The thresholds are 2.5 seconds for LCP, 200 milliseconds for INP and 0.1 for CLS, all assessed at the 75th percentile of real visits. They fail for different reasons — LCP is usually server or image work, INP is almost always JavaScript, CLS is layout — so treating them as one ticket normally means fixing the easy one and declaring victory.
As a ranking signal they are real but weak: page experience breaks ties between comparably relevant results rather than promoting a fast page over a more useful one. The business case is conversion, where the effect is much larger, and it concentrates on mobile, on long journeys and on sites currently rated poor. Our Core Web Vitals ROI simulator puts numbers on it.