If an RNG biases out a used number, that means that, whenever the game generates a random number, that number will not be used again until a certain point, for example, when every other number gets used.

So for example, in a standard deck of cards, you have a 1/52 chance of drawing the Ace of Spades. You draw a card, it's the 3 of Clubs, and then put it in the discard pile. You've therefore biased out the 3 of Clubs, increasing your chance on the next draw of getting the Ace of Spades to 1/51.

This principle is how a pity counter would work, but it's not how the drop rate for girls does - it's more like rolling dice - there's 1,000 faces and on exactly one face, you get a girl. There's no bias towards any given face - you could conceivably roll the same number 1,000 times in a row and thus "win" a thousand Buff T-shirts in a row and nothing prevents the RNG from doing that.