All tor does is provide onion addressing and strong authentication with increased the observation costs for passive observers. Anything beyond that is a user's myopic extension of crypto-is-a-panacea. Cryptography can provide protections for observability, it cannot provide protection against identifiability. Mixnets like remailers or modern traffic mixing like Nym attempt to address identifiability.
>I read off the address: "152 dot" and they repeated back "152 dot". "19 dot" "19 dot" and then they told me the rest of the network address. (I was stunned.) Tor is supposed to be anonymous.
It's hard to tell the author's genuine understanding of Tor is versus what is hyperbolic. How surprising is the quoted feat? IPv4 is roughly 2^32 in size. There's roughly 2.4 million tor users [1], so an observer would need ~22.2 bits to exactly identify them.
The author gives at-least (assuming uniformly random IP address distribution, which isn't the case) ~16 bits of entropy (log(255)/log(2)*2). Which leaves their counter party a 1/32 eg 2^(22.2-16) chance of guessing their IP. Unless your ip space is chock full of tor users, it's not surprising an exit node was able to autocomplete the rest of your ip address. PS If we know the country of their IP, we need at least ~15.3 bits and at most ~19.7 bits
The trick is akin to living on a street with a unique name and a retailer auto completing your address and customer details because you've ordered from them before and you gave them your street name.
If I was a Global Passive Adversary, I would be probing and rerouting traffic to see how systems responded: https://www.ndss-symposium.org/wp-content/uploads/2017/09/ND...
https://www.muckrock.com/foi/united-states-of-america-10/req...
[1] https://metrics.torproject.org/userstats-relay-table.html