Data Chart

Render measured data as static charts in the shared editorial figure language: gridless ink axes, restrained direct labels, and one accent family per page. Titles and legends are optional content, not a fixed page frame. SVG is the primary artifact (text preserved as text); inspect a PNG proof before claiming success.

Routing: chart or diagram?

Decide by the figure's content, not its name:

• Measured data - real numbers, many points, true scales (log axes, distributions, time series) - this skill.
• Structure - boxes, arrows, layers, pipelines, or a few illustrative values inside a larger schematic - the sibling technical-diagram skill. Use drawio-diagram only when native draw.io format is part of the request.
• Borderline (3-8 bars): real measurements that may change → this skill (the chart regenerates from data); a decorative sketch inside a diagram → technical-diagram.

Quick start

1. Read references/chart-language.md for the full rule set (anatomy, tokens, marks, dark mode).
2. Write a small script that imports the style module scripts/editorial_mpl.py: ed.use() → plot with family colors → ed.mono_ticks / ed.axis_label → optional ed.header → ed.save(fig, stem) (writes stem.svg + stem.png). The module reads the vendored canonical values from assets/editorial-tokens.json. scripts/example_chart.py is a working reference for both a line chart and grouped bars; copy its margin setup (subplots_adjust with top≈0.80 clears the header row).
3. Run it. No system matplotlib is assumed - use uv run --with matplotlib python <script>.py (add other deps the same way). findfont warnings about Inter/IBM Plex Mono are expected on machines without those fonts; fallbacks carry the voice and the SVG keeps the full stack.
4. Look at the PNG before finishing. Check: no clipped direct labels or tick text (widen margins, not the font), any required legend fits on one line, one accent family, and no gridlines or unrequested supporting text.