Whenever you're ready, walk me through how you'd approach this.
Given human labelers, labeling tasks, and a model, how would you create a balanced assignment schedule across (human, task), (human, model), and (model, task) pairs?
Explain how you would represent the assignment problem, define balance and quality objectives, generate eligible candidates, and schedule assignments under capacity and deadline constraints. Discuss how the design would adapt to uncertainty, annotator expertise, model disagreement, delayed labels, and changing task distributions. Include offline evaluation, online monitoring, retraining feedback, and safeguards against training-serving skew and feature drift.