Why is the torsion-free condition (rather than allowing torsion) the natural choice for Riemannian geometry?
Think about your answer, then reveal below.
Model answer: The torsion-free condition means ∇_X Y - ∇_Y X = [X,Y], so the antisymmetric part of the connection carries no independent information beyond the Lie bracket. This ensures that geodesics depend only on the metric (not on extra torsion data), that the Christoffel symbols are symmetric, and that the connection is fully determined by the metric. Allowing torsion would introduce additional geometric degrees of freedom that the metric alone does not control. The uniqueness of the Levi-Civita connection — that the metric determines the connection — is only possible with the torsion-free condition.
In physics, torsion appears in Einstein-Cartan theory (where spinning matter sources torsion) and in string theory. But in pure Riemannian geometry, the torsion-free condition is universally adopted because it gives the most economical theory: one piece of input data (the metric) determines everything. If you allowed torsion, you would need to specify the metric AND the torsion tensor independently.