domain-corpus-bootstrap

Start with a topic map decomposing the domain into core topic, sub-topics, adjacent fields, competing approaches, and foundational theory. Then write a checked-in query manifest, download open-access PDFs, triage into priority tiers, and write an analysis of themes, canon, and gaps.

NASA ADS is the primary index, supplemented by NASA NTRS, NIAC final reports, SPIE proceedings, and arXiv astro-ph. General indexes like Crossref and Semantic Scholar under-cover this field's grey literature, so ADS is not optional.

No. Acquisition is limited to URLs advertised as open access by Unpaywall, OpenAlex, arXiv, HAL, Europe PMC, CORE, or publisher OA links. Paywalled items go onto an institutional library request list or a browser-only list for manual retrieval.

Matching proceeds by normalized DOI, then DataCite arXiv DOIs, then arXiv ID, then normalized title only when no DOI exists. A preprint and its version of record become alternate versions on one entry, never two separate papers.

Each entry needs a full citation, resolvable DOI, ISSN for journals or ISBN plus volume for proceedings, and one line of internal justification quoting the claim that depends on it. Keep the list short — around 20 items with reasons gets filled, 300 items get ignored.

A gap analysis derived from search records is a shortlist for reading, not a recommendation — the reference implementation measured a 30% false-positive rate when recommending citations without checking the manuscript text. Claims must be grepped from the live source before any recommendation is written.