Producing a figure that survives publication

A figure is judged at its final size, in its final medium, by someone who has not seen your notebook. Design for that and check the exported file, not the preview.

Rules

1. Decide medium and size first. Email (raster, ~600–800 px wide body), slide (16:9, read from distance), journal column (one or two columns; the journal's own guide is authoritative and changes: check it live), web (responsive, vector). Render at that size; do not scale afterwards.
2. Text legible at final size. Axis and tick labels must be readable in the printed or emailed figure, not only on your monitor. If you need to zoom to read it, so will the reader.
3. Vector for line art, raster for email. SVG/PDF for print and web; PNG for email clients (no SVG support in many). Never upscale a raster to fake resolution.
4. Show the data when n is small. For fewer than about 15–20 points per group, plot every point (dot/strip plot) with the summary on top. A bar of the mean hides the distribution: very different datasets give identical bars.
5. Panels are consistent. Same axis limits for compared panels, same colour for the same entity across panels and across figures in one report, same units, panels labelled (A, B, …) and referenced in the caption.
6. Caption is self-contained. What is plotted, n and unit of replication, what intervals mean, data source and date, any transformation (log, smoothing, normalisation, exclusions). A reader of the figure alone can interpret it.
7. Do not trust defaults. Default tick density, colours, legend placement and line widths are someone else's compromise. Check each against the message.
8. Grayscale and colour-vision check. Series stay distinguishable in grayscale and under common colour-vision deficiencies (line style, marker, direct label). See color-and-accessibility.
9. Reproducible from a record. Keep the spec, data snapshot hash, code version and render settings next to the output so the exact figure can be regenerated and audited. A figure that cannot be regenerated cannot be corrected.
10. Inspect the exported file. Open the PNG/SVG/PDF that will be sent: clipping, overlapping labels, legend covering data, missing series, unreadable ticks, wrong aspect ratio.

ieeVisualizer support

Need	API
Final size	RenderRequest(width=, height=, units="in" or "px", dpi=); a PNG measures exactly request.size_px
Formats	png, svg, pdf from the matplotlib adapter
Themes	a registered theme name or name@version (RenderRequest(theme=) or spec.theme); themes.list_themes() lists them, built-ins include default, minimal, presentation
Determinism	same spec + library versions → byte-identical SVG and PDF
Provenance	ieeVisualizer.artifacts.export(spec, ["png", "svg"], writer) writes figure files, canonical spec.json, manifest.json (spec hash, dataset hashes, provenance, library versions, warnings)
Inspection	build_figure(spec) returns the Matplotlib figure and per-layer artists without exporting; read RenderResult.warnings
Legend placement	legend.position = "auto" moves the legend outside when it would cover data

Limits today: one plot per FigureSpec (render panels separately with equal axes.*.limits); SVG text is drawn as paths (svg.fonttype="path"), so text in the SVG is not editable or searchable; there is no subtitle field, so use the caption for measure, unit and period.

Good / bad

Bad: bar chart of mean task completion time for 4 participants per condition, ± unlabelled bars, exported at 300 px and scaled to a slide. Good: dot plot of all 8 times, mean marked, caption "n = 4 per condition; line = mean", rendered at slide size with the presentation theme.

Bad: email report embeds an SVG and a 2,400 px PNG with 6 pt ticks. Good: 1,200 × 750 px PNG at dpi=150 (rendered at 8 × 5 in), tick labels checked readable at 600 px display width.

Checklist

• Medium and final size chosen before rendering; rendered at that size.
• Text readable at final size (checked on the exported file).
• Format fits the medium (PNG for email, vector for print/web).
• Small-n groups show every point.
• Compared panels share limits, colours and units.
• Caption states n, interval meaning, source, date and transforms.
• Spec, data hash and render settings stored with the output (manifest).
• Exported file opened and inspected for clipping, overlap and missing series.

Adapted in part from ieeVisualizer's own docs (MIT), K-Dense scientific-visualization (MIT), Rougier et al. 2014 (CC0) and Weissgerber et al. 2015 (CC BY 4.0); changed: rewritten and condensed, ieeVisualizer mapping added. Sources and licences: SOURCES.md.