See for example any kind of documentation for any piece of hardware that's older than 5 years. The companies change their support pages structrue so often that it's incredibly hard to track anything down.

Cynic in me thinks they're doing it on purpose.

Those are particularly nasty because model numbers are often so arbitrary that a few years down the line searching for an old model number often becomes a near impossible quest of wading through stuff related to newer models with nearly the same model number.

It's a place where metadata can prove useful.

Sci-Hub works largely through an applied identifier, DOI (though URLs also work), which is a unique-per-article identifier present since the 1990s. Earlier gets a bit difficult.

Title, author, date, and publisher, you know, what your uni essays instructor always insisted on for footnotes, are pretty good identifiers, and will likely be meaningfully unique. Adding a publication location (as in a newspaper byline) also helps.

Ironically, many news organisations fail to include such information not only on individual articles but anywhere apparent on their website. You'll get city, and possibly county, but not a state or province or country. This in an age of, literally, worldwide access.

Otherwise, The Internet Archive does yoeman's work in creating a permanent record, where they're allowed to.