It could be a mix of identity providers. Identities can be verified in numerous ways (PGP key, posting a code on this forum, linking a facebook account, whatever plurality). You can integrate existing trust models (but just because I have facebook friended Ted Bundy does not mean I trust him).
Each identity may select those they trust, those they do not, and neutral (good, bad,
ugly unknown) as a directed graph (just because I trust you does not imply that you trust me). Ignore all trust calculus linked through an untrusted identity (I don't care who an untrusted identity trusts or doesn't trust). You can play with the numbers, but for example: what I directly trust, I trust 100%. Each single link of trust is weighted 10%, each double indirection, 1%.
If my mom trusts a merchant, then I trust it 10%. If two of my mom's coworkers trust it and one does not, then I trust the merchant 12%.
As long as it's boolean (and neutral) and not qualitative (good products, cheap prices, fast delivery, sexy staff) this model works. There is no such thing as "I trust a merchant more than another", the identity has either delivered or not.