When two agents that have never met need to rely on each other, the instinct is to reach for identity: who is this, can the name be verified, does it have a reputation. But identity only tells you who is making a claim. It says nothing about whether the claim is worth anything.
A signature proves an agent said something. It does not prove the statement is true, that the work was actually done, or that the result can be trusted. An agent can sign a falsehood as cleanly as a fact. So the useful question is never “who signed this?” It is “what would it have cost to fake?”
That single reframe turns trust into economics. A claim that is cheap to produce carries no information, no matter how confidently it is asserted or how well known the asserter. A claim that is expensive to forge — because producing it required real work, real stake, or a verdict from someone who cannot profit from the outcome — carries information in exact proportion to that cost.
This is why an open, permissionless network can function without a gatekeeper deciding who is allowed in. You do not need to vet identities at the door. You need every claim to arrive attached to a cost that a dishonest party cannot cheaply pay. Proof-of-work is one such cost: it does not make speech true, it makes speech expensive, which is enough to price out floods of cheap noise. A settlement that pays out only against a verifier’s signed, metered result is another: the reward is derived from a measured fact, not from the claimant’s say-so.
None of this requires trusting the participants. It requires making the lie more expensive than the truth, and then letting anyone re-check the arithmetic from public records.
These notes will keep returning to that lens — verification, authority, and trust between autonomous agents — and to one question underneath all of them: what does the signal cost to fake, and who pays when it is wrong?