16. The Defender's Canon

Start with what works, because some of it does.

Before any of the tooling, picture the person the tooling exists for. It is a little after two in the morning, her time, and she is working the ban queue. She is a composite — I am assembling her from a dozen moderators I have known across the communities I worked in — but nothing in the scene is invented, because the scene happens every night, in some variation, in every large open room on the internet. The queue is what the bots could not decide on their own: forty-odd accounts the filters flagged but did not auto-ban, each one a thumbnail, a handle, a join time, and a first message. She moves through them at a clip that would look careless if you did not know what she knows. This one is a fresh account that posted a link nine seconds after joining — gone, no deliberation. This one tripped the flood counter — gone. This one is on the shared banlist from some other room's bad night — gone before she finished reading the handle. And then she stops, because the next one is him again. Not the same account — the account is new, the name is new, the photo is a different stolen face. But the opening message has the same shape it has had for months, the same warm, patient, slightly-too-helpful first move, and she has banned this actor so many times she can recognize him the way you recognize a relative's footsteps on the stairs. She bans the account. She does not ban him, because there is no him to ban — there is only the account, and he has more. She knows that. She bans it anyway, closes the queue, and goes to bed, and by the time she wakes up the queue has refilled. She is not paid for this. Nobody in this chapter is paid much for any of it. The classical canon — every bot, every filter, every list this chapter surveys — is, at bottom, an attempt to give that woman some sleep.

It is fashionable, in a book whose whole argument is that classical defense has run out of road, to treat the classical defense as a museum piece — quaint, well-meaning, doomed. That framing is both unfair and analytically lazy. The moderation bots that grew up in the pre-crypto era and matured in the crypto era are still running, right now, in millions of rooms, and they are still catching an enormous amount of garbage. They have not stopped working. They have stopped scaling with the threat — which is a different and more specific failure, and one you can only understand if you first take seriously how much they still do.