literature-review

Literature Review & Evidence Synthesis

Guide the full arc from scattered sources to synthesized argument: choosing the right review genre, searching systematically, screening with an audit trail, extracting into structured forms (annotated bibliography, literature matrix), and building the theoretical or conceptual framework the study will stand on. The unifying principle: a literature review is an argument about a field, not a pile of summaries — every stage exists to turn vertical reading (paper by paper) into horizontal synthesis (across papers, by theme, method, and theoretical stance).

Quick Reference

Task	Reference
Choosing a review type; search strategy, logging, and the toolkit's literature tools	Read references/search-and-review-types.md
Screening with audit trails, annotated bibliographies, literature matrices	Read references/screening-and-synthesis-guide.md
Theoretical vs. conceptual frameworks and how to build one	Read references/framework-construction-guide.md

Workflow

Step 1: Establish the Review's Genre and Purpose

Read references/search-and-review-types.md before searching. Ask what the review must do for the project:

• Narrative–theoretical review (the anthropological default): positions the study within ongoing scholarly conversations, traces concepts genealogically, and builds toward a framework. Selection is purposive and argued, not exhaustive.
• Scoping review: maps what exists across a broad or emerging area — how much, of what kind, using which methods — without appraising every study in depth.
• Systematic review: answers a focused question by exhaustively identifying, screening, and appraising all relevant studies under a registered protocol (PRISMA). Common in applied, medical, and policy anthropology; rarely the right default for interpretive projects.

The genre determines everything downstream: search exhaustiveness, screening formality, and what counts as done.

Step 2: Search Systematically and Log Everything

• Build the search from the research question's concepts plus their variants (synonyms, subfield vocabularies, spelling variants).
• Use the toolkit's literature tools where available — the ai-anthropology MCP tools (search_openalex, search_crossref, search_pubmed, search_google_scholar) or the corresponding Colab notebooks — and complement database search with citation chasing (backward through reference lists, forward through cited-by) and grey literature where the field lives outside journals.
• Log every search: database, exact query, filters, date run, record count. A review that cannot state how it searched cannot defend what it found. Note preprints as preprints.

Step 3: Screen with an Audit Trail

Read references/screening-and-synthesis-guide.md.

• Define inclusion and exclusion criteria before screening, from the review's purpose — then apply them in two passes: title/abstract, then full text.
• Record every exclusion with its reason. For systematic reviews, track counts at each PRISMA stage; for narrative reviews, a lighter decision log still pays for itself at the writing stage.
• Disclose AI assistance: when a model helps screen or extract, the write-up must say which tool, how it was configured, and how its recommendations were human-verified — this is a PRISMA 2020 requirement and good practice in every genre. Never let the model's screening judgment stand unreviewed.

Step 4: Annotate and Extract

• Build the annotated bibliography as an analytical instrument, not a formality: each entry gets a summary (aims, methods, findings), an evaluation (credibility, limitations, evidentiary weight), and a reflection (what it does for this project).
• Build the literature matrix in parallel: one row per source, columns for citation, question, theoretical orientation or lens, design and sample, key concepts, findings, and relevance to the study. The matrix is what turns reading into synthesis — patterns, conflicts, and gaps appear down its columns, not within its rows.

Step 5: Synthesize

Work from the matrix, not from memory of the papers:

• Group by theme, debate, and theoretical commitment — not by paper.
• Name the disagreements: where findings conflict, ask whether the conflict is empirical (different populations, periods, methods) or theoretical (different assumptions about what counts).
• State the gaps precisely: a gap is not "little research exists" but "existing work assumes X, has not examined Y under conditions Z."
• Sample-check every claim against its source before it enters the synthesis — verbatim quotes carry page numbers; paraphrases stay faithful.

Step 6: Build the Framework

Read references/framework-construction-guide.md.

• Theoretical framework: an existing theory (or theories) adopted to interpret the phenomenon — it drives the questions and the methods.
• Conceptual framework: the researcher's own synthesis of concepts and relationships, built when no single theory covers the study's terrain.
• Construct in three movements: delimit the phenomenon and gather evidence; extract concepts and map their relationships; verify internal consistency and render the framework as diagram plus narrative. Every element must trace back to the matrix.

Guardrails

• Never fabricate or embellish citations. Every reference must correspond to a real, verified source — check DOIs resolve, verify metadata against the databases, and flag any source the model "remembers" but cannot verify. A hallucinated citation is the single fastest way to destroy a review's credibility.
• Read before citing. No source enters the synthesis on the strength of its abstract alone; if only the abstract was read, the write-up says so.
• Selection is an argument. In narrative reviews, justify what is in and what is out; purposive is not the same as arbitrary.
• Disclose AI participation in searching, screening, or extraction as a methods fact, with human verification described.
• Preprints are labeled as preprints, and retraction status is worth checking for load-bearing sources.

Common Failure Modes

Serial summarizing. One paragraph per paper, in sequence, with no cross-referencing — a book report, not a review. The cure is the matrix: synthesis reads down columns, across studies.

The annotated bibliography as endpoint. Annotations that only describe ("this study examined...") without evaluation or reflection do none of the analytical work the format exists for.

Matrix without synthesis. A beautifully complete matrix that is never read for patterns — extraction is preparation, not conclusion.

Framework as literature dump. A "framework" section that lists theories without committing to relationships among concepts. A framework makes claims; it can be drawn.

Faux-systematic claims. Calling a purposive narrative review "systematic" — the word is a protocol claim, not a compliment. Use the genre's real name and meet its real standard.

Unverified citations. Accepting model-recalled references without resolution against a database. Verify, then cite.

Examples

Example 1: Dissertation literature review

Input: "I need to do the literature review for my dissertation on medical pluralism in urban Ghana."

Output approach: Narrative–theoretical genre. Search OpenAlex/CrossRef for medical pluralism + regional and conceptual variants; chase citations from the field-defining pieces; log queries. Screen purposively with a decision log; build annotated bibliography and a matrix with a theoretical-orientation column; synthesize by debate (e.g., pluralism as system vs. as practice); build toward the theoretical framework the analysis will use.

Example 2: Systematic review in applied anthropology

Input: "We're doing a systematic review of community health worker interventions for a policy client."

Output approach: PRISMA genre. Define criteria and protocol first; parallel database searches with logged queries and counts; deduplicate; two-pass screening with per-exclusion reasons and stage counts for the flow diagram; disclose any AI-assisted screening with verification procedure; extract into the matrix; synthesize with attention to intervention heterogeneity.

Example 3: From matrix to framework

Input: "I've read forty papers on platform labor. How do I turn this into a conceptual framework?"

Output approach: Read the framework-construction guide. Extract recurring concepts from the matrix; classify relationships (what shapes what, what mediates); commit to a diagram with narrative justification; verify every element traces to sources; identify which relationships are well-evidenced versus the study's own contribution.