I guess in some fields of science the effective dependency graph of academic work is very flat, and the true results get plucked and developed by industry (being true results it is actually possible to meet the higher reproducibility bar there). And the citations don't actually reflect the true dependencies, but some political/social graph instead. Too bad.
I think this gets to the major concern with Academia today, as it becomes somewhat of a self-reinforcing feedback loop. Curry citations with political savvy, get awarded grants due to citations and political savvy, show that you are productive due to citations, grants, and political savvy - earning yet more political capital.
This will probably become my go to explanation for why Academic CS research has largely become decoupled from industrial application and industrial research. While political savvy is important in a large corporation, eventually you need to produce results.
So many papers get published, few are read widely, and even fewer are replicated, they'll still get citations if the talk circuit is played right. Citations are what advance a scientist in their career, and anything that could be tossed off as an unfortunate statistical anomaly or error is unlikely to end a career.
In such a world, "optimal play" would be to intentionally or unintentionally P-hack, or just slightly embellish results such that the work is interesting to cite, but not interesting enough to replicate. People who do this will eventually move up ahead of everyone else, ultimately favoring incremental but bogus resuls.