version: "1.0.0" name: sci-research-lifecycle description: Run or audit an evidence-traceable scholarly research lifecycle from question framing and reproducible search through claim mapping, writing, figure/table QA, journal submission, revision, and archiving. Use for manuscripts, literature matrices, research project packs, submission checks, or reviewer responses; do not treat structural validation as proof of scientific truth.

SCI Research Lifecycle

Outcome

Produce the requested research artifact with traceable sources, calibrated claims, preserved history, explicit uncertainty, and a clear human-authorization boundary.

Start

1. Read the nearest project instructions and inspect current source artifacts.
2. Identify project ID, manuscript version, target article type/journal if any, lifecycle stage, deliverable, and authoritative record location.
3. State consequential unknowns. Never infer experiments, results, authorship, affiliation, funding, ethics approval, submission status, or reviewer identity.
4. Preserve raw exports, original data, prior submissions, signed declarations, and formal correspondence.

For a new auditable project pack, run scripts/init_project.py <destination>. For an existing pack, run scripts/validate_project.py <project-directory> before and after material changes.

To normalize a CSV, RIS, or NBIB export without changing the raw file, run scripts/import_literature.py <raw-export> <literature_register.csv> and read references/imports-and-normalization.md. Imported records remain unverified and uncertain until human review.

Route to the smallest protocol

• Question, contribution, protocol, eligibility, or scope: read references/question-and-protocol.md.
• Database search, deduplication, screening, or flow counts: read references/search-and-screening.md.
• CSV, RIS, or NBIB metadata import: read references/imports-and-normalization.md.
• Source/DOI verification or evidence extraction: read references/evidence-integrity.md.
• Manuscript claims, novelty, causality, or synthesis: read references/claim-evidence.md.
• Quantitative tables, conceptual figures, captions, or calculation lineage: read references/figures-and-tables.md.
• Journal matching, submission files, transfer, reviewer response, or revision: read references/submission-and-revision.md.
• Project state, versions, decisions, audit entries, or handoff: read references/memory-and-audit.md.
• Schema fields or validator behavior: read references/project-schema.md.
• Final readiness or release decisions: read references/quality-gates.md.

Read only the references required for the current request.

Working rules

• Verify primary or authoritative sources before promoting a record to verified.
• Assign stable Evidence IDs and Claim IDs; never silently reuse an ID for a different object.
• Classify support as direct, indirect, conflicting, absent, or author synthesis.
• Lower claim strength when design, setting, sample, access, or conflicting evidence limits inference.
• Distinguish reported findings, author synthesis, and proposed hypotheses.
• Recalculate counts, percentages, totals, and flow diagrams from registered inputs.
• Use current official journal/publisher pages for changeable requirements and record the check date.
• Inspect rendered Word, spreadsheet, figure, PDF, and submission-system outputs when layout affects correctness.
• Never activate final submission, approval, withdrawal, public release, or destructive cleanup without explicit authorization for that action.

Completion

Finish only when the requested artifact exists, material claims have traceable support or explicit unsupported status, unresolved items remain labeled, project validation passes or exceptions are documented, rendered outputs are checked where relevant, and the audit record contains the change and next safe action.