Three Strikes: Warn, Suspend, Ban (USA)

This is the combination of two ideas outlined in the Unwelcome Sexual Advances description: most unwanted sexual advances come from a small number of perpetrators, and community specific norms make the reliable classification between wanted and unwanted incredibly challenging at the margins. For these two reasons, a system of escalating penalties for unwanted sexual advances is likely to be highly effective, because the users in the space can be more in control of constructing and policing the norms they want to set in this particular zone, the harmed party is always aware of the harm, and the serial abusers will quickly get kicked off of the platform.

A typical three strike system is:

1. Warn - on first violation (or in this case, first report) warn the user that their behavior was found to be objectionable, and help them reset their expectations for the space.
2. Suspend - after a second violation, send a stronger message and force the user to take a time out.
3. Ban - on third violation, remove the user's account and their ability to continue to use this as a source of harm.

This approach can be tempered by some sort of classifier to ensure that the content that was reported was potentially construable as being a sexual advance. While ML is very bad at ascertaining the cultural distinctions between unacceptable and acceptable behavior, it will be much better at the task of determining the outer-boundaries of "could this be construed as a sexual advance".