Consensus

01 / The bots

Each microbot is an independent agent with its own temperament: stubbornness, openness to persuasion, confidence, influence, and how much it cares about winning. No two bots are identical, and each browser session seeds a fresh swarm, so no two visits unfold the same way.

02 / Speaking

Nobody schedules the speakers. Every bot builds an urge to talk from its own situation — being insulted by name, hearing an opposing view, losing, gloating while ahead, having just switched sides, or the clock running out. When the urge crosses a threshold, the bot speaks. Some moments are crowded; some fall silent.

03 / Voices

What they say is written live by a language model. At each speaking moment, one batched request describes every speaker — its side, whether it is winning, its stubbornness, its greed, its mood, the last thing said, and the seconds left on the clock — and the model returns one short, childish line per bot. If the model is slow or fails, the bots fall back on a small scripted vocabulary, so the debate never stops.

04 / Drift

Bots listen to every statement. Agreement, the speaker's influence and the listener's openness pull a bot's position; social pressure pushes it toward whichever side is currently ahead. Positions drift continuously, and bots visibly cross the floor when they change sides.

05 / Incentive

After each round the leading side is rewarded with influence, and influence buys speaking time. The apparatus rewards winning, not being right — the single control on the main page sets how hard this loop bites.

06 / The clock

Every debate is capped at two minutes, and the bots know it. They reference the time left, and the final thirty seconds turn the losing side desperate. A winner is declared when one side pulls clearly ahead — or when time expires and the lead, however thin, is called.