Hierarchical clustering is a stronger extension of one of today's most\ninfluential unsupervised learning methods: clustering. The goal of this method\nis to create a hierarchy of clusters, thus constructing cluster evolutionary\nhistory and simultaneously finding clusterings at all resolutions. We propose\nfour traits of interest for hierarchical clustering algorithms: (1) empirical\nperformance, (2) theoretical guarantees, (3) cluster balance, and (4)\nscalability. While a number of algorithms are designed to achieve one to two of\nthese traits at a time, there exist none that achieve all four.\n Inspired by Bateni et al.'s scalable and empirically successful Affinity\nClustering [NeurIPs 2017], we introduce Affinity Clustering's successor,\nMatching Affinity Clustering. Like its predecessor, Matching Affinity\nClustering maintains strong empirical performance and uses Massively Parallel\nCommunication as its distributed model. Designed to maintain provably balanced\nclusters, we show that our algorithm achieves good, constant factor\napproximations for Moseley and Wang's revenue and Cohen-Addad et al.'s value.\nWe show Affinity Clustering cannot approximate either function. Along the way,\nwe also introduce an efficient $k$-sized maximum matching algorithm in the MPC\nmodel.\n