Where the tree of life is empirically resolved, and where it is not: an open atlas of species-level phylogenies and their archival uncertainty

Comparative and macroevolutionary biology depends on knowing two things at once: where the species-level tree of life is empirically resolved, and where it is not. Yet the phylogenies that anchor downstream inference are routinely treated as fixed inputs, even though their empirical substrate is deeply heterogeneous in ways that vanish once a tree is downloaded. Trees differ in inference framework, in dating regime, in the balance between species supported by direct molecular data and species inserted by taxonomic constraint, and in whether per-node uncertainty estimated during inference survives the path from primary analysis to deposited derivative. These choices accumulate silently into cross-clade analyses, and no existing resource maps the current state of phylogenetic knowledge uniformly across Bacteria, Archaea, and Eukaryota. We offer a starting point: a first systematic inventory of what is, and is not, empirically resolved. We assemble an atlas of 264 species-level phylogenies, mapped onto a single global species dictionary of roughly 637,000 labels and organised into 62 partitions of described biodiversity, with 36 further partitions explicitly retained as unrepresented so that phylogenetic dark matter remains visible at the same resolution as the represented clades. For every canonical tree, a per-tree provenance ledger records inference framework, dating regime, the fraction of tips anchored in molecules rather than in taxonomy, and an audit of how much inferred uncertainty survived deposition; an explicit sensitivity envelope of inclusion criteria, from permissive to strictest, brackets every coverage statement against its own auditable rule. Because knowledge of the tree of life is itself moving, the atlas is released as a versioned, continuously updated resource through the companion R package phyloatlas and a per-partition wiki, so that downstream comparative analyses can propagate provenance and inclusion-criterion uncertainty into their own inferences rather than silently committing to one set of choices.