How does NumPy handle dtype promotion in chained arithmetic expressions?

NumPy promotes dtypes step-by-step using a fixed hierarchy during expression evaluation.

In chained expressions like int + float + bool, NumPy evaluates promotion rules at each ufunc stage. Each intermediate result may have a different dtype depending on operand combination. This ensures precision is preserved but can increase memory usage due to repeated upcasting.