Abstract
<title>Abstract</title> <p>aims institutional tags are distorted by consortium over-attribution and by author-profile merge/split errors, so naïve location and seniority assignments are biased. Using German-speaking (DACH) radiation oncology as a worked example (17,270 articles, 2000–2025), we present an openly reproducible pipeline that derives locations and seniority from the raw affiliation strings rather than from the disambiguated institution tags. Its core components are: a robust, checkpointed OpenAlex client; a union corpus combining a topic with eight core journals through a dual-stream OR; a stem-based department filter separating radiation oncology from excluded specialties; a curated, corpus-grounded city gazetteer with word-boundary matching; a single-string consortium-clause guard that removes phantom locations from network-grant (“DKTK”) strings; and a person-gated fallback that recovers genuine senior locations while excluding non-radiation-oncology applicants. We define three transparent location measures (strict, gated, presence), report the data-quality failure modes encountered (P1–P7) with their corrections, and emit a full machine-readable provenance manifest. On a manually coded random sample, the raw-string measures localise last authors far more accurately than a naïve institution-tag baseline (F1 ≈ 0.92 vs. 0.64). The pipeline is field-agnostic and transferable to any discipline-and-region slice of OpenAlex.</p>