That's actually super easy : obscure data sheets for embedded devices, private R&D that companies everywhere hoard, pre-internet content that nobody cares enou

That's actually super easy : obscure data sheets for embedded devices, private R&D that companies everywhere hoard, pre-internet content that nobody cares enough about to digitize, any content that rotted or was taken down by the original hosters before being mirrored, things that (dominant) search engines don't archive, things that was left behind by the SEO arms race in a priorty bucket so low it might as well not exist for any reasonable amount of effort, and etc... . Things get even worse if you consider legal questions e.g. the SciHub saga or the YouTube copyrights fiasco I'm sure there is more if one thinks carefully enough, the internet is wonderful memory, but it's _memory_, it _will_ fail you at some point or another. Distribution merely decreases the probability of faliure, but any arbitarily small probability can be made arbitarily large by multiplying by enough population size.