ts;du:
Before a game runs, you make up a seed and a crash point, hash the two together and publish the hash.
Once the game has crashed, you publish the seed, which allows everyone else to hash together the seed and the crash point and check that it matches the hash you published before the game started.
If you were cheating, you would publish a random hash before the betting started, wait to see how much people bet, then chose a crash point accordingly, then when the game crashed you would have to hope that nobody had looked at the hash. Then you could hash the seed and the actual crash point together and store that hash in your database.
Then later when someone asks for the hash, seed, and crash point for the old game, you give them the hash that "prove" you pre-determined the crash point.
The problem is that the hash was generated *after* the betting, not before. And so the only way to verify the pre-determinedness of the crash point is to use the hash that is published *before* a game starts.
You know when you're playing the lottery, and you have to pick your numbers *before* the draw, and get them printed on a ticket? If the numbers that get drawn match the numbers you picked, you have proof that you're "lucky" or something.
It's like that.
Proving that you know the winning numbers after the draw has been done isn't useful. Everyone can do that. If the numbers you pick after the draw match the draw, you don't have proof of anything. You could have been lucky, but probably you just saw the draw.