Engineering

Picture a table in Munich. Five friends, a Thursday night, a campaign two years deep. The whole session happens in German: the threats, the bad accents, the argument about whether the paladin’s oath covers arson. Then the recap arrives, and it reads like a BBC correspondent summarizing their game from across the channel. If you play D&D in German, French, Portuguese, or anything else that gets called a “non-English market” in a planning doc, you already know this experience. Every tool in this hobby assumes your table speaks English. Character builders, VTT interfaces, AI note-takers: English in, English out, and if your session happens to be in another language, you get whatever falls out of the machine. ...

Your table’s session has exactly one copy. Six people spent four hours making something that will never happen again, and the only record of it lives inside a bot that promised to be listening. If you’re a GM deciding whether to trust a recording bot with that, you should want to know how these things fail. So here is how ours fails, all thirteen ways, because the day we started paying for traffic we commissioned an adversarial audit of our own recording pipeline with one mandate: “from /record to /end we minimize the data we lose, always, and always package it up correctly.” ...