You're often honest, but this note has many lies:
1. the outbox model is supported in many clients, and the ones that do not support it are adding support as we speak, so it's unfair to use this as an argument to say it is "complicated".
2. the outbox model wasn't ever an afterthought, sure, implementation details weren't clearly described in the spec (or in my mind), but it was always how Nostr was supposed to work, and Mike Dilger (npub1acg…p35c) was the first who materialized it.
3. the outbox model is not unreliable -- or at least I have never seen any claim about it being unreliable or have any reason to believe it is unreliable, but I'm open to hear why you think it is.
In fact everything in my experience DHTs are unreliable. I know you're 100% sure that they are perfect, but even if they are very good as you claim they can still fail. Maybe I used them wrong every time? Maybe the peers and network conditions were doing something wrong? Probably, but even then if that is the argument I can also claim that the outbox model, if used correctly and perfectly will also always work.