Data Visualization And Figures

Help the reader make the scientific comparison accurately and easily. Use this skill when visual choices remain unresolved: chart form, evidence layers, scales, color, image presentation, diagrams, layout or export. Mixed code-and-design requests qualify when they require that judgment. Pure syntax fixes, plumbing, unchanged reruns and implementation of settled specifications do not call for redesign.

Work from the reader's task

1. Identify the operation: compare amounts, distributions or effects; follow change; inspect relationships; understand a process; read image evidence; or look up exact values. Determine what each mark represents, the independent unit, relevant grouping/pairing and intended medium. Use supplied context before asking for missing details.
2. Preserve settled specifications and meaningful existing styling. For unspecified ordinary trajectories, start with solid lines, circles if markers help, a restrained palette and clear labels. These are editorial starting preferences. Use dashes, shapes or stronger colors when they carry meaning or solve an actual reading problem.
3. Read the narrowest relevant references below. Compare plausible designs, including what each reveals and hides. Resolve scientific meaning before cosmetic polish; do not let a generic rule override the task or evidence.
4. Make important comparisons visible. For continuous groups, consider center, spread, shape and observations together. A box/violin/point combination can help; sparse groups can omit density while retaining points and a named center, and dense groups need not show every dot. Do not confuse descriptive spread with uncertainty of an estimate.
5. Use hierarchy deliberately. Position, labels, facets and whitespace may already identify groups; extra color, dashes and shapes are not automatically necessary. Keep essential marks readable and equally important groups comparable in emphasis.
6. Inspect the actual rendered figure at the intended size. Check scientific decoding, labels, scale, contrast, important structures and relevant accessibility conditions. If the artifact is unavailable, keep exact appearance claims conditional. Report checks actually performed.

Read according to the decision

• references/chart-selection.md: reader operations, chart families, distribution diagnostics, composition and tables.
• references/statistics-and-uncertainty.md: distribution layers, estimands, pairing/nesting, intervals, smoothing and uncertainty displays.
• references/color-and-accessibility.md: palette meaning, restrained emphasis, identification cues and actual accessibility checks.
• references/specialized-figures.md: heatmaps, networks, genomic views, intersections, projections, maps and dense time series.
• references/biological-images.md: morphology versus intensity, scale, channels, annotations, representative images and analysis reporting.
• references/layout-and-typography.md: panel relationships, axes/transformations, labels, tables and final-size readability.
• references/conceptual-figures.md: graphical abstracts, mechanisms, spatial claims, arrows and known versus hypothesized relations.
• references/publication-technical-requirements.md: medium, formats, editable masters, reproducible revisions and export inspection.

Use checklists/publication-readiness.md for an actual readiness pass and examples/ for concrete applications. Consult references/handbook-access.md optionally when a fuller explanation or worked figure would help. The bundled reasoning works without the handbook; a text edition alone cannot establish visual quality.

Deliver the requested result

Give the figure, critique or recommendation the user asked for. Explain consequential choices through their purpose and tradeoff. A short request need not receive a full checklist report. When uncertainty matters, distinguish observed findings, reasonable assumptions and information still needed.

Reject invented metadata, hidden transformations, unsupported causality and unreported changes to the unit or estimand. Avoid automatic palette replacement, marker cycles, mean-only displays that conceal consequential structure, and complexity added solely to satisfy a rule. Prefer the least complicated design that preserves the important scientific information.