Discovery: publications and software around a topic

Most research work begins in a landscape someone else has already mapped in part. Discovery is the deliberate survey of that landscape before building, writing or claiming novelty: the publications that define the state of the field, and the software that already solves pieces of the problem. Doing it early is cheaper than discovering a competitor in review, and the survey itself becomes reusable material for the paper's related-work section and the proposal's state of the art.

Finding publications

Work from open, keyless APIs first; they make searches reproducible and scriptable:

• OpenAlex covers most of the scholarly record with abstracts, citation links and topic filters; Crossref serves DOI metadata; Semantic Scholar adds influence signals and TLDRs; arXiv covers preprints in its fields. Zenodo and JOSS matter specifically for research software: JOSS papers ARE software descriptions, and Zenodo records often carry the software itself.
• Google Scholar casts the widest net (theses, reports, gray literature) but has no API and rate-limits scraping; use it interactively for coverage checks, not in pipelines.
• Snowball in both directions from the two or three most relevant hits: who do they cite (backward), who cites them (forward - OpenAlex and Semantic Scholar both serve citing works). One good seed paper beats twenty keyword guesses.
• Search by concept variants, not one phrasing: field synonyms, British/American spellings, method names old and new - and record the queries used, so the search is repeatable and its gaps are visible (rseng-reproducibility's spirit applied to searching).

Finding software

Cast several nets; each finds things the others miss:

• Research software registries: Research Software Directory instances (rseng-software-reuse ships offline snapshots and the search discipline), domain registries (bio.tools for life sciences, ASCL for astronomy, and their field equivalents).
• Package indexes for the stack (PyPI, CRAN, conda-forge, npm): search by task words and by the names publications mention; libraries.io searches across ecosystems at once.
• Public forges: code search on GitHub/GitLab finds the unregistered majority - search READMEs by domain terms, filter by language and activity, and check the topics/tags of anything close.
• Curated awesome lists collect a community's known tools; find the list for the domain, then treat it as a lead generator - lists rot, so verify each candidate is alive.
• Software archives: Software Heritage preserves code whose forge home vanished; JOSS and Zenodo double as software discovery for peer-reviewed and archived tools.
• Papers with Code links publications to implementations in ML-adjacent fields.

From findings to decisions

Discovery feeds the pack's decision skills rather than replacing them:

• Candidate software worth adopting goes through the six-axis intake vetting (rseng-dependency-management); fit judgment and citation duty live with rseng-software-reuse.
• Competitor and alternative surveys become an honest comparison table: what each neighbor does, its stack, license, activity and how the user's project differs - the state-of-the-field material that software papers (rseng-software-peer-review), communication (rseng-science-communication) and documentation's related-projects section (rseng-documentation) all need. Differences stated honestly build more trust than silence about competitors.
• Everything found gets verified before it is repeated: resolve the DOI, open the repository, confirm the claim matches the source (rseng-fact-checking, rseng-citation-hygiene) - discovery collects leads, verification makes them citable.
• Record the survey: queries, sources searched, date, and the shortlist with reasons, in the project record (rseng-project-tracking) - a documented search can be updated; an undocumented one gets redone from scratch.

Working with this skill

This skill is source-independent: its authority is the discovery services linked below. It is the front end of a chain: discovery finds, rseng-software-reuse and rseng-dependency-management judge, rseng-fact-checking and rseng-citation-hygiene verify.

Learn more (verified):

• https://help.openalex.org - OpenAlex API
• https://arxiv.org - arXiv
• https://www.semanticscholar.org - Semantic Scholar
• https://scholar.google.com - Google Scholar (interactive)
• https://zenodo.org - Zenodo
• https://joss.theoj.org - JOSS
• https://bio.tools - bio.tools registry
• https://libraries.io - cross-ecosystem package search
• https://github.com/sindresorhus/awesome - the awesome-list index
• https://huggingface.co/papers - Hugging Face Papers

Related skills

Check whether any of these applies before moving on:

• rseng-citation-hygiene - verifying surveyed references
• rseng-dependency-management - vetting discovered software
• rseng-fact-checking - verifying claims before repeating them
• rseng-science-communication - related-work narrative for audiences
• rseng-software-peer-review - state-of-field for JOSS paper
• rseng-software-reuse - candidate fit and citation duty