stable determine how one configuration is selected from the valid in-sample candidates in each Walkforward segment.
Both use the same Walkforward score, but Stable first narrows the candidate group.
The Walkforward score
The score compares Net P&L with Max Drawdown.
When Max Drawdown is 0, the score uses Net P&L.
This expresses the amount of Net P&L relative to the drawdown recorded for the candidate.
For example:
Score example
- Net P&L: 1,000
- Max Drawdown magnitude: 500
- Score: 1,000 / 500 = 2.0
A higher score represents more Net P&L relative to the drawdown used in the calculation.
Best Score
Best Score applies the Walkforward score to all valid in-sample candidates.
The configuration with the highest score is selected for that segment.
The process is:
- evaluate all valid in-sample combinations
- calculate the Walkforward score for each combination
- select the highest-scoring configuration
- apply that configuration to the out-of-sample period
Stable
Stable narrows the candidate group before applying the same Walkforward score.
The process is:
- rank all valid in-sample candidates by Net P&L
- keep the configured Top Combinations
- calculate or compare the Walkforward score within that retained group
- select the highest-scoring configuration
- apply that configuration to the out-of-sample period
Top Combinations
Top Combinations controls how many of the highest Net P&L candidates remain before Stable makes the final score-based selection.
Top Combinations = 20
- The 20 highest Net P&L candidates remain for the final score comparison.
Best Score and Stable compared
Best Score
- All valid candidates
- Compare Walkforward scores
- Highest score selected
Stable
- All valid candidates
- Rank by Net P&L
- Keep Top Combinations
- Compare Walkforward scores
- Highest score selected
The selection is performed again for every eligible Walkforward segment.
The selected configuration may therefore change as the in-sample period changes.
Where to go next
Continue with Configuring and running Walkforward to see where these choices appear in the research setup.