Name a science topic. A Pollinations model writes competing explanations, each with a prediction you can test. Then you run experiments: pick the explanation that matches what was observed, and watch the ones that don't get falsified. A survivor is not a proof โ it is an explanation that has not failed yet.
BYOP: approve a scoped key on the Pollinations consent screen (5 Pollen, 7 days). One build is one chat completion billed to your Pollen.
Everything below is generated from the model's reply and rendered by this page. No generated code runs: the predictions are drawn from a fixed set of shapes.
Experiments
Each experiment is one observation. The observation falsifies every explanation it contradicts, and that is permanent. Your pick decides whether you named the surviving explanations exactly โ it earns the point, not the falsification.
Generated on request by a language model. It can be wrong in a way no experiment here catches. Treat a survivor as a live hypothesis, not a fact.