0.10.13 — not by language, by place

The third screenshot of the same evening. On the English entry card stood “3 MINUTY CZYTANIA” and “CZYTAJ →”, the quote was wrapped in Polish quotation marks, and the widget that tells a human from a bot said “Powodzenie!”.

The guard from the previous release let all four through. The reason: the words “czytaj”, “minuty” and “czytania” contain not one Polish letter, and a quotation mark is not a letter at all. Searching by language looks sensible right up to the moment you meet a sentence that does not give its language away.

The right question is a different one: **not “what language is this in”, but “where does it stand”.** Every text a reader sees belongs in the instance’s strings — whether it is Polish or English. The guard was rewritten: bare words as element content, attributes visible to the reader, and typographic quotation marks.

That immediately turned up the mirror image of the original fault: the not-found page and the error screen were hardcoded **in English**, which means a foreign language was standing on the Polish side of the site too. The same mistake in the other direction — and for three releases nobody saw it, because everyone was looking for Polish.

Quotation marks turned out to be a lesson of their own. Polish opens at the baseline, English at the top; it is not the same character and it cannot be left in the code. The anti-bot widget was given the instance’s language instead of guessing from the browser.