PubTrivia sells in 24 currencies and speaks one language. That combination produces a specific and slightly strange localisation posture, and I had never measured it, so here it is in numbers.
The short version: money is localised with real care, words are not localised at all, and the request header that exists to tell a server what language you want appears nowhere in the codebase.
Zero occurrences of Accept-Language
$ grep -rni "accept-language" --include="*.ts" --include="*.tsx" . | grep -v node_modules $ Nothing. Not in the proxy, not in a page, not in a test. The only request header that changes what you are served is the country header the edge attaches, which picks your currency. What language you asked for is not consulted, which is defensible because there is exactly one language to serve, but it means the decision is implicit rather than written down anywhere.
The HTML agrees with itself about this, at least. Across all 111 URLs in the sitemap:
$ grep -o 'lang="[^"]*"' ptpages/*.html | sed 's/.*://' | sort | uniq -c 111 lang="en" One lang attribute per document, the same value on every one, zero hreflang links anywhere, and no dir attribute on any element of any page.
Twenty-four currencies, 24 pinned locales
Each currency in the table carries its own formatting locale, chosen so that separators and symbol placement look native. Calling our own formatter with the real £30 Pro constant gives the exact strings a visitor is shown:
gbp en-GB "£30" usd en-US "$42.99" eur en-IE "€36.99" cad en-CA "$58.99" chf de-CH "CHF 34.99" mxn es-MX "$719.99" sek sv-SE "401,99 kr" brl pt-BR "R$ 220,99" nok nb-NO "398,99 kr" aud en-AU "$59.99" dkk da-DK "272,99 kr." nzd en-NZ "$71.99" isk is-IS "5.190 kr." inr en-IN "₹4,044.99" pln pl-PL "157,99 zł" jpy ja-JP "¥6,736" czk cs-CZ "883,99 Kč" sgd en-SG "$54.99" huf hu-HU "13 208 Ft" hkd en-HK "HK$331.99" ron ro-RO "191,99 RON" zar en-ZA "R 684,99" try tr-TR "₺2.026,99" ils he-IL "124.99 ₪" Twenty-four currencies produce 16 distinct symbols. Six of them print a bare dollar sign: USD, CAD, MXN, AUD, NZD and SGD, so $54.99 and $59.99 and $719.99 are three different currencies written the same way. Four print kr: the Swedish, Norwegian, Danish and Icelandic kronor, two of them with a trailing full stop and two without. That ambiguity is why every price carries a line naming the code underneath it, and the function that writes that line says so in a comment: a dollar sign alone could be four different currencies.
Three currencies are written with no decimal places at all: the króna, the forint and the yen. That list is hardcoded rather than read from the platform, and the reason is worth quoting because it is a hydration bug waiting to happen:
Fixed here rather than read from Intl, because the price is rendered twice, on the server and again in the browser on hydration, and JavaScript engines disagree about forints: V8 writes them without decimals, while engines that follow ISO 4217 strictly give them two. A price that depends on the engine is a price that can differ from the charge. The euro is the one currency that cannot have its own locale
Fifty-eight country codes map to a currency, and 27 of them map to the euro: the 21 euro area members plus Andorra, Monaco, San Marino, the Vatican, Montenegro and Kosovo. One currency, 27 countries, and the formatter takes exactly one locale. Ours is en-IE.
Here is what that costs, running the same amount through every euro area locale:
en-IE "€36.99" <- what we actually send de-DE "36,99 €" fr-FR "36,99 €" it-IT "36,99 €" es-ES "36,99 €" nl-NL "€ 36,99" pt-PT "36,99 €" fi-FI "36,99 €" el-GR "36,99 €" sk-SK "36,99 €" mt-MT "€36.99" de-AT "€ 36,99" Seventeen of the 19 euro locales I tested write that number differently from the way we write it. Sixteen of them put the symbol after the amount and use a comma for the decimal separator. Malta agrees with Ireland, which makes sense: they are the two euro countries whose main business language is English.
So a landlord in Lyon is shown €36.99 where every price in their country is written 36,99 €. It is readable, it is unambiguous, and it is subtly foreign in the way a translated product usually is. Picking Ireland is not wrong, it is just the only choice available to a single locale field, and the field being single is the actual design decision.
One honest hedge: the Hebrew string above is produced by the formatter locally and includes two Unicode right-to-left marks around the shekel sign, which I have never seen in a live document because I cannot geolocate myself to Israel. On a page with no dir attribute anywhere, those marks are the only bidirectional hint in the markup.
The dates and the scores do not follow the same rule
Money is pinned per currency. Everything else in the codebase splits into two habits:
| Call | Count | Locale |
|---|---|---|
toLocaleDateString('en-GB', ...) | 5 | pinned, British |
toLocaleDateString('en-GB', { timeZone: 'UTC' }) | 1 | pinned, and the only one that pins the zone too |
toLocaleString() with no argument | 10 | whatever locale the renderer has |
The last row is the interesting one. Ten calls format a number with no locale argument, which means on the server they take Node's default and in the browser they take the visitor's. A German visitor's browser would write 1.234 where the server wrote 1,234, and the only reason this has never produced a visible hydration warning is that the numbers in question are small enough to carry no separator at all most of the time.
The copy still has pounds in it
The two pages that read your country, the homepage and the pricing page, are the only two that can show you your own currency. The other 109 are prerendered once and served from the edge to everybody, so any money in their prose is in pounds forever. Across the full crawl:
- 10 of 111 pages contain a
£in the body, 106 pound signs in total - 7 of them name
£30specifically, including four comparison pages and two venue pages - the structured data on the pricing page carries
"priceCurrency":"GBP"with"price":"30.00"and"price":"100.00"even when the visible price on that same page is€36.99
That last line is deliberate and I have written about it before: the structured data is in the currency the plans are defined in so that search results do not disagree with the plan. The result is still a page that says two true things in two currencies at once.
And then there is our own profit calculator, a free tool with no sign-up, which offers a currency selector with three options:
const CURRENCIES = ["£", "€", "$"] as const Three symbols against the site's 24 currencies, and the formatting is toLocaleString("en-GB") regardless of which one you pick, so choosing euros gives you €1,234.56 with British separators. For a back of an envelope tool that is a fine trade. It is also the clearest illustration of the posture: the money the business charges is localised properly, and the money the customer plays with is not.
What the measurement cannot show
Every live observation came from one IP address in central Europe, so the only two currency renderings I have actually seen in a response are GBP and EUR. The other 22 strings come from calling the site's own formatting function locally, which is the same code path the page uses, but they are computed rather than observed. I also cannot tell you whether any real customer has ever been confused by €36.99, because nothing here measures behaviour, only markup.
Check it
# one language, 111 times curl -s https://pub-trivia.app/sitemap.xml | grep -o '<loc>[^<]*' | wc -l curl -s https://pub-trivia.app/faq | grep -o '<html[^>]*>' # the header that changes nothing curl -s -H "Accept-Language: de-DE,de;q=0.9" https://pub-trivia.app/pricing | grep -o '€[0-9.]*' | head -1 curl -s -H "Accept-Language: ja-JP" https://pub-trivia.app/pricing | grep -o '€[0-9.]*' | head -1 Both of those last two return the same figure as a request with no language header at all. Then open the pricing page and a page with pounds baked into the prose, such as hiring a quizmaster, in the same browser session, and watch one page agree with your country while the other quietly does not.