The session-below-threshold test measured lifetime from adoption, so it also absorbed the handoff to the test thread and a synchronous request before the close was scheduled, leaving too little slack for a loaded runner and risking a spurious stable reading. Scale the threshold and the hold together so the ratio the test depends on is unchanged while the absolute margin grows. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>