Despite the deluge of data generated by over a decade of highly parallel sequencing, molecular approaches to date have failed to deliver a fully-resolved metazoan phylogeny in the face of strongly supported incongruence. Here, I attempt a partial explanation of this paradox, illustrating some outstanding dilemmas of the discipline with two empirical datasets, and outlining prospects for future momentum. Firstly, incorporating new genome assemblies spanning the extant diversity of Placozoa, I investigate pre-bilaterian relationships. I see strongly supported discordance among gene sets for the position of Placozoa even in the presence of site-heterogeneous mixture models, but show that genes that pass a sensitive compositional heterogeneity test favor a Cnidaria+Placozoa clade, suggesting the familiar position of Placozoa outside a Bilateria-Cnidaria clade is a compositional artifact. I also observe strongly supported discordance vis-à-vis the identity of the earliest-branching animal clade, depending on whether the matrix is encoded as amino acids or as Dayhoff groups. In a parallel analysis, I examine pan-metazoan relationships, with a large (165+ taxa) matrix comprising genome or transcriptome assemblies from all animal phyla (excluding Mesozoa). I discuss strategies for overcoming multiple sources of error at this scale, e.g. by considering the value of taxon-specific gene sets, and the detrimental impact of rogue or data-poor taxa on convergence in Bayesian mixture model analyses. Finally, I present a new technique for amplifying long-insert (3-10 kb) genomic libraries with little coverage bias via PCR from picogram-scale input DNA quantities. Combined with nanopore sequencing, we show the potential of this technique to fill in the remaining gaps in genomic resources across the animal tree of life, by enabling highly contiguous reference genome (and metagenome) assemblies derived from single meiofaunal animals.