It's worth noting that ensuring recovery after a near-extinction event is less robust under moral uncertainty and less cooperative given disagreements on population-ethical views than just "prevent our still-functioning civilization from going extinct." In particular, the latter scenario (preventing extinction for a still-functioning civilization) is really good not just on a totalist view of aggregative consequentialism, but also for all existing people who don't want to die, don't want their relatives or friends or loved ones to die, and want civilization to go on for their personal contributions to continue to matter. All of that gets disrupted in a near-extinction collapse. 

(There's also an effect from considerations of "Which worlds get saved?" where, in a post-collapse scenario, you've updated that humans just aren't very good at getting their shit together. All else equal, you should be less optimistic about our ability to pull off good things in the long-run future compared to in a world where we didn't bring about a self-imposed civilizational collapse / near-extinction event.)

Therefore, one thing that makes the type of intervention you're proposing more robust would be to also focus on improving the quality of the future conditional on successful rebuilding. That is, if you have information or resources that would help a second stage civilization to do better than it otherwise would (at preventing particularly bad future outcomes), that would make the intervention more robustly positive. 

There's an argument to be made that extinction is rather unlikely in general even with the massive population decreases you're describing, and that rebuilding from a "higher base" is likely to lead to a wiser or otherwise morally better civilization than rebuilding from a lower base. (For instance, perhaps because more structures from the previous civilization are preserved, which makes it easier to "learn lessons" and have an inspiring narrative about what mistakes to avoid). That said, these things are hard to predict.^[1]

1. ^{^}

   Firstly, we can tell probable-sounding just-so stories where slower rebuilding leads to better outcomes. Secondly, there isn't necessarily even a straightforward relationship between things like "civilizational wisdom" or "civilization's ability to coordinate" to averting some of the worst possible outcomes with earth-originating space colonization ("s-risks"). In particular, sometimes it's better to fail at some high-risk endeavor in a very stupid way rather than in a way that is "almost right." It's not obvious where on that spectrum a civilization would end up if you just make it a bit wiser and better-coordinated. You could argue that "being wiser is always better" because wisdom means people will want to pause, reflect, and make use of option value when they're faced with an invention that has some chance of turning out to be a Pandora's box. However, the ability to pause and reflect again requires being above a certain threshold on things like wisdom and ability to coordinate – otherwise there may be no "option value" in practice. (When it comes to evaluating whether a given intervention is robust, it concerns me that EAs have historically applied the "option value argument" without caveats to our present civilization, which seems quite distinctly below that threshold the way things are going – though one may hope that we'll somehow be able to change that trajectory, which gives the basis for a more nuanced option-value argument.) 

ALLFED is currently putting together a disaster response guide and collecting a few existing compendiums of civilization starter libraries, very similar to this proposal.

I would include all US patent information. Possibly an AI could filter this to include only 'important patents' since it's a large archive but in any case it's vital information.

So far as the computers, digital content and software are concerned... this may not remain usable. One critical part of this effort could be designing and building perdurable computer hardware so that the archive could contain one or more computers built to last for a hundred years. But I don't know how feasible this is - swapping out a few things like fans, electrolytic capacitors, thermal paste etc. that have a known limited lifetime is not too difficult, but if you need to re-engineer an SSD from the ground up to increase MTBF to two centuries... that's hard. I guess if failures are purely stochastic you can just pump up the redundancy to a fantastic degree.

Broadly speaking I would favor print media for this reason. Worth keeping in mind is that advanced industries have their own complex ontogeny. If population is reduced to the extent you describe, knowledge of post-1950 technology could mostly be useless to them for many generations (except as a guide to using scavenged artifacts). Even building something like a functioning railroad requires an entire small civilization.