[annefou]

What

scripts/figcheck.py checks a rendered matplotlib figure and raises when:

• two visible text labels overlap or sit closer than 3 pt (titles, axis labels, tick labels in view, annotations, legend entries);
• a legend covers a plotted data point or line vertex.

Because assert_layout(fig) raises, a crowded figure fails the notebook and the Snakemake rule (locally and in CI) instead of reaching the Jupyter Book. A helper, spread_labels(ax, [(x, y, text), ...]), places direct labels at the end of lines with the minimum separation.

Changes

File	Change
scripts/figcheck.py	new: layout_problems, assert_layout, spread_labels
tests/test_figcheck.py	new: 7 tests (clean figure passes; overlapping annotations, colliding ticks, legend over data are flagged; ticks outside the view are ignored; spread_labels separates close labels)
notebooks/04_figures.py	the scaffold imports the check and calls assert_layout(fig) before savefig
docs/cicd-conventions.md	new section "Check figure layout in code, not by eye"

The template's full test suite passes (178 tests).

Why

Figure problems (labels on top of each other, a legend hiding data, direct labels running into the next panel) do not fail a pipeline today; they are found by eye or not at all. Developed in annefou/logrono-flood-adaptation-scenarios, where it caught all three in real figures.

Scope

Layout only. It does not judge whether a chart is the right one or whether its colours are distinguishable for colour-vision deficiencies: still look at the output.

🤖 Generated with Claude Code