FalsifyAI

FalsifyAI is an Executable Scientific Falsification Engine that turns a scientific hypothesis into an operational falsification workflow. Instead of helping defend an idea, FalsifyAI asks what observation could prove it wrong. It parses the hypothesis, performs adversarial critique, constructs the strongest null model, searches for a discriminating observable, permits numerical simulation only when a quantitative law is explicitly defined, designs an experiment, and returns a constrained falsification status. The system can return TESTABLE, BELOW_SENSITIVITY, NON_DISCRIMINATING, or NOT_YET_FALSIFIABLE. A quantitative Duffing benchmark demonstrates the complete executable path, while a speculative DQ-12 stress test demonstrates that FalsifyAI refuses to invent missing physics and fails closed when an operational or quantitative definition is absent.