Literature Review

You find what the published research actually shows on a question, judge how much weight each study can bear, and say where the evidence is strong, weak, mixed or missing. The review is reproducible: someone else following your protocol would find the same studies. It is honest about its type: a rapid structured review by one reviewer is not a full systematic review, and you say which it is.

When to use

• "What does the research say about four-day weeks and productivity?"
• "Is there evidence this intervention works?"
• "Find the key papers on X for our grant application."
• "Summarise the academic work on our market or technology."
• Before a decision where the owner wants evidence, not opinion.

What you need

• The question, why it matters, and how the answer will be used.
• Constraints: date range, languages, study types of interest, depth and deadline.
• Access: free databases through web_search, web_extract and your browser (Google Scholar, PubMed, OpenAlex, Semantic Scholar, arXiv and other preprint servers, SSRN, CORE, DOAJ). If the owner has institutional or paid access, they can sign your browser in. Use Unpaywall or the authors' own repository copies to find legal open-access versions. Never use pirate sites.
• Any papers the owner already knows about, as seeds.

Method

1. Frame the question with a structure that fits it:
   • PICO for interventions: Population, Intervention, Comparison, Outcome
   • PEO for exposures or experiences: Population, Exposure, Outcome
   • for broad topics, define the concept, the context and the population Agree the framed question with the owner if there is any doubt.
2. Write the protocol before searching and save it: databases, search strings (with synonyms and Boolean operators, for example ("remote work" OR telework OR "work from home") AND (productivity OR performance)), date range, languages, inclusion criteria (study types, populations, outcomes) and exclusion criteria. Changes later are allowed but logged.
3. Search and log. Run each string in each database; record the date, string and number of hits. Export or copy results into a single list with execute_code and remove duplicates by DOI or title.
4. Screen in two passes, logging counts and reasons:
   • titles and abstracts against the criteria
   • full texts of the survivors, recording the reason for each exclusion Report the flow in the style of a PRISMA 2020 flow diagram: records identified, duplicates removed, screened, excluded, full texts assessed, excluded with reasons, included.
5. Snowball. Check reference lists of included studies (backward) and works that cite them (forward, through Google Scholar's "cited by" or OpenAlex). Log what this adds.
6. Appraise every included study with the checklist in references/appraisal-checklists.md for its design. At minimum record: design, sample size and population, what was measured and how, the main result with effect size and confidence interval where given, risk of bias, funding and conflicts of interest, and whether it is peer reviewed. Check each study against retraction notices (Retraction Watch data, now distributed through Crossref) and note journal concerns where relevant.
7. Extract into an evidence table: one row per study per outcome. Do it in a spreadsheet or CSV, not prose.
8. Synthesise by theme or outcome, not paper by paper:
   • what most studies find, and how large the effect is
   • where results differ, and plausible reasons (population, method, measure, setting, date)
   • which findings rest on strong designs and which on weak ones Avoid vote counting ("7 studies positive, 3 negative") as the main argument; weight by quality and size.
9. Rate certainty per main conclusion using GRADE-style levels:
   • High: further research is very unlikely to change the conclusion
   • Moderate: further research may change it
   • Low: further research is likely to change it
   • Very low: any estimate is very uncertain Downgrade for risk of bias, inconsistency, indirectness, imprecision and suspected publication bias.
10. Name the gaps: populations, outcomes or settings with little or no evidence.
11. Write it up in the structure below and state the review type and its limits.

Output

# Literature review: <question>
Type: rapid structured review (single reviewer) | <other>   Date: <date>
Protocol: <link to saved protocol>

## Bottom line
<3 to 5 sentences: what the evidence shows, how certain, and the main gap.>

## Search and selection
<Databases, dates, strings (or link), and the flow counts.>

## Findings by theme
### <Theme or outcome>
<Synthesis with citations, effect sizes, certainty level and why.>

## Conflicts and uncertainty
## Gaps
## Limitations of this review
## Evidence table (link) and references (full citations with DOI)

Put the bottom line and certainty levels on a show_card.

Checks before you finish

• The protocol is saved and any changes are logged.
• The flow counts add up from identification to inclusion.
• Every included study has an appraisal and appears in the evidence table.
• Every claim in the synthesis cites the studies it rests on.
• Preprints and non-peer-reviewed sources are labelled as such.
• Certainty is stated for each main conclusion, with the reason.
• The review type and its limitations are stated.

Pitfalls

• Reading only abstracts. Abstracts overstate. Base the appraisal on the full text.
• Cherry-picking. Searching until you find support for a view is advocacy, not review. The protocol prevents it.
• Treating all studies as equal. One large well-designed trial outweighs several small uncontrolled ones.
• Ignoring publication bias. Positive results are published more often; say so when the evidence base is small and all positive.
• Confusing correlation with causation. Observational studies rarely justify causal language.
• Over-claiming the review type. Call it systematic only if it truly followed a full systematic protocol with independent screening.
• Sign-off. For medical, clinical or safety questions, the review informs a qualified professional; it is not advice to act on.

Appraisal checklists: references/appraisal-checklists.md. Credits: references/CREDITS.md.