| name | alterlab-figure-qa |
| description | Verify publication figures with a render-then-check QA pass — data-fidelity against every underlying row, axis/label floor-and-ceiling legibility, bounding-box collision detection for overlapping text/markers, and 300-dpi print-readiness. Use when proofing or auditing a finished figure for correctness and print quality, catching mislabeled or overlapping elements, or confirming a plot faithfully represents its data before submission. For CREATING the plot prefer alterlab-matplotlib (or alterlab-seaborn / alterlab-plotly); for multi-panel publication layout prefer alterlab-scientific-viz; for schematic diagrams prefer alterlab-scientific-schematics. Part of the AlterLab Academic Skills suite. |
| license | MIT |
| allowed-tools | Read Write Edit Bash(python:*) Bash(uv:*) |
| compatibility | Runs under `uv run python` with the plotting stack already used to make the figure (Matplotlib and, for image/text-box checks, Pillow). No GPU, no account. Operates on a rendered figure plus the source data — it re-renders and inspects, it does not need network access. |
| metadata | {"skill-author":"AlterLab","version":"1.0.0"} |
Figure QA (render-then-verify)
Overview
A finished figure can look fine and still be wrong: a series plotted from the wrong column,
a legend covering a data point, tick labels clipped at the axis edge, or a 150-dpi export that
pixelates in print. This skill is a render-then-verify QA pass — it re-renders the figure
and checks it against a concrete, measurable checklist before submission. It complements
the plotting skills (which make figures); it does not create plots.
When to Use This Skill
Use this skill when the user wants to:
- Proof a finished figure for correctness and print-readiness before submission.
- Confirm the plot faithfully represents its data (every row/series accounted for).
- Catch overlapping or clipped labels, legends, and markers.
- Verify resolution / export settings (e.g. 300 dpi, vector where required).
Does NOT Trigger
| Scenario | Use instead |
|---|
| Create/plot the figure in the first place | alterlab-matplotlib / alterlab-seaborn / alterlab-plotly |
| Lay out a multi-panel publication figure | alterlab-scientific-viz |
| Draw a schematic / diagram | alterlab-scientific-schematics |
| Build an infographic | alterlab-infographics |
Core Capabilities
1. Data-fidelity check
Verify the rendered marks against every row of the source data: series count, point count
per series, min/max/range on each axis, and that categorical labels match the data's
categories. Flag any series plotted from the wrong column or a silently dropped subset.
2. Label floor-and-ceiling legibility
Check that tick labels, axis titles, and annotations are within legibility bounds — a font
floor (nothing below the journal's minimum pt at final size) and a ceiling (titles not
so large they crowd the panel) — and that nothing is clipped at the axis boundary.
3. Bounding-box collision detection
Compute the bounding boxes of text, legend, and markers and detect overlaps/collisions
(legend over data, colliding labels, out-of-axes text). Report each collision with its
location so it can be nudged.
