We identify the qutrit as the minimal Hilbert-space dimension in which coherence amountand coherence sharing become operationally distinguishable. The starting point is theexact purity decomposition into population predictability, Hilbert–Schmidt coherence, andmixedness. Although this closure is a state-space identity, in dimension three it selects aEuclidean coherence coordinate that can be compared nontrivially with normalized ℓ1coherence. Their mismatch defines a positive deficit Δℓ1 that vanishes if and only if thethree pairwise coherence moduli are equal. The deficit therefore diagnosescoherence-sharing anisotropy rather than total coherence magnitude alone. We show thatΔℓ1 separates two physically distinct mechanisms of coherence reduction: populationanisotropy in symmetric channels and branch asymmetry in balanced qutrits. In a purifiedqutrit–meter model the same Euclidean language acquires a dynamical interpretationbecause reduced mixedness equals normalized linear-entropy entanglement across thesystem–meter cut. We clarify the resource-theoretic status of the diagnostic by showingthat it is not a coherence monotone, but a witness of unequal sharing among pairwisechannels. Finally, we give a reduced-state reconstruction protocol based on populationsand three embedded SU(2) Ramsey blocks, supported by finite-sample simulations. Thesame quantity has a direct three-path interferometric implementation in terms of pairwisefringe visibilities and calibrated path intensities, where it measures anisotropicmutual-coherence sharing rather than a universal violation of classical coherence theory.