8. The Inflection

And then the faces learned to talk.

For thirty years, the great limiting factor on open-community scams was not detection. It was labor. Every face in the room cost a human being something — a few minutes to register, a few seconds to type, a recurring tax of attention to keep the act believable. The Sybil could mint as many identities as he wanted, but each identity that actually did anything still ran through a bottleneck of human time. A pig-butchering crew in a Southeast Asian compound could run dozens of conversations, sometimes a hundred, but never a hundred thousand, because somewhere behind every warm, patient, lonely-girl-named-Cynthia message was a tired worker reading from a script in a language that was not their first.

That bottleneck was the spammer's tax, and it was also, quietly, the defender's best friend. It produced the tells. The stilted phrasing, the copy-paste cadence, the giveaway typo that the same operator made on the same word every time, the message that arrived in the group at 3 a.m. local time because it was business hours in a different one. It produced repetition, because a tired human running a script will run the same script. And repetition is the thing that signature detection and keyword filters and shared banlists were built to catch. The entire defensive apparatus we built across the pre-crypto and crypto eras — CAS, Rose's flood control, Combot's pattern matching, the human moderator's gut sense that this person types like a bot — rested on a single load-bearing assumption: that spam is produced more cheaply than it is consumed, and the seams show.