Towards an intent-centric topology [draft]

[degregat]

@cwgoes

Truly permissionless systems do not exist in the real world, because they have no way to prevent DoS. The difference between the permissioning of the Ethereum mempool and the permissioning of someone’s API is merely the distribution function of the tokens providing authentication.

@cwgoes

Second, and more fundamentally, I think there is a subtle, implicit assumption in play here: that whether a system is permissionless or permissioned and private or public is a choice made by the intent system or intentpool designer. I do not think this assumption holds. At an operational level, intents are just messages sent between peers. What messages to accept (submission permissioning) and who to allow to read them (privacy) are choices made by individual nodes, not intent system designers.

If you make these two points, I think making the following explicit would make sense:

Corollary (?)

Trust-aware intent sharing is one model to span (large parts of) the spectrum. Any agents can decide based on their trust how much to share or accept, keep track of the outcomes over the iterated game and adjust the amount of information they reveal and ingest based on their measurements over time.

If execution only depends on being able to see intents, this dimension is also included.
If we want to restrict anything beyond “visibility == full capabilities to act on it” maybe one might want to model an intent as a resource, or at least include a predicate in it.

Example:

• I see you on the network and don’t know who you are, but I want to test if we want to cooperate
• I share my lowest utility/locally unsolvable intents/whatever low risk experiment I can perform with you
• I observe if you reciprocate
• Depending on how that goes, we build on that
• If you defect at some point, I stop sharing again

This would fit naturally with the kudos approach.

Side notes, which might be interesting for a blog post

I also think the “sharing views of the state space and composition happens via consensus” is close to that, but on the state space you decide that you want to produce common history for a subspace. Here you need to decide that you want to traverse (parts of) the frontier of what that future history might be together.

Here, a message is just a partial view, and with intents we don’t even need to spin up a new identity as we need to do with signed histories of the state space when we want to break down the view of one identity.

Side note, likely out of scope for a blog post

If you want to go to fully generalize higher order N-Intents, you should be able to stack this ad infinitum, by traversing frontiers over intents (over intents, …), and we could also run consensus to prevent being front run or so, at least in a single round unless someone figures out how to do perpetual or dynamic committments.