Metric battlecards for vision reviews
Recall which computer-vision metric to use when stakeholders challenge model readiness.
Accuracy trap "The model is 96 percent accurate, so why are we still reviewing it?" Defects are rare and missed defects are expensive. Your line Accuracy is a headline; for rare defects we need recall on true defects and false alerts per shift before reducing review. Letting the biggest number become the readiness decision. It exposes class imbalance and ties the metric to operational cost. Precision pressure "Every alert should be right, or people will ignore it." The use case is safety screening. Alert trust matters, so let us report precision, but we should choose the threshold with recall visible…
Sign up free — one personalized lesson every day, matched to your role and goals.
Already have an account? Sign in