Panels, alignment, and the multi-panel figure
Lesson 5 of 6 · Figures
1 · Learn the move · Template pattern
Multi-panel figures fail in committee for layout reasons before anyone reads the data: panels that almost align, label letters in three styles, axis fonts that shrink panel by panel, a shared axis that is not quite shared. The cure is a template filled once per figure, before any plotting: grid dimensions, target width in millimeters at final size, panel letter style and position, which axes are shared and which are deliberately not, font sizes as numbers, spacing. Hand the filled template to the model and it writes one assembly script where every panel inherits the same constants - alignment by construction instead of by nudging. When a reviewer wants panel C twice as tall, you change one template line and rerun, and nothing else drifts.
Here is my figure template, filled: GRID: [rows x cols]. FINAL WIDTH: [mm, per journal spec]. PANEL LABELS: [style, position]. SHARED AXES: [which, and which are intentionally independent]. FONTS: [sizes in pt at final size]. SPACING: [between panels]. PANELS: [A: what it shows, from which script/data ... ]. Write one assembly script where every value above is a named constant used by all panels. Flag any template field I left blank as [NEEDS: ...] rather than choosing for me.
2 · Your turn. You write the prompt
Figure 2 is four panels from four separate scripts - a time course, a dose-response, a bar summary, and a schematic placeholder - each currently exporting at whatever size it felt like. The journal wants 180 mm double-column width. Fill the template and get one assembly script that ends the nudging.
Remember: the AI sees only your prompt, not this page. If the situation isn't in your prompt, it doesn't exist.
Optional. These shape the output when you run your prompt below, not your score.