---
name: research-manuscript-evaluation
description: Use when reviewing or scoring a research manuscript against explicit, evidence-linked criteria.
---

# Evaluate a research manuscript

Establish the manuscript version, audience, purpose, authorized review scope and available evidence. Determine whether the user supplies a rubric or requests a qualitative review. Do not imply access to underlying data, code, references, peer-review history or completed analyses that were not actually provided.

Use criteria appropriate to the question and design, such as clarity of contribution, soundness of assumptions or methods, evidence supporting claims, transparency, reproducibility, treatment of uncertainty and intelligibility. For each material finding record its location, observed evidence, likely consequence, severity and a feasible correction or verification step. Distinguish a factual defect, an unresolved question and a stylistic preference.

If a rubric is supplied, apply its stated dimensions, weights and scale consistently, and show how each score is supported. Do not invent a required rubric, compress missing evidence into a high score, or imply that a numerical grade proves correctness or acceptance. If the rubric is absent, propose clearly labeled criteria or give a qualitative assessment with the user's intended use in mind.

For arithmetic on a user-supplied weighted rubric in this package, run `python -X utf8 -B tools/rubric_score.py review.json` from the package root. This validates recorded scores and evidence notes and writes a JSON summary to standard output. A score requires both a specific evidence note and a rationale. Use `null` for an unassessed criterion; the tool reports assessed-weight coverage and withholds a complete mean until all criteria have scores. It does not choose the rubric, read the manuscript, judge the evidence, or assign ratings. See [rubric scoring format and limits](../../docs/RUBRIC-SCORING.md).

Return a prioritized review with strengths, consequential limitations, specific evidence pointers and outstanding checks. When edits are authorized, track what was changed and recheck affected claims and references; otherwise provide review comments without modifying the artifact. Keep reviewer and manuscript material confidential within the user's authorized workflow.