Aero StudioType scales that survive translation
A typographic system that works beautifully in English has a decent chance of falling apart in the third locale you ship. We have been on the wrong end of this enough times to have opinions.
Length is the obvious problem
German runs roughly 30% longer than English, and individual strings can be far worse — short UI labels are the danger, not body copy. A button reading "Save" becomes "Speichern"; "Settings" becomes "Einstellungen". Body text absorbs expansion gracefully because it reflows. Buttons, tabs and table headers do not.
The practical rule we use: design every fixed-width label with the assumption it may be twice as long, and decide in advance what happens when it is. Wrap, truncate with a tooltip, or let the container grow — pick one per component and write it down. The failure mode is not choosing, which means each engineer chooses differently.
Vertical metrics are the less obvious problem
This is the one that catches people. A line height that looks generous in Latin script can be visibly tight in Greek or Cyrillic, and genuinely broken in Thai, Devanagari or Arabic, where marks extend well above and below the nominal em box. Set a line height of 1.2 on a heading and Thai diacritics will collide with the line above.
We now treat 1.4 as a floor for anything that might carry a script with tall ascenders or stacked marks, and we set heading line heights in a way that leaves headroom rather than optically fitting them to Latin capitals. It costs a little tightness in English. It costs a lot less than a redesign.
Choose a typeface for its coverage, not its specimen
A typeface's marketing page shows Latin. Check what happens in the scripts you actually need, and check it in the weights you actually use — it is common for a family to have good Cyrillic in regular and nothing usable in the semibold you designed the whole interface around.
Where a family does not cover a script, you are into fallback territory, and fallbacks need designing. Matching x-height between the primary face and the fallback matters more than matching style. A mismatched x-height reads as a bug; a mismatched style reads as a different voice, which is survivable.
Do not build the scale from a ratio alone
Modular scales are appealing and produce sizes that are hard to defend individually. A 1.25 ratio from a 16px base gives you 12.8, 16, 20, 25, 31.25 — and now you are rounding, and your 25 and your 20 are doing very similar jobs while nothing sits comfortably between 31 and 25 for a page title.
We start from the ratio and then adjust by hand for the sizes that carry real work, keeping the total count low. Six sizes is usually enough. Nine means two of them are decorative and someone will use them inconsistently.
Test with real strings early
Pseudo-localisation — replacing English with an accented, expanded version of itself — catches layout problems in an afternoon and requires no translator. Run it before the design is signed off, not after the strings come back from the translation vendor with a launch date attached.
And get the actual longest strings from whoever owns the content. There is always one legal disclaimer or one product name that is far worse than anything you would have invented, and it is better to discover it during design.