Facing an observed pair of mean values <x> and <y> Jacob Cohen (1923-1998) knows two things: ____1_____Each one is liable to have random fluctuations, ____2_____The difference, or each one mean, should be scaled by a standard deviation. If fact he got that a large difference <x> - <y> for very dispersive data is much less important/significant than for a very close together one. So, he defines a ´suitable difference (my say) ´ as ______d= (<x> - <y>)/s (Cohen?s d). And because he had no idea how to define s: ascribing to the undisturbed y or to the treated x or both, left it to the reader?s expertise to chose. No matter, Wikipedia states he is famous by the find . . . supposedly he was the first psychologist to think about . . . No important . . . This and other avatars I come to know about NHST is that Statistics teachers did not emphasize that a single term change all things, namely _______W= (<x> - <y> - D)/s , When the shift D is considered as a constant W it follows the same distribution than the difference <x> - <y> and in consequence one is immediately liable to calculate a Confidence interval for D at a preset level. Cohen as an inventor . . .