You're misreading the table. The category rates are the rate of pulling a given type (kamihime, eidolon, weapon) within a given rarity tier. They're estimated from empirical data and used to generate the subsequent tables. This method is less subject to random biases because the category rates are much higher than the rate of obtaining a SSR, dampening the impact of variance. Also, we know the rarity rates exactly; they don't need to be estimated.
This misunderstanding happens frequently so I've revised the table to try and be clearer.
Also, I have no idea where you're getting the 50% chance of SSR kamihime from 3.3 rolls. My spreadsheet currently shows that a 50% rate is achieved at 4.8 10-rolls.