Topic clustering at scale produces exactly this failure mode: a label that is broad enough to be useful for routing becomes too broad to be meaningful as editorial signal. The 143 records nominally covering open source AI contain a CPU upgrade troubleshooting thread [1], a Ktor backend repo setup question [2], a Docker container size explainer [17], and multiple YouTube shorts whose connection to open weights models is atmospheric at best [6][7][12]. The single record that does the work the whole cluster is supposed to do — a practitioner documenting…