Two ways a lagging detector sample could tear down a live managed-hook channel: A stale session read could roll back a newer one. After SessionStart(S1) then SessionStart(S2), a detector still reporting S1 drove a session-change that revoked the freshly verified S2 channel and rolled record.sessionID backwards. An announced rotation now retires the superseded session for every runtime, so a detector read of it is ignored, while a genuinely new session the hook has not announced still distrusts the channel until it re-announces. A transient process sample could revoke the launch generation. If one proc_listpids dropped the launcher while keeping the forked engine child, the launch root flipped to the child and read as a process replacement, permanently revoking the registration and its private settings file. The launch process is now kept while it is still a live ancestor of the identified process. Also restores the symlink cwd-comparison test dropped during the session takeover change. Claude-Session: https://claude.ai/code/session_01AQ4X2bx8DnU74wwThhV9D8