The outrage spiral, drawn as a loop

Each camp runs the same logic: the more provocation we perceive, the more outraged we get; the more outraged we get, the more we post; and our posts are the next provocation the other side reacts to. That’s two balancing loops (B) in intent — each side trying to restore its sense of safety by pushing back.

But they share one variable: perceived provocation. A’s defense raises B’s provocation, B’s defense raises A’s, and the two balancing intentions chain into a single reinforcing loop (R). This is the Escalation archetype — the same structure as an arms race. Nobody is escalating on purpose; the structure escalates them.

In a face-to-face argument, the spiral eventually exhausts itself. Online there’s a third party that never tires. Outrage generates engagement, engagement is what the feed optimizes for, so algorithmic amplification turns up distribution on exactly the most provocative posts — which raises perceived provocation for both camps at once.

Run the simulation above and watch it: the amplification edge closes the loop into a faster, tighter spiral than the two camps could sustain on their own. The algorithm isn’t taking a side. It’s taking the spiral’s side, because the spiral is what feeds it.

The two instinctive moves are symmetric, and both are inside the loop. Fight back harder directly feeds perceived provocation — you’re pulling the reinforcing edge tighter. Wait for them to stop assumes the other camp controls the spiral, but no single actor does; the structure does.

Meanwhile polarization accumulates with a delay. Long after a given fight burns out, the gap it widened stays — so the next spark starts from a higher baseline. That’s why feeds feel angrier year over year even when no individual fight seems worse.

Escalation only breaks where the reinforcement is cut. Two places work.

• Unilateral de-escalation. One camp deliberately under-responds — refusing to treat the latest provocation as a trigger. It feels like losing, which is exactly why it’s rare and exactly why it works: it’s the one move that weakens the shared loop instead of feeding it.
• Change the amplification metric. The highest-leverage edge isn’t between the camps at all — it’s the algorithm’s goal. Optimize the feed for something other than raw engagement, and you sever the link from engagement to algorithmic amplification that turns a human squabble into a machine-accelerated spiral.

Arguing better inside the loop can’t win. Changing what the loop rewards can.