Literature review

Collect and synthesise sources without invented references or conclusions.

Result

You produce a source matrix and a bounded synthesis in which every substantive statement can be traced to a reviewed source location. Evidence, interpretation, disagreement, and missing information remain distinct.

Minimum source matrix

Use stable bibliographic identifiers plus: access level (full text or abstract only), selection decision, study context, method, sample or material, reported result, limitations, exact supporting location, reviewer note, and verification status.

Actions

1. Define the review question, scope, date boundary, eligible publication types, and selection criteria.
2. Create and test the source-matrix fields on two known papers before processing the full set.
3. Import bibliographic data without inventing missing fields; keep stable identifiers and the source database.
4. Read the permitted material and record the exact page, section, table, figure, or quoted passage supporting each extracted point.
5. Mark access as full text, abstract only, metadata only, or unavailable, and limit claims accordingly.
6. Group studies by a stated substantive basis such as method, population, mechanism, or result; preserve contradictory findings.
7. Draft the synthesis from verified matrix rows and attach each claim to its source locations.
8. Run a separate citation and claim audit before acceptance.

Check

Sample the synthesis from both directions: each substantive claim resolves to an available source location, and each cited source supports the sentence that cites it. Dates, author names, identifiers, quotations, and access-level labels match the source matrix. Disagreement and evidence gaps remain visible.

If it fails

Remove unsupported claims, narrow them to the available evidence, or turn them into explicit open questions. Correct the matrix before regenerating prose. Never ask a model to add plausible references or infer full-text findings from an abstract.

Next: Evidence package.