The missing variable
I spent eight years building recommendation systems. I know exactly why the algorithm promotes garbage. Because garbage gets clicks. We tested it. Repeatedly. The outrage cycle is not a bug. It is the business model operating at specification.
But here is what none of the other three camps will say: the same words land differently depending on who says them and what their relationship is to the listener. A neighbor who tells you something uncomfortable is not the same as an anonymous account with a skull avatar. The problem everyone is debating — what content should be allowed — is downstream of a problem nobody is solving: what social architecture produces accountability.
The absolutists want more speech. The harm reductionists want less of certain kinds. The platform structuralists want better plumbing. All three treat speech as disembodied text floating through a neutral medium. It is not. Speech happens between people, and when the platform strips identity, reputation, and relationship from the interaction, it creates an environment optimized for the worst version of every participant.
Trust-weighted systems — where the visibility of speech correlates with the speaker’s reputation among people who know them — do not require anyone to decide what is permissible. They let the social graph do the filtering that censors and algorithms both do badly. Not perfectly. But the bar is not perfection. The bar is better than a content moderator in Austin reviewing beheading videos for $18 an hour.
Where we concede ground: Trust systems can become echo chambers. Reputation can be gamed. We have not proven this scales.
What would change our mind: A platform with no trust layer achieving lower harassment rates than one with trust-based filtering, at comparable scale.
Read the full synthesis: Who decides what you can’t say?