I have already written about how the question bank was built: the rules a question has to pass, how the wrong answers are designed and how difficulty is graded. This post picks up where that one stops. Once a question is approved, three more things have to go right before it is any good to you. It may need a picture, it needs to work in your language, and it has to land in the right match at the right moment.
Visual questions
Some knowledge is not in words. You might not remember a side character's name but you know their face the moment you see it. Visual questions test that: an image of a character, an object, a symbol or a scene, with four options underneath.
The images come from QuiJutsu's own library and from a community library that players help build. A community submission does not go live on its submitter's say-so. Other players review it, and an entry only counts as well known once it has either two confirmations backed by a source or four players recognising it. Anything that only appears in the manga is excluded, the same as for text questions. If someone thinks an entry is wrong, they can challenge it, and it gets one more community round before a curator makes the final call.
The rules for visual options are the same as for text: four of the same kind of thing, equally well known, no aliases. A picture of a main character never sits next to three names nobody remembers.
Translation without giving the answer away
QuiJutsu is played in English, Japanese, French, Spanish and Brazilian Portuguese. Translating trivia is harder than translating ordinary text, because a translation can be perfectly correct and still break the question.
The biggest decision was to stop letting translation guess names. Every character, place and technique has one agreed name per language, written down once and used everywhere. That is how a fan expects it. In Japanese, for example, Solo Leveling's characters keep their Korean names written in katakana, because that is how fans there know them, rather than the names from the Japanese localisation.
The trap I did not see coming was what I call a script leak. In Japanese or Chinese, some names correctly stay in Latin letters. If the correct answer is the only option written in Latin letters and the other three are in Japanese, you can pick the odd one out without reading a word. Every one of those cells is a correct translation, and the question still gives itself away. Death Note's L is the classic case. Now every question is checked for this, and where a proper local form exists, it is used so all four options look alike.
Translation also turned out to be a second review of the English. When two options that look different in English become the same word in another language, it means the English question had two correct answers all along. Attack on Titan's "Wall Sina" and "Wall Sheena" are two spellings of the same wall, and that was only caught because the translation collapsed them into one.
The game's menus and buttons are translated too, but anything only admins see stays in English. I would rather spend that effort on what players read.
How a match picks its eight questions
A match is eight questions, ten seconds each. Which eight you get is decided on the server, not in your browser, so nobody can tweak it.
In a standard match, visual questions sit at positions two, four and six, and text questions fill the rest. Within each kind, questions get harder as the match goes on. How hard depends on the players: when two players have different levels, the lower level sets the mix, so a newer player is never thrown straight into the deep end. Newer players see mostly easy questions; the strongest players get a hard one in the mix. Tournament matches use one fixed mix for everyone, so every bracket plays on the same footing.
The other rule is no repeats. The server first looks for questions neither player has seen. Among those, it prefers the ones that have been dealt least often across the whole game, so the entire bank gets played instead of the same favourites over and over.
I will be honest about why that rule is so strict. In October I found a bug in how visual questions were slotted into a match. Adding the three visuals cut the text questions down from eight to five, and it kept the five easiest. For a month, nothing above easy reached a match, and most of the approved bank had never been dealt once. The fix was the design above, and since then every approved question has a fair chance of showing up.
Anime only, always
One rule runs through all of this. Every question, text or visual, in any language, has to be answerable from the anime as it aired. No manga panels, no light novel exposition, no fan theories. If you only watch the anime, you should never lose a point for it.
When something slips through
With thousands of questions in five languages, some mistakes will get past all of this. Every question can be reported from inside a match, and I read those reports. A bad translation, a wrong answer, an image that is not who it claims to be: report it and it gets looked at.
Want to see how it all comes together? Jump into a match.