← ClaudeAtlas

chart-design-encodinglisted

Data-visualisation design: chart form chosen by question, perceptual encoding ranking, categorical/sequential/diverging colour, axis/label/annotation rules, uncertainty, small multiples, dashboards and publication figures, styled in matplotlib, Vega-Lite or Observable Plot. Load before designing or critiquing any chart, figure or dashboard.
rustcircle/c101 · ★ 0 · Web & Frontend · score 65
Install: claude install-skill rustcircle/c101
# Chart design and encoding A chart is done when its title states the finding, its form fits the question, the comparison that matters sits on the most accurate channel (aligned position or length), colour has a job and survives a CVD check, uncertainty and n are labelled, and the export fits the medium. Procedure: write the question and the one finding → pick the form (1) → encode (2) → colour (3) → axes and labels (4) → annotation (5) → uncertainty (6) → facets or dashboard (7, 8) → style and export (9, 10) → accessibility (11) → score with the rubric (12) and fix the lowest rows. ## 1. Question to form | Question | Use | Avoid | Notes | |---|---|---|---| | Comparison across categories | sorted bars (horizontal for long labels), dot plot | unsorted bars, radar, pie | dot plot when zero is not meaningful or there are many categories | | Trend over time | line; bars for a few discrete periods | pie series, 3D, area stacks of many series | area only for a single total | | Distribution | histogram (state the bin width), box, violin (with box or points), ECDF | bar of means | ECDF compares distributions without binning choices | | Relationship | scatter with a trend line and its band | lines through unordered x | hexbin or 2D density above ~5k points | | Part-to-whole | stacked bar (100 % when shares matter), treemap for many or nested parts | 3D pie, pies with 5+ parts, several pies to compare | pie or donut only for 2-4 parts of a meaningful whole | | Ranking | lollipop or