scientific-visualization

Scientific Visualization

Adapted from K-Dense for EasyResearch's Figures-owned figures/ root, project-local dependencies, and platform-native shell policy.

Routing

Use this Skill for empirical data visualizations: lines, points, intervals, distributions, heatmaps, images, multi-panel comparisons, accessible color, or export inspection. Use the sibling drawio and drawio-academic-skills for architecture, workflow, roadmap, network, taxonomy, and replicated schematic diagrams.

Require accepted data/result paths and enough evidence to define variables, units, replicates, missingness, transformations, uncertainty, comparison groups, and target medium. Never invent or manually improve values. If the evidence or a consequential visual choice cannot be derived, return blocked through the caller without asking the user directly.

Artifact Boundary

All final plot artifacts stay under figures/:

figures/
  .venv/                    # created only when a data-plot task needs packages
  <figure-name>.py          # reproducible source
  <figure-name>.provenance.json
  <figure-name>.pdf|svg|png|tiff|html

Temporary files may use a clearly named subdirectory under figures/. Do not write plotting code or exports into experiment results/; read accepted results there and preserve their paths in provenance. An explicitly supplied existing layout may be followed only when the dispatch names it.

Integrity Rules

• Preserve raw tables/images, exclusions, missing codes, analysis source, normalization, binning, image adjustments, and random seeds.
• Do not hide inconvenient observations, connect missing observations, treat missing as zero, upsample as new detail, or tune axes/dual axes to exaggerate.
• Name the estimator and uncertainty type: SD, SE, CI, percentile, posterior, or another declared interval. State n and the independent unit.
• Show raw observations when feasible and keep jitter from obscuring values.
• Bars/areas normally include zero; a nonzero point/line axis needs context and disclosure. Avoid decorative 3D and misleading area/radius encodings.
• Record log transforms, handling of zero/negative values, smoothing/binning, normalization, and sensitivity choices.
• Keep compared panels on compatible scales unless a difference is explicit.
• Preserve original images; disclose whole-image processing and add valid scale bars where applicable.
• Verify current venue rules from an official public source immediately before final delivery. A dated profile is planning guidance, not certification.