Keep in mind that insurers will dictate your setup to a degree. They won't just insure any scheme you figure out for yourself, or receive much worse terms.
Still could be worth it but it's a serious tradeoff, along with privacy implications.

Can you write up your experiences with the DoS attacks so we learn how to protect our services better?

Exactly what I was thinking the other day.
I'll have a word about flotilla with someone I know, hope he's a good lead and contacts you.

That's cool and all but realistically bunker connections can be flaky and lots of footguns with implementation from both sides.

If you want to retain some users better expose an nsec / ncryptsec login as well.

Do you think Spark and Ark require comparable fees to exit unilaterally? Have you seen a similar study for Ark?

This data liveness requirement is painful but I think comparable to the pain of expiring VTXOs in Ark. There you already always have the state to publish for the unilateral exit but you (wallet or watchtower) must monitor the expiry deadlines.

Of course the other thing is that ark operators are less trusted since they cannot conspire to steal funds.
However I still love Spark because at the end of the day operators don't have enormous liquidity requirements and this will make a big difference in adoption.

Also in Ark has that constant periodical refresh from user wallets, that doesn't look good to me. I don't necessarily open my wallets every month and this is going to be a problem when a lot of users realize this (or have to pay extra for a refresher service).

All in all love the innovation of the L2 space

Data-only works well. Phone numbers and SMS can be hard to troubleshoot, much depends on other parties

Anyone implementing the same set of rules ie the subset of the nostr protocol most agree on will see the same events and the rules can be enforced server side optionally as well.

What I struggle to understand:
1. If you have your own self hosted relays and git servers where you can enforce rules to the extent you want how is that different moderation from Forgejo
2. If I fork and mirror your forgejo instance and maintain a bridge, then develop a client that views both sources, how is that different from what you describe as the issue with nostr?
Forgejo is open source so anyone can build this easily today. You're not bothered by that?

Nostr gives you signatures from sovereign IDs and standard data modeling, so it's more neutral.
Everything else I see comes down product specifics.