This is also a serious problem in Computer Science. Anything involving data or empirical results is susceptible to these sorts of issues. So, machine learning, computer vision, performance benchmarks, all these areas are rife with the sorts of issues discussed above.
This is not new -- I recall a final exam in a geology class where the answers
came out of the "geosynclinal book" and an hour later I sat in a lecture hall
and listened to a plate tectonic talk which disclosed that the big oil companies
had used plate tectonics to identify target regions on the globe to explore over
the previous ten years.
The omission in all of this is the effective application of "deprecation" and retirement.
Consider the bogus paper on "Autism and measles-mumps-rubella vaccination"
and the associated controversy. Technical journals of the future must have
on line free errata. Even if the original paper is behind a pay wall! Those doing
modern work need to know what can be known about the foundations of prior
art that they depend on. Another internet issue is a glorious lack of dates.
This allows content to look new when it is a reissue with a new cover.
I recall mapping a fault in a region of central Arizona that was obviously
a normal fault but consulting the old maps from the USGS it was marked
as a thrust fault (no map in those old days lacked a thrust fault as was the fashion).
In Nevada other old map thrust fault structures are now clearly wrong in light of awareness of the
modern understanding of volcanic activity.
Computer science has update processes and within some bounds a structure
that allows the replacement of some foundation library when a bug has been
discovered and fixed but that does not cover the case of the bad decision
of null terminated strings. Buffer overflow and OBTW the world has more alphabets
than just ASCII.
The patent offices of the world are in serious need of a pile of Rosetta Stones.
And they have need of international digitization of computer science literature
to apply those Rosetta Stone translations to. Other literature searches for
prior art stumble over new language and news standards... A 7 layer OSI stack
in contrast to a TCP/IP model or the five layers of the VINES protocol stack.
RAID vendors all have their own tools and names for features, functions and devices.
RAID technology has science behind it but vendors like to differentiate or add value
so the names get changed for good or selfish reasons and subtle improvements
to the science might make it into a public scientific journal within a decade.
That is to say that some science has value and publication is delayed.
Some inventiveness is hidden in patent applications (locked and confidential) then
revision after revision attempts to lock in the flow of knowledge as the initial work
is not "quite right" but the .... well that is another topic to rant on.