What is the Stability Score in darwintIQ?
A model that looks good on average can still be hiding something. The Stability Score finds it.
The trading model stability score measures something that aggregate metrics miss: whether a model's returns are spread consistently across an evaluation period, or concentrated in a narrow window that may not persist.
A model that produces most of its returns in a single favourable stretch looks acceptable — even strong — when you examine its Sharpe Ratio or Profit Factor alone. The Stability Score is specifically designed to surface this kind of concentrated performance and flag it as the fragility it represents.
What consistency means in a trading context
Aggregate performance metrics summarise results across all trades in a period. They are useful, but they cannot tell you whether performance has been distributed reasonably throughout that period or generated by a small number of outlier events.
Consider two models with the same average return, the same win rate, and the same drawdown over a 30-day window. One produces results spread relatively evenly across the month — some better days, some worse, but consistently contributing. The other produces a strong first week, then flat to slightly negative behaviour for the remaining three weeks, ending with a similar aggregate number.
Both look identical in aggregate. But the second model's performance pattern should raise questions: is the first-week surge repeatable, or was it a specific short-lived condition?
The Stability Score distinguishes between these two models. It captures the temporal distribution of returns — not just whether returns exist, but how evenly they are generated across the evaluation period.
How the Stability Score relates to other metrics in darwintIQ
The Stability Score works alongside the Robustness Score and the Sortino Ratio to build a multi-dimensional picture of model quality.
The Robustness Score focuses on whether the model's performance holds up when its parameters are slightly varied, and whether it is likely performing on genuine market logic rather than a narrow fit to historical data.
The Sortino Ratio describes the relationship between return and downside volatility — how much the model earns relative to the losses it produces on the downside.
The Stability Score captures something neither of these addresses directly: the evenness of the return stream over time. A model can have high robustness and a strong Sortino Ratio, and still exhibit poor stability if its strong risk-adjusted periods are concentrated in a short window. All three metrics contribute distinct information, and together they describe a quality profile that any one alone cannot.
Reading the Stability Score in practice
A high Stability Score indicates that the model's trading results are distributed throughout the evaluation window — the model is contributing at roughly consistent levels rather than in concentrated bursts.
A low Stability Score is a warning sign even when other headline metrics appear strong. It suggests that the model's results depend on conditions that only prevailed briefly, or on a small number of trades that happened to be unusually favourable. That concentration makes it harder to be confident the performance will continue.
When assessing models in darwintIQ, the Stability Score is most useful as a filter after the initial screening. Once you have identified models with acceptable returns, drawdown, and Profit Factor, the Stability Score helps distinguish those with genuine consistency from those whose numbers rest on concentrated episodes.
A model showing strong headline metrics but a weak Stability Score deserves closer inspection before being prioritised over one with slightly lower returns but reliable, distributed performance.
Why the rolling window makes stability particularly relevant
darwintIQ evaluates models on a rolling 4-hour window. This means the Stability Score is always calculated on recent behaviour — which is precisely what makes it timely and meaningful.
A high Stability Score on the current window reflects consistent behaviour under the conditions that currently exist in the market. It is not a historical average from months ago; it is evidence that the model is performing steadily right now, not just in aggregate across a period that has already passed.
Models that maintain strong Stability Scores across consecutive evaluation windows — rather than occasionally spiking and then dropping — demonstrate the kind of structural consistency the Genetic Algorithm selects for over time. Consistency that persists through changing conditions is considerably more valuable than a single-period episode, however impressive it appears.
Final thoughts
The Stability Score is not a shortcut to model selection — it is a lens that reveals what aggregate metrics can hide. Used alongside returns, drawdown, and the Robustness Score, it is one of the sharpest tools available for distinguishing models that are genuinely performing consistently from those that merely look good in summary. When the distribution of performance matters as much as its size, the Stability Score is where to look.