ps
tells different things in different stories or even within 1 story.
option 1: the shuffle the deck (like in a casino)
With 52 cards, there are more than 8*10^67 possibilities (different deck orders), thats an 8 with 67 zeroes!
Interstestingly, ps says this is 5*10^67; somehow they missed 30%.
option 2: take a not-shuffled deck (e.g. always 2-A * 4) and take a random card.
Although the result is the same, most likely they use option 2, as it is much more easy to implement.
Both methods use a random number generator RNG.
The basics the RNG must comply with is that 'any card can be the next'. Mathematically this result in at least the following two:
1) on average all cards are selected the same number of times
2) the average interval between selection of a specific card is 52, while the frequency distribution is flat
On the accreditation/verification:
1) a large part will be on 'is the procedure followed'
2) verify the randomness, is not 100% possible! Even ps claim some 100 billion hands, well there were 8*10^67 different decks, so 100*10^9 is a rediculously small sample. Analyzing any data set will result in a % conformity to the expectations, resulting in a accept or reject.
Luckily the RNG does not need to be 100%, as poker is also skill based.
As long as the RNG 'defect' is negligible compared to the skill differences, the RNG has no biased effect on the result of the game.