REPLY NETWORKS

Two people can post heavily in the same text channel without speaking directly to each other, so reply edges are more specific than co-presence. They still do not automatically prove friendship, conflict, intimacy, or hierarchy. Context remains rude enough to be necessary.

Richard's Corpse supplies the counterexample every network chart needs tattooed on its forehead. Rich appears central to cooking, streaming, and watch-session interactions while leaving almost no text in some sequences. Momo explicitly says Rich's voice startled her awake despite Rich contributing no corresponding text to that exchange. A text-only social graph can therefore undercount the person the room is literally reacting to.

Mugen's Lounge has zero surviving voice-channel text in the recovered package. The site will not convert that into zero VC use because the Richard's Corpse examples prove the category error: spoken presence and typed residue are different evidence streams.

A reaction records recognition. A reply records directed textual response. Co-presence records shared room-time when available. Voice residue records only the keyboard leakage from a synchronous space. Future relationship maps should preserve those layers rather than collapsing them into one authoritative-looking friendship score.

Alias changes, renamed accounts, deleted accounts, and the pooled Deleted User aggregate already make naive graphs dangerous. Voice adds another problem: missing text is structurally missing, not random. The prettiest possible lie would be a precise graph of only the behavior the archive happened to make countable.