75 websites · 87 font families · July 2026
Romanian writes ș and ț with a comma below. In 1987 the encoding standards handed it the cedilla instead, the hook used by French and Turkish, and in 1999 Unicode corrected the error and left a one-sentence instruction next to the old character.
Twenty-seven years later, 2,981 of the 26,861 letters in that family served by 75 Romanian-language pages are still the Turkish letter. In central government the proportion rises to 13.2 per cent. The Senate spells its own Constituţia with a cedilla eight times on its front page, Bucharest City Hall spells Bucureşti nine times, and the National Institute of Statistics writes Naţional with a cedilla twelve times.
The uncomfortable half of the measurement comes next. On most of those pages the error cannot be seen. The font repairs it, every time the page declares its language, and nobody has any way of noticing what the memory actually holds.
A cedilla grows out of the foot of the letter and stays attached to it, as in the French ç. The comma below is a separate mark, set under the letter without touching it. At newspaper size the difference is a few hundredths of a millimetre. At headline size you can see it from across the room.
For a program the distance between them has nothing to do with the drawing. They are two different numbers, two different characters, as unrelated as S and 5.
The same letter, the same size, two characters
the character 0219 should be used instead for Romanian
The annotation at U+015F, Unicode Character Database
The sentence has been in the standard since 1999, three lines from the character that 32 of the 36 central government sites still serve.
The error began as a saving of space in 1987, was confirmed by documentation in the nineties, corrected in 1999 and delivered to the user in 2007. Every step took longer than the one before it.
ISO/IEC 8859-2, known as Latin-2, divides 96 positions among the languages of Central Europe. Romanian is on the list, and for the sounds written ș and ț it is handed the positions occupied by S and T with cedilla, the hooked mark used by French and Turkish. Cedilla and comma end up in the same drawer because they look alike at small print sizes and because space is expensive.
Unicode 1.1 fixes Ş, ş, Ţ and ţ at U+015E, U+015F, U+0162 and U+0163. The same edition encodes the comma below at U+0326 as a combining mark, so the mark Romanian needs enters the standard at the outset. The letters carrying it do not, and the cedilla forms remain the only thing a system can type, store and transmit.
In the Unicode 1.1.5 documentation the four cedilla characters are described as suitable for both Turkish and Romanian. Font vendors and operating system vendors read exactly what suits them: one form, one drawing, no extra work. Romanian orthography, which has required a comma since the nineteenth century, has no standing to contest a technical note.
The revision of ISO/IEC 8859-2 passes without the amendment the Romanian side requested. The final text merely notes that the cedilla letters may be used as a substitute for the comma letters. That sentence turns the substitute into the official form for another decade of documents.
The block U+0218 through U+021B enters the standard: Ș, ș, Ț, ț, letters with a comma below. Next to the old U+015F, Unicode attaches a one-sentence annotation: the character U+0219 should be used instead for Romanian. In the same year the Romanian standard SR 13411 fixes the comma form for the national character set.
ISO/IEC 8859-16, Latin-10, includes the comma letters. It arrives at a point when the world is already moving to Unicode, so its practical reach is small. Its value lies elsewhere: from 2001 nobody can claim that the international standards have nowhere to put the letter.
At the request of the group revising the keyboard standard, the Institute of Linguistics of the Romanian Academy confirms in writing that the mark below S and T is a comma. The document closes the linguistic question. The technical one remains, and that depends on vendors.
SR 13392:2004 defines two layouts and ties them explicitly to the comma-below Unicode codes. A keyboard standard only takes effect if the operating system ships it, and the dominant operating system still did not.
Microsoft releases the European Union Expansion Font Update, adding the comma letters to Times New Roman, Arial, Trebuchet and Verdana for Windows XP SP2, Windows Server 2003 and Windows Vista. Vista makes the correct layout the default and renames the old one Romanian (Legacy). Romania had joined the European Union eight months earlier.
Twenty-seven years after the correction, 2,981 of the 26,861 letters in the Ș and Ț family served by 75 Romanian sites are still the cedilla form. In central government the proportion rises to 13.2 per cent. On most of those pages the error is invisible.
Three switches decide what the reader sees: the character in memory, the language the page declares for itself, and the font family. Only one of them has anything to do with what the page actually says.
The bench starts with “Şedinţe publice” misspelled twice, once with Ş and once with ţ, in a very widely used font, on a page that declares its language as Romanian. Look at both letters before you touch anything.
Şedinţe publice
What the memory holdsU+015EU+0065U+0064U+0069U+006EU+0163U+0065U+0070U+0075U+0062U+006CU+0069U+0063U+0065
The font repairs Ş and leaves ţ as it is. The same error, in the same word, hides on one letter and shows on the other. Eleven of the 84 families behave this way, Roboto among them.
Seventy of the 84 families that have the letter declare this substitution, but only 59 carry it through. Eleven repair S and leave T untouched, so a word like “Şedinţe” comes out half corrected. It is a well-meant typographic courtesy, and it is the reason the error survived a quarter of a century without inconveniencing anyone.
On 29 July 2026 I requested the front page of 82 websites publishing in Romanian once each and counted, in their text, how many letters of the Ș and Ț family arrive with a comma and how many with a cedilla. Seventy-five answered with a page.
The bar shows the proportion for each. To the right is the cedilla word that site repeats most often, which is usually the way the institution spells its own name.
commacedillacomma share · most frequent cedilla word
The word most often written with a cedilla is “şi”, the commonest conjunction in the language: it appears that way on 46 of the 75 pages and is the most repeated cedilla word on 29 of them. After it come the fittings of a website: “Anunţuri” in the menu of five institutions, “Citeşte” on the read-more button of four publications, “Bucureşti” and “Administraţia” in the masthead of Bucharest City Hall. On gov.ro, madr.ro and economie.gov.ro the very same pair turns up, “foloseşte” and “exprimaţi”, the text of a single cookie notice copied from one place to another. The wrong letter has left today’s article and settled in the labels nobody ever rewrites.
On the Senate front page, two of the 93 cedillas are legitimate: the name of Numan Kurtulmuş, a Turkish politician. One more foreign cedilla word exists in the whole sample, “Tolışi” on Wikipedia. The rest are Romanian words. The 1987 collision shows up twice, exactly as it was designed.
They appear in no total. The Chamber of Deputies, the High Court, the Ministry of Justice, the Ministry of Development and the legislative portal answered no request at all, browser included. The Trade Register rejected the request, and Casa Jurnalistului never finishes loading its content.
I downloaded the files Google Fonts serves for 87 families and read them from the inside: whether the letter exists, what it is built from, how far the mark falls, and whether the family carries a Romanian localisation.
The result is unexpectedly good. All 84 families that have the letter draw it with a real comma, built from the mark at U+0326, rather than a cedilla moved further down. The typographic layer of the problem is solved. What is not solved is the text.
Lato loses it completely: the latin-ext subset Google serves has 47 glyphs and no Romanian diacritics whatsoever. Righteous has ş with a cedilla and lacks ș with a comma, so it rewards precisely the error. Prata has no latin-ext subset, so all its diacritics come from another font.
The latin subset stops at U+00FF in every family measured. The letter arrives only from latin-ext, and a page that subsets fonts itself to save weight ends up without it.
The family declares a localised substitution for Romanian and swaps the cedilla glyph for the comma glyph when the page says the text is Romanian. Seventy families declare the feature, but only 59 cover all four letters. Eleven repair S and leave T untouched, Roboto among them, the most used font on the web: on a Romanian page “şi” corrects itself and “ţara” does not.
The distance between the foot of the letter and the top of the comma, in thousandths of the em. The median across the 84 families is 46, which is four and a half pixels in a 96-pixel headline. At 94 the comma starts reading as punctuation that wandered onto the line below.
Four of the families, Roboto among them, still build the letter from a glyph named uniF6C3. The name comes from the private use area of the Adobe glyph list of the nineties, put there because the list refused the Unicode assignment. To this day the correct comma travels under a code invented to route around an error.
The same pair of words, Constituția spelled correctly and Constituţia spelled wrongly, passes through three systems that judge it differently. All three answers were verified in the browser you are reading this in.
If the page declares its language as Romanian and the font carries a complete localisation, the two words look identical. Fifty-nine of the 84 families measured behave this way, Open Sans, Montserrat and Playfair Display among them. Another eleven repair some of the letters, so they leave a visible trace at random.
Romanian collation treats cedilla and comma as equivalent, at every level of sensitivity. A correctly ordered list of names gives no sign that half of it is spelled with the wrong letter.
String equality does not use collation, it uses characters. U+021B is not U+0163, and canonical decomposition does not bring them closer: one yields s followed by U+0326, the other s followed by U+0327. The compatibility form does not unify them either.
| check | result |
|---|---|
| 'și' === 'şi' | false |
| 'și'.normalize('NFD') | U+0073 U+0326 U+0069 |
| 'şi'.normalize('NFD') | U+0073 U+0327 U+0069 |
| NFKD('și') === NFKD('şi') | false |
| new Intl.Collator('ro').compare('și', 'şi') | 0 |
| new Intl.Collator('ro', { sensitivity: 'variant' }) | 0 |
| new Intl.Collator('tr').compare('și', 'şi') | -1 |
| /și/i.test('şi') | false |
A state keeps its records in the third layer. That is where exact filters live, and joins between two registers on the name field, and deduplication passes, and the regular expressions in forms, and the comparisons inside database migrations. A citizen whom one institution spelled with a cedilla and another with a comma is, to any system that puts the two together without collation, two people with the same name.
That is why the error held. It produces no ugly page, raises no alert and troubles no reader. It costs only in the places nobody looks at.