Replication package (anonymised for peer review): Compositional Integration: Anthropology's Four Fields on One Shelf, 1888-2025

This version (2026-09-09) accompanies a retitled and rewritten manuscript. The analysis is reorganised around a decomposition of cross-boundary citation into a compositional component, supplied by what a discipline's shared journals publish, and an elective component, obtained by balancing each decade's subfield-by-subfield citation matrix to equal margins. Three defects in the previous version are corrected: communities smaller than 100 papers were pooled into a single residual label that then behaved like the largest module; the evolutionary-biology and molecular-biology comparison corpora were superseded on disk and have been recomputed; and the claim that anthropology exports nothing to the fields it imports from is withdrawn, having been measured and found false. New scripts: 71-79 (concept tests and the decomposition), 82-86 (tables, figures, supplement and title-page word count), 98/99 (the machine audit and its configuration). Code, derived tables and figures for a distant-reading study of four-field anthropology. The corpus is 86,436 research articles from 36 English-language anthropology journals (1876-2025), fused from OpenAlex, Crossref and Semantic Scholar, spanning sociocultural, biological, archaeological and linguistic anthropology in two tiers (four-field/generalist journals and subfield flagships). Contents: the full pipeline (collection, multi-source fusion, review/paratext filtering, SPECTER2 embedding, Leiden community detection, CD index, Uzzi novelty, BERTopic lifecycle, canon analysis); a four-field text classifier validated against held-out outlet subfield; a continuous science-interpretation classifier with dip/GMM bimodality tests; the cleaned corpus; per-paper subfield, orientation and community assignments; all derived tables and figures; and the comparison-panel audit (outputs/panel_audit.json) documenting why philosophy of science, history and literary studies are excluded from the panel. Comparison corpora (sociology, psychology, economics, philosophy, evolutionary biology, molecular biology, physics, mathematics) were built with the identical pipeline; their derived tables are included, their raw corpora are not redistributed. Note: author identity and the project plan are withheld during peer review; this version contains the anonymised manuscript. The record will be re-issued with full metadata on acceptance.