Evidence Synthesis
Rapid evidence assessment without shortcuts that collapse confidence. Every finding carries a source and a confidence grade. Gaps are named, not hidden.
When to use
Trigger for evidence review work: programme design questions, evaluation framing, policy briefs, funder questions about what works, contribution analysis needing prior-evidence assembly.
Do not trigger for news scans, stakeholder mapping, or quick factual lookups. Those are different tasks.
Required inputs
Ask in one batch. The first two are required.
- Question: what Ane needs the evidence to answer, as a specific question (required)
- Purpose: programme design, evaluation framing, policy advocacy, donor response, academic input (required; shapes depth and format)
- Scope constraints: geography, population, intervention type, time range (optional; will default if missing)
- Sources Ane trusts or distrusts: organisations, journals, or author groups to prioritise or treat cautiously (optional)
- Timeline: how much time Ane has — determines whether rapid (2 hours), standard (1-2 days), or rigorous (1-2 weeks) synthesis (default: rapid)
Method
Step 1 — frame the question
Use the framework that fits:
- PICO for intervention-effectiveness questions: Population, Intervention, Comparison, Outcome
- SPIDER for qualitative or mixed-method questions: Sample, Phenomenon of Interest, Design, Evaluation, Research type
- SPICE for service evaluation questions: Setting, Perspective, Intervention, Comparison, Evaluation
Show the frame explicitly before searching.
Step 2 — define inclusion and exclusion criteria
For rapid synthesis: minimum criteria are population, intervention or phenomenon, outcome, timeframe (default last 10 years), language, study type.
State each criterion. Explain any exclusion decision that would surprise a peer reviewer.
Step 3 — select synthesis approach
Choose one, name the choice:
- Thematic synthesis: when findings group into recurring themes (Thomas & Harden 2008)
- Narrative synthesis: when studies are too heterogeneous to pool but a structured summary is needed (Popay et al. 2006)
- Realist synthesis: when the question is "what works, for whom, in what circumstances, and why" (Pawson et al. 2005)
- Meta-analysis: only when effect sizes from comparable quantitative studies can be pooled. Usually out of scope for rapid work.
Step 4 — search and screen
If Ane provides sources, use them. If not, ask where to search:
- Grey literature: WHO, UNFPA, UNAIDS, IPPF, Cochrane, 3ie, relevant IGOs
- Academic: PubMed, Scopus, Global Health Database, Cochrane
- Ane's resource library: subfolders matching the topic
Never invent citations. If a search cannot be conducted in this session, ask Ane to paste the top sources or point to the library subfolder.
Step 5 — extract findings
For each source, extract:
- Full citation (author, year, title, journal or publisher, DOI if available)
- Setting and population
- Intervention or phenomenon studied
- Key finding, in the source's own framing
- Study design and sample size
- Confidence flags: low sample, non-random, self-report, author conflicts, funder bias
Step 6 — grade confidence
For each finding or theme, grade using a GRADE-adjacent scheme:
- High: consistent findings across multiple rigorous studies; mechanism understood
- Moderate: consistent findings but with methodological limitations or narrow context
- Low: few studies, or conflicting findings, or serious bias risk
- Very low: single source or methodologically weak studies only
Explain the grade in one clause. No inflation.
Step 7 — apply the relevant lenses
For SRHR or gender-related questions, apply:
- Feminist lens: whose voices shaped the research questions? Are women and girls subjects of the research or objects? Cornwall & Rivas (2015) framing.
- Decolonial lens: where was the research conducted, who funded it, whose knowledge is centred? Chilisa (2020).
- Intersectionality: does the evidence disaggregate to current standard (age, gender identity, disability, geography)? Flag when it does not.
Step 8 — identify gaps
Name what the evidence does not answer. Use
format. Distinguish:
- Gaps in research (the study has not been done)
- Gaps in context (research exists but not from relevant settings)
- Gaps in population (research exists but excludes the target group)
- Gaps in method (research exists but with weak designs only)
Output structure
Produce an evidence brief with these sections:
- Question — as framed in Step 1, with the framework named
- Method — inclusion criteria, search approach, synthesis approach, limitations of the rapid format
- Key findings — organised by theme. Each finding:
- One-sentence statement
- Confidence grade
- Supporting sources (author year)
- Lens observations — feminist, decolonial, intersectional notes
- Implications — what this means for the stated purpose (programme design, evaluation, etc.). Be concrete.
- Evidence gaps — entries
- Sources — full citations, alphabetical by first author
Citation requirements
Every finding cites at least one source. Method references:
- Thomas & Harden (2008) for thematic synthesis
- Popay et al. (2006) for narrative synthesis
- Pawson et al. (2005) for realist synthesis
- GRADE Working Group for confidence grading
For SRHR-specific framings:
- Cornwall & Rivas (2015), Chilisa (2020) for lenses
- WHO/UNFPA (2023), UNFPA HRBAP for rights-based framing
Writing rules
Follow CLAUDE.md house style. In this skill specifically:
- Never summarise a finding in language stronger than the source supports.
- Never present contested findings as settled.
- Never use "evidence shows" without a specific citation.
- Flag when a finding is contested, and by whom.
Limitations
Rapid syntheses are not systematic reviews. State this limitation in the Method section. Do not invent effect sizes or pool findings across incompatible studies. If Ane needs a systematic review, route to a proper review protocol (PRISMA) rather than inflating rapid-review scope.