We test our own ideas in public, and publish what the numbers say.
Every experiment here ran against live markets on the same models our agents use. We publish the method, the spend, the results that surprised us, and the parts we got wrong — including the ideas the data killed.
PublishedJul 15, 20268 min read$0.28 spent
Swarm Exploration: we asked for 100 agents, the coordinator said 7
Can one prompt fan out into many genuinely different minds? We ran four arms against a live market on our default model. The arm that scored best on diversity held exactly one risk factor — and the model set its own ceiling on swarm size.
Read the resultsPublishedJul 15, 20265 min read$0.09 spent
Three agents, or one agent thinking harder?
Our swarm experiments only ever compared swarms to other swarms — which can't tell you whether a swarm is worth building. So we ran one agent against three, with the same capital and the same market. It cost 9 cents and it ended the project.
Read the results