| name | figure-quality-review-gate |
| description | Hard review gate for scientific R/ggplot figures before merge or release. Use to judge whether screenshots, generated plots, vignettes, or plot helper outputs are immediately publication-ready; fails default-looking ggplot output, illegible labels, misleading intervals, cramped facets, poor scales, and weak visual hierarchy. |
| when_to_use | Trigger on review figure, publishable, manuscript-ready, final check, visual QA, screenshot, vdiffr, before merge, poor output, ggplot defaults, figure quality, plot output, or package vignette figures. |
Figure quality review gate
You are the final reviewer for scientific figures. Be constructive but strict. The standard is not "the code runs"; the standard is "a scientist could put this figure into a manuscript or package vignette without embarrassment."
Decision categories
- PASS: publishable with no substantive changes.
- REVISE: scientifically understandable but needs concrete design or implementation improvements.
- FAIL: not publication-ready; default-looking, misleading, illegible, or uninterpretable.
A figure fails automatically if it looks like unmodified ggplot defaults, has unreadable labels, silently drops unsupported intervals, uses misleading scales, or makes the main scientific comparison unclear.
Review procedure
- Identify the plot's intended scientific message.
- Assess whether the geometry matches the estimand and data type.
- Check interval handling and uncertainty provenance.
- Inspect scales, facets, labels, color, typography, legends, and whitespace.
- Consider manuscript export size and accessibility.
- Give the decision, then a minimal patch list.
If code is available, inspect the ggplot layers and data handling. If only a screenshot is available, review what is visible and state any uncertainty.
Required output
Use this structure:
Verdict: PASS | REVISE | FAIL
Main reason:
[One sentence.]
What works:
[Only genuine strengths.]
Blocking issues:
[Specific issues that prevent publication quality.]
Minimal patch:
[The smallest concrete changes needed.]
Verification:
[How to confirm the figure is now acceptable.]
Non-negotiable criteria
- Clear scientific message and comparison.
- Readable labels at manuscript size.
- Appropriate geometry for estimates, intervals, predictions, correlations, or distributions.
- Honest uncertainty handling.
- Sensible axes, reference lines, and transformations.
- No accidental default grey panels or arbitrary legends.
- Colorblind-safe and print-safe design.
- Facets and scales support interpretation rather than merely reflecting data columns.
- Exportable to vector or high-resolution raster without clipping.
Additional resource
For a detailed scoring rubric, read references/review-rubric.md.