Reconstructing Journal Thematic Identity from Article-Level Topics: A Reproducible Framework and a Preregistered Temporal Validation

Abstract

Journals are conventionally described by editorial subject classifications, yet what a journal actually publishes need not coincide with its assigned label. We present a framework for reconstructing a journal's thematic identity from article-level evidence while maintaining a strict separation between what is observed and what is inferred. From the distribution of OpenAlex subfields across a journal's articles, we define an observed thematic representation and, from it, three preregistered comparative descriptors: alignment with the editorial classification, intrinsic coverage of the thematic nucleus, and relative position, which yield an operational two-dimensional typology. Over a transparently constructed cohort of journals, we characterize thematic concentration, temporal stability, and the discriminant structure of the descriptors. We then submit the framework to a preregistered prospective replication that shifts the observation window forward in time, and to a documentary control that recomputes the same analysis on document- rather than citation-weighted distributions. All thresholds were fixed in advance, and code, data, preregistrations, and a code-to-documentation audit are released with cryptographic version identifiers. This is a preprint, prepared for submission to Quantitative Science Studies; it has not yet been peer reviewed. This deposit is the reproducible research package accompanying the manuscript, including canonical analysis code, data manifests, preregistration documents, and a code-to-documentation audit, with SHA-256 version identifiers linking every reported quantity to the implementation that produced it. The manuscript, documentation, and data are licensed under CC BY 4.0. Source code contained in the code/ directory is licensed separately under the MIT License