Document Major Construction of BM25 Postings in leCore

Sparse retrieval implementations can spend substantial indexing time examining term and document combinations that never occur. This paper evaluates a document-major postings construction change in leCore's BM25 implementation. The change traverses observed terms within each document instead of probing every document for every vocabulary term. A preserved offline experiment compared four exact source revisions on 808 overlapping chunks of a pinned Rust repository, with ten source questions and fixed retrieval parameters. The isolated document-major revision reduced median full index construction time from 915.70 to 217.11 milliseconds across five recorded fits per version, a ratio of medians of 4.22. It produced bit-identical postings and query score arrays to the baseline on the recorded corpus and questions. Full evidence coverage within the top three chunks remained eight of ten questions. A separate query-term-frequency correction and a later upstream snapshot are reported independently to avoid attributing their behavior to the traversal change. The result supports a specific implementation optimization on one machine and corpus. It does not establish a new retrieval algorithm, a retrieval-quality improvement, or a general speedup across datasets and systems. Preparation of this report was assisted by Codex. The research programme is directed toward human agency, financial inclusion, and evidence-based accountability. Intended social benefits are distinguished from measured technical results in the manuscript. The manuscript and newly authored analysis are offered under CC BY 4.0. Any supplied source files retain the license terms stated in the supplement; the record license does not replace third-party source licenses.