I did see a bit (experiment in April 2024), and I've implemented a slight modification to the algo.

During a random walk, if a cycle occurs (a --> b --> ... ---> a), then the walk stops.

This doesn't influence ranks of normal users because the probably of a cycle is super small give that on average people follow 100 others. However, some bad actors follow each others to boost their rankings, and this simple modification lowered their ranks by a factor of ~3.