literature-review-general

name

literature-review-general

description

General-purpose rigorous literature review workflow for any research field or interdisciplinary topic. Use when Codex needs to perform early-stage scoping, mid-stage argument refinement, or late-stage paper-writing literature review with current source lookup, verified bibliographic metadata, field-appropriate databases and venues, DOI/author/title/venue/URL cross-checking, citation and quality constraints when provided, milestone contributions, latest progress, hot topics, unresolved problems, evidence grading, and explicit uncertainty marking.

Literature Review General

Core Rule

Never invent bibliographic facts, claims, citation counts, journal metrics, venue rankings, conclusions, or field narratives. If a source, metric, or claim cannot be verified from reliable evidence, mark it explicitly instead of filling the gap.

Use current web/database lookup for recent literature, citation counts, journal metrics, venue rankings, author affiliations, publication metadata, and any "latest", "hot topic", or "current open problem" request. Prefer primary publisher pages, DOI.org/Crossref, field-specific scholarly databases, citation indexes, official journal or conference pages, and official metric/ranking sources when available.

Workflow

1. Clarify the review stage: early scoping, mid-stage argumentation, or late-stage writing. If the user does not specify, default to early scoping and state that assumption.
2. Identify the field, subfield, neighboring disciplines, publication culture, and relevant evidence standards.
3. Define the search space: topic terms, theories, methods, data types, populations/systems, time range, languages, geography, and exclusion criteria.
4. Build a query map across at least three angles when relevant: concepts/theory, methods/evidence, application/context, and debates/gaps.
5. Select field-appropriate databases, publishers, journals, conferences, preprint servers, and citation sources using references/search-and-validation.md.
6. Search iteratively, starting with authoritative field-specific venues and reviews, then broadening as needed. Log candidate sources with provenance.
7. Apply the user's inclusion constraints. If the user gives no constraints, use transparent relevance, source quality, citation influence, recency, and field norms instead of pretending a universal threshold exists.
8. Read every included source's abstract, summary, or equivalent. When full text is accessible, inspect conclusion/discussion sections before making substantive claims.
9. Cross-validate DOI or persistent identifier, authors, title, venue, volume/issue/pages or proceedings details, year, URL, and citation/metric source where applicable.
10. Organize results by contribution type: earliest verified contribution, most influential contribution, major findings, latest important progress, active hot topics, unresolved key problems.
11. Mark uncertainty using references/evidence-and-uncertainty.md.
12. Produce a stage-appropriate output using references/stage-outputs.md.

Field Adaptation

Do not use one evidence model for every field.

• In STEM and medicine, prioritize peer-reviewed articles, systematic reviews, meta-analyses, clinical guidelines, datasets, benchmark papers, methods papers, and preprints when recency matters.
• In computer science and engineering, include top conferences, proceedings, standards, technical reports, arXiv, and code/data artifacts when they are field-relevant.
• In social sciences, include peer-reviewed journals, books from academic presses, working papers when customary, policy reports when appropriate, and methodological debates.
• In humanities, include monographs, edited volumes, critical editions, archives, translations, and field-defining essays; citation counts may be weak or misleading.
• In law, policy, and business, separate scholarly literature from statutes, cases, regulations, standards, market reports, and institutional reports.

State which evidence model is being used and why.

Required Review Axes

For most literature reviews, explicitly cover:

• earliest known or earliest verified contribution
• canonical, highly cited, or field-defining contributions and why they matter
• major discoveries, theories, mechanisms, methods, datasets, cases, or debates
• latest important progress, with exact publication dates when relevant
• currently hot topics and why they are active
• unresolved key problems, including whether they are theoretical, empirical, methodological, computational, translational, conceptual, or normative
• disagreements, failed assumptions, controversies, and limitations where the literature is not settled

Quality Gate

Before finalizing, check that:

• every factual source claim is attached to a cited source
• every included source has a clear inclusion reason
• every source has DOI, persistent identifier, or URL status noted
• abstracts, summaries, introductions, or equivalent source summaries were read
• discussion/conclusion sections were checked when full text was available and the claim depends on them
• field-specific quality signals are stated, not silently assumed
• unverified items use uncertainty labels, not confident prose
• "latest" and "hot topics" are based on current search, not memory

Output Discipline

Use tables for evidence inventories and concise prose for synthesis. Keep bibliographic records structured enough to audit:

Authors | Year | Title | Venue/Source Type | DOI/ID | URL | Citations/Metric | Quality Signal | Inclusion Reason | Evidence Checked | Notes

Separate verified findings from interpretation. When synthesizing, preserve provenance: say which source supports which claim, and distinguish a source's own conclusion from a broader inference across sources.

References

• Use references/search-and-validation.md for source inclusion, database selection, venue quality, and metadata cross-checking.
• Use references/stage-outputs.md for early, mid, and late stage review depth and deliverable formats.
• Use references/evidence-and-uncertainty.md for uncertainty labels, confidence grading, and anti-hallucination rules.