Read the explanation
The built-in synthetic sample contains two true positives, one false negative, three false positives and four true negatives. False positive rate is three over seven, about forty two point eight six percent, while false negative rate is one over three, about thirty three point three three percent. The bars use eight pixels per percentage point. These sample labels are retrospective fixtures, not validation of any named model safety claim. With false positive cost ten and false negative cost twenty, observed gate cost is three times ten plus one times twenty, equal fifty. Always restrict costs seven benign cases times ten, equal seventy. Always allow costs three harmful cases times twenty, equal sixty. The two bars show observed gate fifty and always restrict seventy at six pixels per cost unit. Costs are analyst units and not observed monetary losses. Holding the sample fixed but changing false negative cost to five makes observed gate cost thirty five and always allow cost fifteen. The bars use twelve pixels per cost unit. The source ranks these totals; its modal heading always says the gate beats both blankets even when this new input contradicts that heading. Read the calculated totals and exported audit rather than treating the fixed modal title as an unconditional conclusion.