Well that only happens because most scientists are pressured to get "new" ideas working for reputation reasons (which is not at all excusable) or because (utterly) big money consortiums want to hide the truth for concurrence or liability reasons. Not even saying that most of them know that sometimes several years or decades are needed to double check...
Anyway...
Much of this has to do with statistical studies. Based on statistical studies, this year coffee can be good for you, while next year coffee can be bad for you. This is not logical and will not appear, reproducible even though the data will be processed with logic. It has to do with the statistical foundation premise. If you have two fuzzy dice data points, curves can have a range of slopes and still touch both points because each point has an area instead of being just a point. This allows slants to logic.
Say we have a study that demonstrates X. Once this is accepted and published, X becomes a premise for the next level of logic; extrapolate from X to get Y. If this follows and Y is accepted, Y becomes a new premise for further logic; cite previous studies, to get to Z.
Say X changes due to new data and later studies, this impacts both Y and Z, since a foundation premise has changed on which both are built. Often Y and Z will remain, since they were already published. Sometimes you need to trace things back to the beginning, to make sure the chain was not broken.