systematic-review

name	systematic-review
description	Write a PRISMA 2020 systematic review — structured abstract, reproducible search/eligibility methods, a mandatory Mermaid PRISMA flow diagram of records identified/screened/excluded/included, risk-of-bias assessment, and synthesis of results. Use when the deliverable must make the evidence-selection process legible and reproducible end to end. Anti-trigger; for a general scholarly write-up with no mandatory flow diagram or registered protocol use academic.
argument-hint	<the review question or body of evidence to synthesize>

systematic-review

Produces a PRISMA 2020 systematic review: a reproducible, auditable account of how a body of evidence was identified, screened, appraised, and synthesised. Its center of gravity is the PRISMA flow diagram — the report is not conformant without one reconciling the count of records at every stage. This genre follows PRISMA 2020 (Page et al., BMJ 2021), the current reporting standard for systematic reviews and meta-analyses. Before authoring, verify the current PRISMA guidance live — confirm which statement is in force and treat no item count or flow-diagram template as fixed until checked against the live guidance.

Pattern (industry: PRISMA 2020 systematic review and meta-analysis)

1. Title / Structured Abstract — background, objectives, eligibility criteria, information sources, methods of synthesis, results, and conclusions, each named as its own clause of the abstract.
2. Introduction — the rationale for the review against existing knowledge, and the explicit objectives (the review question, framed as PICO or a comparable structure where applicable).
3. Methods — the reproducible protocol:
   • Eligibility criteria — inclusion and exclusion criteria for evidence.
   • Information sources — where evidence was sought.
   • Search strategy — the search approach and terms.
   • Selection process — how records were screened and selected.
   • Data items — what was extracted from each included study.
   • Risk-of-bias assessment — how the validity of each included study was appraised (name the tool, e.g. Cochrane RoB 2).
4. Results — the study selection and synthesis:
   • Study selection, with the PRISMA flow diagram: the counts at every stage — records identified, records screened, records excluded (with reasons), and studies included.
   • A risk-of-bias / verification-verdict table for the included studies.
   • Synthesis of results — the synthesised findings across included evidence.
5. Discussion — interpretation, limitations of the evidence and the review process, and conclusions.
6. Registration & Protocol — the registration record and protocol availability (PRISMA 2020 "Other information"); state explicitly when none exists.
7. References — the full reference list.

Rules that keep it a systematic review

• The PRISMA flow diagram is mandatory and never silently omitted. Render it as a Mermaid flowchart (top-down, flowchart TD) — never ASCII art, an image link, or Graphviz/DOT. If the underlying data cannot support a stage, say so in prose rather than dropping the diagram.
• The flow diagram must reconcile: records identified, minus those removed or excluded at each stage, equal the studies included. A diagram whose counts do not add up is a defect, not a rounding artifact.
• Every claim traces to a cited source; no orphan facts and no uncited assertions.
• Report the appraisal outcome of every included study explicitly. A study excluded on eligibility or risk-of-bias grounds is recorded as an excluded record in the flow diagram with its exclusion reason — never quietly dropped from the corpus.
• Exhaustive coverage: synthesize from the full set of eligible studies — every included study carries its own evidence (population, comparator, outcome, citation), never condensed to a cherry-picked subset.
• Numbered (Vancouver-style) inline citation markers [1], [2] resolve to a references list; a risk-of-bias/verification-verdict table sits alongside the flow diagram as a plain Markdown table.
• The Methods section is reproducible on its own: another reviewer must be able to repeat the search and selection process from the prose alone.

Anti-triggers — do not use this genre for

• A general scholarly write-up with no mandatory flow diagram or registered search protocol — that is academic: a formal research report (abstract, background, method, findings, discussion) with a selectable citation style, but no PRISMA-mandated flow diagram or reproducible multi-stage search protocol.
• A single already-made technical decision — that is adr.
• What to build and why, before design — that is prd or feature-spec.

MIF frontmatter

type: semantic — a systematic review is declarative synthesized knowledge, not a time-bound event or a step sequence. Use namespace: systematic-review/<area> (e.g. systematic-review/clinical-nlp). Climb to L2 with modified, title, and tags when the review context supplies them. Gate every output with mif-validate at its target level; the floor is --level 1.

Why machine-readable — the point of MIF here

Question an agent asks	Answered by (frontmatter)
Is this evidence synthesis still current, or has newer evidence appeared?	temporal.validFrom / ttl
Where did each claim come from; can I trust it?	provenance (W3C-PROV) + citations[]
Which studies were included vs. excluded, and why?	the PRISMA flow-diagram counts, cross-checked against citations[]
What downstream document does this review inform?	typed relationships[] (relates-to, realized-by)

The same document still reads as a human systematic review and projects losslessly to JSON-LD and back — one artifact, two readers.

The L1 -> L3 climb (two exemplars)

• templates/good-l1.md — L1 floor: id, type, created + body. A complete, valid PRISMA review, but opaque to a machine consumer.
• templates/good.md — L3 (highest this genre supports): adds namespace, modified, temporal validity, W3C-PROV provenance, citations[] tied to the methodology and appraisal tool, and a typed relationships[] graph (e.g. relates-to a PRD the review's conclusions inform). Validate with mif-validate --level 3.

Author at the highest level the drafting context supports (grade down rather than fabricate). templates/bad.md shows the antipattern: a review that reports a study count and a "results" narrative with no PRISMA flow diagram anywhere in it.