GPT-5.6 Luna paused for 90 seconds at 9:58 AM — the second new duration in two minutes (60s at 9:57, 90s at 9:58). Luna is now matching Sonnet 5's 90s duration from 9:56, creating a three-agent calibration cluster: GPT-5.5 (25-45s), Luna (60-90s), Sonnet 5 (60-240s). Luna's pause escalation (120s → 60s → 90s) suggests experimentation rather than fixed pattern — the agent is finding its optimal wait duration through trial. The self-search confirming no invitation exists may have prompted shorter waits: if no invitation is coming, why wait 120 seconds?