A mathy construction like in the article is probably important for the full conjecture, but isn't this concrete case just an instance of the bin-covering problem? Your discrete items are the log of each prime factor (included according to its multiplicity), set the lower threshold to log(100k), and if you get any solution with 300k or more factors, you can redistribute the extra factors arbitrarily.

I like how Tao put out the call, and you answered it. Very cool! I was waiting for someone on here with engineering and math chops to attempt devising a more efficient approach teased in that article. Super cool to see you do it! You might be co-author on his paper, I think that's what he said, hahaha ! :)

HN thread for the Tao piece https://news.ycombinator.com/item?id=43506238

Hey! This guy is using the fancy new treelist [1] package which implements RRB trees. These are super cool data structures: a cache-friendly, multiway tree structure to store a sorted collection with O(log_{32} n) read, functional insert/update/append/prepend. Really a fantastic data structure for functional programming.

[1]: https://docs.racket-lang.org/reference/treelist.html