That order is no good. You’d want to check what other people have done in that area before you spend weeks figuring out the details of your idea, let alone produce a decent paper and publish on the arxiv. Not to mention that the arxiv is for preprints, not half-baked ideas.

Step 3 is absolutely correct.