Einstein Engagement Frequency was updated in Oct 2020, with a new, improved, ML model. The older version of Einstein Engagement Frequency predicted a more general, overall email send frequency range that was applied to all your subscribers/contacts. For eg. "Einstein recommends sending 7 - 10 emails to all your contacts within a 30-day period."
The new, and improved version predicts the optimal email send frequencies for every single contact at the individual level. It does this by optimizing for maximizing the message open rates, and minimizing the unsubscribes. Now, every single contact in your business unit has their own individual preferred frequency range (based on their past engagement). Manually setting saturation rules (like capping frequency at 4 emails per week), is not a true AI-driven solution and has its limitations. There still could be contacts within your BU who might be saturated at 4 emails per week, because their open rates are maximum at 2 emails per week and begin to drop after that. Einstein predicts this ideal send frequency for every single contact, ensuring higher open rates, and avoids causing message fatigue for your contacts.
There is a new 'Frequency Split' that you can use while designing journeys on Journey Builder, and in real-time Einstein can now intelligently segment your contacts based on their preferred frequency. You can try reducing the number of emails that are sent to your saturated segment of contacts, and increase the frequency for your undersaturated segment as you have an opportunity to send them more messages and drive up open rates.
Please see this trailhead live video for more details.