shazow pfp
shazow
@shazow.eth
Interesting constraint of "decentralized but offchain" like Bluesky: Everything is signed by secret keys, so either it has to be individually owned or delegated to a trusted intermediary, there is no programmable middle ground. Whereas onchain, anything that is ownable can also be owned by a collective (or an arbitrary program). This is very powerful, for example: Posts and block lists and labellers must be owned by an individual, but... wouldn't it be better to have accounts posting owned by teams, or block lists owned by collectives with governance, or labellers owned by verifiable programs, etc? Can't do that without onchain.
1 reply
0 recast
7 reactions

Brent Fitzgerald pfp
Brent Fitzgerald
@bf
Yeah onchain provides a collective “right here right now” for coordination to happen. But you can do a lot with just signatures.
1 reply
0 recast
0 reaction

shazow pfp
shazow
@shazow.eth
Can you? I was trying to imagine how you'd approach similar functionality with just signatures, and best I could come up with is somekind of MPC or ZK but signature/state aggregation is still an issue and additive signatures require consensus/resistance against peers withholding messages, seems you just slowly reinvent a blockchain?
1 reply
0 recast
0 reaction

Brent Fitzgerald pfp
Brent Fitzgerald
@bf
Right, “a lot” is fuzzy. I didn’t mean you can approach that functionality with just sigs and I agree the chain state unlocks a radical level of coordination. But you could use MPC to split a secret so messages are only signable as a group. Or you could have verifiable anon polling with ZKs. And maybe those even compose interestingly.
2 replies
0 recast
0 reaction

shazow pfp
shazow
@shazow.eth
I was going down this path too, but then we need to coordinate the MPC (unless we require all actions to be coordinated out of band and committed as a unit), and if it's responsible for any ownership/permission then an incomplete multi-step action can act as censorship against permission changes, so at minimum we get state splits that don't always converge because there's no way to guarantee ordering or timestamp correctness. I think at best we could do MPC time-agnostic content creation that is coordinated out of band, but we couldn't do anything more complex where ordering changes the semantics of the action.
0 reply
0 recast
0 reaction