The summary (and article) is also missing all those kinds of relevant data :) I thought I had added a comment to my post that I was ignoring things like suitability ranking, but apparently I didn't actually type it. However, it should be unsurprising that my calculations don't include data that's not available, no?

True. I thought I had added a note that I'm explicitly ignoring anything other than raw numbers, like suitability ranking, but apparently I forgot to actually type it.

73% of 130ish applicants asian, so 94.9ish asian applicants. Let's call it 94, and 36 non-asian. Then we have 17 non-asian hires and 4 asian hires. So the question is, using random sampling without replacement from that pool of 94A and 36N, what're the odds of getting 17N and 4A?

I'm using the formulas from http://people.wku.edu/david.ne...
So in our case the figure we want is C(36,17)*C(94,4)/C(130,21) - ways to choose 17N and 4A divided by ways to choose just any 21.
Conveniently, you can google "36 choose 17" to do the calculation, but the formula for N choose K is N! / (K! * (N-K)!)
36 choose 17 is 8597496600
94 choose 4 is 3049501
so C(36,17)*C(94,4) is about 2.6218074e+16 (yeah, we lost a little precision but we can live without it)
130 choose 21 is 8.7664606e+23
So our final result is 2.62e16 / 8.77e23, or 1/3.35e7

one in 33 million isn't one in a billion, so I would also be interested in seeing their math, but the conclusion that the selection was biased seems to be fairly well supported.

cap equal to his own costs or $1000, whichever is more (so doing things to minimize your own costs, like representing yourself, doesn't trivialize losing)

exactly. That means they can't be requested to remove the information, so it's still going to be available.

we should really round that up to a power of 2, so make it 512 bits.

Then we can assign addresses to the universes next door too, once we can get there.

both of which are likely to be common methods for covering up a straw purchase. You're basically depending on granny seeming harmless and credible either way.

because then they wouldn't be able to apply it to images that don't contain actual children.

I think you're interpreting it as "the catholic church bought (land let alone people) in the BC era" but I think it was meant to be interpreted as "(the catholic church bought land) let alone (people in the BC era)"

licensing is intrinsically artificial scarcity, so even if it's stupid-ass, it counts.

the problem is that they can spend more on lawyers to make your actions "bad faith" while theirs are still "good faith". I think a more effective route would be to hold this up as an example of their automated scanner being so bad that trusting it is no longer "good faith" but simple blind stupidity, with an eye to forcing them to expose the scanner internals and actually fix the damn thing.

the difference (in my mind a significant one) is that you didn't leave the customer on the hook to pay for something they never wanted.

I was thinking "our contracts specify that if something goes wrong we have to fix it in under 30 minutes or we pay a lot of money. The only way to do that is to have the spare parts and people on site." but I like your method :)

nope, one party's images were copied illegally onto a server and they're complaining that someone else published a pointer to the server with the illegal copies.

Your rules seem to boil down to "embedding is bad, pointing is okay", which doesn't seem unreasonable. But in this case the judge is dinging the news site for pointing to a pointer, so now we're back to "how do we make sure our pointers are safe".