A knowledge graph requires a stable address for its evidence even when the literature offers competing classifications with no consensus. The address chosen for retrieval often passes as the only available classification, which occurs when records gathered under one name refer to units that differ in boundaries, rank, or counting criteria. In Sinitic dialectology the same localities appear in schemes whose units vary in membership and rank; a unit may change affiliation when its diagnostic evidence changes. Two tone counts, each quoted faithfully, may still count different objects, causing the field’s taxonomy to rest on a contradictory literature frame. Verifying a quotation or searching for conflicting values does not reveal these differences.
I separate the working address that gathers candidate evidence, the category defined by the source, and the decision that admits a passage to a particular comparison, arguing that these decisions together with inter‑record dependencies belong to the provenance of a comparison. On four reproducible cases, the PaRAGraph Casebook shows where such differences arise among claims extracted by a language model and verified before entering the graph. The platform now stores these decisions as objects for one dimension of comparability—the address of a record: the placement of a locality in each source’s scheme, the decision that a name in a given publication denotes a unit rather than a place, and the origin of every address. The working taxonomy can be revised as new sources yield new claims, while the path from each comparison to its evidence remains fixed.
Review