Abstract
Choosing the technical content of a figure skating program requires making decisions about both elements and their order. I propose a framework that combines predictions from historical competition protocols with mixed-integer linear programming. For each skater and event, candidate elements are drawn from the preceding two years, and the main model selects a complete program satisfying the encoded International Skating Union composition rules. The analysis covers 13,876 complete programs from 2018 to 2024. In a holdout of 2,500 programs allowing changes in content and order, optimized layouts score 8.25 points higher on average when both layouts are evaluated with the main model. Positive estimated differences remain under a difficulty limit based on the skater's highest recent total base value, and element selection is stable under the tested changes in predicted scores. A dynamic extension allows for carryover after severe execution errors and accumulated load under supplied element load profiles. Calibration supports a small association between earlier errors and later execution, while the additional load penalty is zero in all eight examined profiles. This framework provides program alternatives for assessment by coaches and choreographers.