– Calibrated on your data. Brussle fits a correction to the decision’s number. Brussle applies the correction only when it beats the engine’s own numbers on outcomes that the fit did not see. In one customer’s own tests, the correction cut log loss on held-out outcomes by 55% on contract decisions and by 69% on vehicle-complaint decisions, while which records ranked above which barely moved.
– Run again when data changes. Brussle runs an answer again when data it read changes, and runs no other. It lets a burst of changes settle first, skips the engine call when an input has not changed, and caps the cost with a budget.
– Fitted again when the model changes. Brussle starts a new labelled epoch when the engine’s behaviour changes, then replays the inputs that your outcomes measured and fits the correction again for the new epoch.
