Telling from my past long challenge experiences, I always tried to think what I am missing ( is it knowledge of some theory, or is it not being able to observe some important fact).
Like for example, the first time I learned segment tree. From problem : ZOMCAV
The problem was basically something like this : You had to make many range queries. But I didn’t know how to do range queries in under O(N) time. So I looked up resources that can help me understand how to do this in under O(N) time.
(PS: that problem is solvable in O(N) time and with much easy implementation using Difference Array)
Another example is where I learnt mergesort tree : From problem : LAZER
After thinking for hours I concluded that ,I had to kind of get the Kth largest element in the range. And again I looked up everywhere how to implement that (found PBDS but it was too slow), then came to merge sort tree in the end.
(PS: again, that problem is solvable more easily by making some observations and using sweep-line event based approach)
One example where observation plays a role is : LAZERST
This one had me scratching my head for days. Ultimately I came to the conclusion that it’s not possible using any DS/Algo I had known of. So I started making some other observations and finally managed to solve it.
What helped me during these times is I had already gone through https://cp-algorithms.com/ and cp-handbook (CSES) theory once. Don’t misunderstand these points, I never actually understood how those algorithms worked at first. I just had kind of remembered what is possible to implement and what is not. And that kind of helped me a lot in distinguishing whether I am missing theory knowledge or some crucial observation.
So I think , you should pick up some good resource and go through all of it . No need to immediately try to understand all implementations , just keep in mind what is possible and what is not. That helped me a lot and I hope it will also help you a lot too.