I want to add that silencing an instance removes it from the Federated Timeline and Mutes all notifications from that instance.*

• Except for users one follows, their toots/retoots will show up on the Federated Timeline and one will recieve notifications from that user.

I would favour a three-strikes policy. If after three requests they can't herd their cats in manners consistent with social.coop's accepted rules, and their behaviour persists, that means that they're incapable of or unwilling to keep their instance users under check (independently of whatever their stated policies declare). That makes their content inherently incompatible, and thusly shouldn't be accepted / federated.
Taking their declared intent at face value would be a blank cheque, that wouldn't work because all of these trolls and other pests claim to be for "free speech".

When they identify things they don't like (i.e. left wing politics, brown and black people) they hide behind the flag of free speech to conceal their actual motives (racism, fascism).
I believe a three-strikes notice (or some other similar method) for behaviour inconsistent with our rules would pre-emptively render the free speech discussion moot.

Less waste of time and energies.

I think that the three-strike policy isn't that necessary since this is an instance-to-instance policy.

Ie, the three ways (incompatible CoC, or shitposting, or free speech* as you described) in which an instance would be silenced would change slowly and an instance would either "not-mind" or would have a path to become compatible with the Federation Policy.

This is a small step to clearing the federated timeline and notifications and we would still need to takes steps towards a Code of Conduct and a Moderation Policy.

I see what you mean. If it's intended as a small cleanup step, then that's fine. Leave the more complex edge cases for later then.