---
name: systematic-review
description: Plan and conduct protocol-driven evidence reviews with explicit eligibility, reproducible searching, study linkage, screening records, appraisal, and transparent synthesis.
---

# Systematic review

Use this Academic Research Assistant skill when evidence selection must be explicit, reproducible, and auditable. Follow the [research contract](../../docs/RESEARCH-CONTRACT.md). Choose the review type deliberately: an effect review, qualitative evidence synthesis, diagnostic review, methodological review, and scoping review answer different questions. A scoping review is not simply an effect review with the appraisal step omitted.

## Inputs and protocol

Obtain the review question, intended use, disciplinary scope, team capacity, time, access, and any required reporting or registration guidance. Establish inclusion criteria for populations or materials, concepts, settings, designs, outcomes where appropriate, dates, languages, and publication types. Justify restrictions and explain the likely coverage loss.

Write a protocol before screening at scale. Specify discovery sources, search strategy, deduplication, screening stages, reviewer roles, disagreement resolution, extraction fields, appraisal approach, synthesis plan, and amendment handling. Verify current requirements if registration or a reporting guideline is requested. Drafting a registry entry does not mean it was submitted or accepted.

## Execute the review

1. Run the documented search through [literature search](../literature-search/SKILL.md). Preserve exact queries, search dates, filters, counts, and exported records. Include supplemental discovery methods where justified by the subject and log their contribution.
2. Deduplicate records while retaining provenance. Link multiple reports of the same underlying study. Distinguish the counts of records, reports, and studies throughout the review.
3. Pilot eligibility criteria on varied examples. Clarify ambiguous criteria before broad screening and record protocol changes. During screening retain include, exclude, and uncertain decisions rather than forcing uncertainty into an exclusion.
4. Perform title/abstract screening and full-text assessment according to the protocol. Record a primary full-text exclusion reason that reflects an actual criterion. Track inaccessible reports separately from excluded studies.
5. Use independent reviewers when required by the protocol or methodological standard. If only one reviewer or one assistant performed a task, state that fact and its implications; do not invent independence, consensus, or inter-reviewer agreement.
6. Extract evidence with a piloted codebook through [evidence extraction](../evidence-extraction/SKILL.md). Record study/report relationships, outcome definitions, time points, and values capable of affecting the synthesis.
7. Appraise evidence with criteria suited to each design through [research appraisal](../research-appraisal/SKILL.md). Decide prospectively how judgments affect synthesis and sensitivity analyses. Avoid turning unrelated quality domains into an unexplained total score.
8. Synthesize according to the question and compatibility of evidence. Use [meta-analysis](../meta-analysis/SKILL.md) only when pooling has a defensible interpretation. A structured narrative synthesis should explain patterns, uncertainty, dependence, and differences in methods; it should not count significant studies as votes.

## Boundaries across evidence types

For quantitative reviews, define effect measures, unit-of-analysis issues, missing information, subgroup hypotheses, and handling of multiple eligible outcomes. For qualitative synthesis, specify an interpretive or aggregative approach and preserve context, meaning, and researcher judgments. For theoretical or computational reviews, formulate reproducible eligibility around problem definitions, assumptions, algorithms, evaluation regimes, and artifact types. Do not force these works into intervention-study categories.

Assess confidence in the body of evidence using a framework appropriate to the review purpose when needed. Explain reasons for uncertainty, including bias, inconsistency, indirectness, imprecision, and missing evidence where relevant. Do not infer that a long checklist makes the conclusion certain.

## Deliverable and verification

Deliver the protocol and amendments, search log, screening decisions, study linkage, extraction codebook and dataset, appraisal records, synthesis, and a flow accounting. Reconcile counts with actual logs; do not invent numbers to complete a diagram. Include the last search date and an update plan if the intended use requires currency.

Verify a sample of exclusions and conclusion-driving extraction values against sources, following the actual review team's agreed checks. Clearly state uncompleted stages and access limits. A protocol, a pilot search, or partially screened corpus should be labeled by that stage and must not be presented as a completed systematic review.