Every prediction is a specific, checkable claim: that a stock will out- or under-perform a market benchmark over a set time window. When the window closes we compare what actually happened to what was predicted and record a hit or a miss.
Direction against the benchmark is the whole rule. A hit by a hair and a hit by a mile grade the same, and so do the misses. Every graded call is counted — the ones we got wrong are published beside the ones we got right.