[alexey-salmin]

Imagine you proudly present to the public your obviously flawed version of the algorithm even though the correct version is known for decades. If only you've read a single book on the topic.

If that's not overconfidence then it's hard to find what is.

[caminante]

You're just restating your initial claim and not addressing the issue I raised with the latter.

[alexey-salmin]

What is the issue? Not at all clear from your comment. You're saying above it's better than security by obscurity but it's beside the point.

[caminante]

> but it's beside the point

Why is it beside the point?

You haven't established their intent for gross negligence and give no charity to the fact this was 30 years ago (pre-Wikipedia and the search breadth we have today). Since then, people have continued to expose severe RNG design flaws in other systems designed by very smart people. It happens...

[alexey-salmin]

30 years ago it wasn't dark ages. Wikipedia didn't exist but books on probability theory and statistics did.

When you do a shuffling algorithm in a sensitive context (money or security), you have prove that it returns all the possible permutations with equal probability plus put lower bounds on the amount of entropy you need from the underlying RNG. If you're unable to prove it, you shouldn't move forward with the algorithm. Any irregularities in the output distribution can be exploited. This is textbook knowledge pioneered in early encryption works and perfected by the time WWII ended. Evidently the effort to prove correctness was never made.

Now the original article can indeed misrepresent or omit important facts. I'm definitely open to reconsider my conclusion if more facts become available. However "there was no Wikipedia" isn't one of them, it doesn't count as an excuse not to do your job properly.

If it turned out, for example, that "ASF Software" wasn't even aware that their shuffling algorithm was used to shuffle cards and just shipped it along with 200 other basic algorithms as a some sort of "better standard library", this would change the situation. However from the quick googling it seems that their product wasn't a standard library, it was specifically software for Texas Hold'em. This is a "you had one job" kind of situation.

> Since then, people have continued to expose severe RNG design flaws in other systems designed by very smart people. It happens...

Absolutely, but we're not talking frontiers of Computer Science here.

* If you seed your RNG with at most 86 million unique values, you get at most 86 million unique random sequences out of it.

* If your code should have M possible equiprobable outcomes, it has N equiprobable outcomes, and N doesn't divide M, you're in trouble.

[caminante]

> 30 years ago it wasn't dark ages...

I didn't say or imply books didn't exist. You can't credibly say it was as readily available, and I promise you that people are still making these mistakes, today.

> When you do a shuffling algorithm in a sensitive context (money or security), you have prove that it returns all the possible...If you're unable to prove it, you shouldn't move forward with the algorithm.

Ideally, of course! This is a really high standard that I'm afraid isn't enforced in a lot of commercial or even sensitive applications. 86 million permutations is probably good enough and even if someone was clever enough to synch clocks and narrow to 200k permutations, then I'm not convinced there was actually any harm.

Do you have any proof of harm?

And there are plenty of smart people in the 90s and beyond not realizing that relying a system clock to seed values is attackable. These guys, to their credit, patched their system by openly providing their algorithms.

Even if their clients had been harmed, they'd published the algorithm so that their "sophisticated" clients could audit the algorithm.

> I'm definitely open to reconsider my conclusion if more facts become available.

This is circular as you're taking the article's narrative at face value without getting any primary sources confirming gross negligence or "arrogance" as you imply.

[alexey-salmin]

> Ideally, of course! This is a really high standard that I'm afraid isn't enforced in a lot of commercial or even sensitive applications. 86 million permutations is probably good enough and even if someone was clever enough to synch clocks and narrow to 200k permutations, then I'm not convinced there was actually any harm.

Of course not, this is ridiculous. If your job is to shuffle the deck, shuffle it well. It's like doing a 80/20 coinflip and arguing that 50/50 is a "really high standard". And that for a company that sells bet-money-on-coinflips software.

If you don't know how to do it well -- read a book or use std::random_shuffle. Somehow Stepanov was able to do it right (assuming a good RandomNumberGenerator input) from the first try in 1993, without Wikipedia poor guy. And this wasn't even his main job, random_shuffle was one of a dozens of algorithms he envisioned and implemented for the STL.

> This is circular as you're taking the article's narrative at face value without getting any primary sources confirming gross negligence or "arrogance" as you imply.

I did some quick research and it seems that ASF Software had indeed developed the Planet Poker online platform. Which comes down to failing at your main job, I don't really see what other evidence you expect here?

I strongly believe that people in general and software engineers in particular should be held up to high standards. Finding excuses for how school-level math is too hard for them is condescending. It is disrespectful to the very people you're talking about.

If you say they couldn't even understand that N^N is not divisible by N! you basically say that they're mentally challenged. I on the contrary say that they most certainly would've been able to understand it if they made an effort -- which they didn't. So negligence.

[alexey-salmin]

UPDATE. I think I should also address this:

> if someone was clever enough to synch clocks and narrow to 200k permutations, then I'm not convinced there was actually any harm.

I don't think you understand the situation at all. In Hold'em in the end you see 7 cards: 2 in your hand and 5 on the table. That's 52x51x...x46 = 674B different sequences of open cards.

This means that by the time you see these cards you can know exactly which of the 200k permutations the engine had chosen for this hand. There's only one that precisely matches one of the 674 billions possible open cards combination that you observe.

In fact, by the time you see the flop (2+3 open cards, 311M variants), you know everyone else's cards.