Research Systematic Literature Review

Quick start

1. Require a clear domain and research question before substantive review work. In paper-context mode, infer the domain from the target paper summary only when unambiguous; otherwise stop.
2. Declare one review profile: comprehensive-systematic, bounded-systematic, critical-evidence-map, rapid-scan, or novelty-prior-art.
3. Collect optional inputs, including whether the user wants standard or detailed_math exposition. Apply non-domain defaults only when safe and log every assumption.
4. Check search, corpus, and Zotero access. Stop or narrow the review profile when the requested assurance is impossible.
5. Initialize the full review pack with scripts/init_review_pack.py for systematic profiles; use paper-context mode only for bounded contextualization.
6. Complete discovery assurance before expensive extraction: coverage map, visible seed set, optional withheld challenge set, multi-channel search decisions, backward and forward citation searching, search repair, recall audit, stopping rationale, and candidate-corpus freeze.
7. After freeze, deduplicate, resolve preprints to canonical publications, screen, extract evidence, synthesize, grade confidence, and run adversarial checks.
8. Record a canonical publication_url for every included work, preferring the published or accepted venue record over a preprint when one exists.
9. Generate PRISMA flow accounting and validate the declared process profile with scripts/validate_review_pack.py.
10. State exactly what the review assurance supports. Structural validity and PRISMA consistency do not establish that all important publications were found.
11. If the user asks for project-level sequencing or a decision across multiple stages, invoke research-pipeline-planner first.

Review profiles

Comprehensive systematic