Classical Nucleation Theory (CNT) predicts nucleation rates in hard-sphere colloids that differ from experimental measurements by up to 22 orders of magnitude—the largest such discrepancy in condensed matter physics. Recent experiments (2025) have deepened the crisis, finding that the critical nucleus size remains constant independent of supersaturation, directly contradicting CNT predictions. We propose that this discrepancy has a geometric origin: CNT assumes spherical nuclei, but Frank demonstrated in 1952 that supercooled liquids exhibit icosahedral local order. Icosahedra cannot tile three-dimensional space, creating geometric frustration that CNT does not capture. Drawing on icosahedral geometry, we derive a parameter-free prediction: the nucleation barrier ratio ΔG*/Q* = e⁵/1640 ≈ 9%, where e⁵ is the icosahedral structure ratio. We show that the 22 orders of magnitude discrepancy arises from CNT missing this e⁵ factor at approximately four stages of the nucleation process, giving (e⁵)⁴ ≈ 10⁸–10⁹ as the core geometric error.