The “Odds of Detection” section is referencing a probability from 2016. In 2024, the total number of relays is now about 8,000: 4,344 are guard relays; while 2,459 are exit relays; with the rest being middle and bridge relays.
https://metrics.torproject.org/rs.html#aggregate/all
So, looking more closely at the “Odds of Detection” section, it assumes that an attacker only controls one “poisoned” Guard and Exit relay. Assuming that consensus weight has no bearing on the probability, and that the attacker is not providing a bridge relay instead, that means 1⁄4344 x 1⁄2459 = 1 in 10,681,896.
Assuming that Tor circuits are always forcefully updated every 10 minutes, plus the amount of guard and exit relays remaining exactly the same as it is now for 10,000 hours, along with all prior assumptions, that leaves us with 60,000 ÷ 10,681,896 = 0.0056169804; approximately half of one percent.
Multiple reasons, but it comes down to cost: an attacker would need to deploy and contribute relays to the Tor network for an extended period of time, with a greatly delayed payoff of at least 60 days due to how non-exit relays are ramped up to guard. Due to consensus weight, certain relays are favoured over others for Tor circuit building, so attackers would need to compete with them in order to attract more network traffic for easier end-to-end correlation.
Related: