ranges

The opt-in ranges rule turns a hyphen between two numbers into the locale's range dash, with the guards that protect dates, phone numbers and decimals.

Updated 21 October 2026 · spec 1.12.0

ranges rewrites the hyphen in a numeric range such as 1914-1918 or 5-10 days into the dash the locale prescribes: an en dash in English, German, Swedish, Finnish, Polish and Czech, an em dash in Russian and Ukrainian. When the dash is tight (no spaces around it), the rule also binds it to the digits with zero-width word joiners so the range cannot break across lines.

This rule is off by default. Without it, 5-10, Figure 5-10 and 9-11 come through byte for byte unchanged. Turn it on with rules: { ranges: true }.

Why it is opt-in #

A hyphen between two numbers is ambiguous without the word in front of it. takes 5-10 days is a range. Figure 5-10 is a compound label (figure 10 of chapter 5), and 9-11 can be a date or a proper noun. Telling them apart needs a per-locale list of label words, and the rules in polytypo scan code points, not words. So the rule makes the choice yours: with ranges on, a compound label that has the same shape as a range is converted too.

Notice that nothing changed. The same text with the rule on:

All three converted, including Figure 5-10. That is the accepted trade-off.

How it works #

A hyphen (or a doubled hyphen, or a dash already there) between two numbers is a range candidate. The rule then checks five guards. All of them must pass; otherwise the text is left exactly as written, and the dashes rule does not touch it either.

1. No letter before the number. MP3-4 and H2-2 are rejected.
2. No chain. The numbers must not sit next to another dash. 2026-08-15, 978-3-16-148410-0 and 212-555-1234 are rejected.
3. Not a decimal or a path. The number must not follow ., , or /, and must not be followed by /. 1.5-2.5 and 01/02-03/04 are rejected.
4. Digit-run lengths. Both numbers have the same number of digits, or the left one has one digit and the right one two, with the right one not starting with 0. 1914-1918 and 5-10 pass. ISO 8859-1, 555-1234, 1-800, 2020-24, 10-7 and 9-05 do not.
5. Not decreasing. The left number must not be larger than the right one. 20-10 is rejected.

If the guards pass, the rule looks up dash.range for the locale and replaces the hyphen with that form. Spaces around the original hyphen are replaced along with it, so 5 - 10 also becomes a tight range in a tight locale.

Closed-up symbols #

A symbol written directly against both numbers is repeated on both sides, and the range is recognised. The symbol must be the same character on both sides: $ against $, % against %. The recognised symbols are the currency signs (including $, £, ¥ and the whole Currency Symbols block), %, ‰, ‱ and °. A symbol on one side only does not block the range, so $15-20 and 15-20% convert.

Unmatched symbols are declined: $15-€20 is a currency conversion, not a range. A symbol written with a space (15 kg-20 kg) is not a closed-up symbol, and the range is not recognised. Prefixes and suffixes of more than one character, such as US$15-$20 and 15°C-20°C, are not recognised either.

The invisible binding #

For tight forms the replacement is a word joiner (U+2060), the dash, and another word joiner. The joiners are zero-width, so the bound range looks exactly like an unbound one in any font, and in a diff the line shows a change with no visible difference. The site's examples make them visible; if you inspect output yourself, look at code points rather than glyphs. A range that is already in the target form (5–10) is left alone and not bound.

Examples #

English, en dash, tight:

The same input in Russian, where the range is an em dash:

German also uses a tight en dash, and a spaced hyphen between numbers is replaced too:

French has no verified range convention (dash.range is none), so the rule does nothing even when enabled:

Guards at work. Dates, phone numbers, ISO standards, shorter-right ranges, decimals and part names are untouched:

Closed-up symbols:

Declined shapes:

In markdown mode the rule runs on prose only:

Locale data #

The rule reads dash.range, one of em-tight, em-spaced, en-tight, en-spaced or none. The same field was always used for range styling; only the rule that reads it changed.

none means the locale has no verified range convention. The rule emits nothing there, and does not fall back to the parenthetical dash.

Options #

Enable the rule explicitly:

transform(text, { locale: 'en-US', rules: { ranges: true } })

ranges: false is the default. There is no other setting.

Edge cases #

• Digits are ASCII only. Arabic-Indic, Devanagari and fullwidth digits are not recognised.
• Decimal ranges (1.5-2.5) and abbreviated year ranges (2020-24) are left alone.
• Scores such as 5-0 are rejected because the left number is larger than the right, while 0-5 converts.
• ranges runs before dashes. A range candidate belongs to ranges alone, whether or not it is enabled: dashes never reinterprets a digit-flanked hyphen as a parenthetical dash. With both rules on, a - 5-10 becomes a—5–10 (the parenthetical dash first, then a bound range).
• A bare number with a space-separated unit is not a closed-up symbol. 15 kg–20 kg stays out of scope.
• Because a range candidate is wider than a plain digit flank, text such as $15 - $20 that dashes once converted by default is now left alone unless ranges is on, and is left alone in the none locales whatever you enable.

Related #

• dashes: the parenthetical dash, which never touches a range candidate.
• Rule order
• Invisible characters
• Options

Normative text: spec/rules/ranges.md.