They used a pretty cool method to come up with a list of 10 million crash points while proving Stake could not have influenced any of them. First they chose a bitcoin block that wasn't mined yet and announced publicly that the hash of that block would be used to seed the result of the first game along with a non-public server seed. Then they hashed that hash for the second game, and then repeated the process 10 million times.
So you can take the hash of any crash game and hash it, and you've got the hash of the previous game. With just one seed, you can go back and figure out the result of every single game that has been played since the seeding event (July 2019 I think), but because the hashing function is one way - it's impossible to go the other way.