First of all, this map[i] = vi(); manual initialization kinda prevents all the hacks, because to hack unordered_maps we want only very specific 'evil' keys to be inserted in the map, and that is not just 0, 1, 2, .... They should be (multiples of a speicifc prime) + (a constant).

I tried a version with that line erased, but there still is a problem. To make an unordered_map $$$\mathcal{O}(n^2)$$$, we also need at least a range of $$$\mathcal{O}(n^2)$$$ possible keys. In this code, the keys can only be in range $$$[0,199999]$$$, so we can never make it run as slow as $$$200000^2$$$.

There is a small trick to make it run in approximately $$$\mathcal{O}(n^{1.5})$$$ when the key range is $$$\mathcal{O}(n)$$$. To do this, we need an evil value that is close to $$$n^{0.5}$$$, which could be multiples of $$$541$$$ in case of C++20. We can insert $$$~370$$$ different multiples of $$$541$$$, and then repeat the 'worst' key for the rest. From my experiments, the 'worst' key is the first element inserted and it gets faster for the elements inserted later. So this unordered_map will take approximately $$$\mathcal{O}(n^{0.5})$$$ time for each of the $$$\mathcal{O}(n)$$$ accesses, resulting in $$$\mathcal{O}(n^{1.5})$$$.

So yeah, I tried it, but it took only 1405 ms because its constant is not really that high. Sadge. Maybe there could be a better strategy though...