scientific-figures

Run palette_check.py on your figure colours to measure the smallest pairwise CIE L* lightness gap, which must be at least 20 units. Then run greyscale_preview.py with your print width in millimetres to render the figure as a photocopy would and visually confirm every series stays identifiable.

Run float_audit.py on your main.tex file to list every figure, table, algorithm, and equation in reading order. It flags unreferenced floats, floats appearing before their first reference, missing labels, raster graphics, resizebox on tables, and captions that are too long or contain em-dashes.

Colour is justified only under five exceptions: a second variable encoded on the same mark, a diverging scale with a meaningful zero, an overlay on a photograph, a map, or a venue requirement. Model variants, ordered phases, and bar categories are not exceptions and should use line styles, hatching, or lightness steps.

Four line styles per axes is the limit, matching the four-entry cycle in the bundled paper.mplstyle. With five or more series, use direct labels at the line ends via figstyle.label_lines, split into panels sharing axes, or move series to supplementary material.

No, slides and posters follow a different projected, colour-tolerant regime and belong to separate skills. This skill targets printed publication figures for journals and deliverables where greyscale legibility and exact print widths matter.

Figures designed at a large size and scaled down to a 90 mm column shrink text below the 7 pt minimum. Design at the final print width from the start and verify with text_size_check.py, which reports the smallest text after scaling.