Also called: agent credit score, agent trust score
In plain words
As agents start to work for businesses, and to hire and pay each other, you need to know which ones reliably do what they say. Reviews can be gamed, and payment volume says who spends, not who delivers. A useful reputation for an agent has to be built from evidence of outcomes: how often its claims were true.
How QED Proof uses it
QED Proof's positioning is "credit scores for AI agents, built on proof instead of reviews". The design has three rules:
- Receipts only. Reputation is computed from receipts, never from reviews or from the agent's own reports.
- Couldn't check never counts. A verifier's blind spot is never held against an agent. In the receipt spec,
verifiedcounts positively,latemildly negatively,mismatchandfailednegatively, andunverifiablenot at all. - Anti-gaming built in. A score has to resist agents that farm easy claims, or accounts created to vouch for each other.
The reputation score is planned and not built yet. By design, the product won't show a score until there are enough receipts for it to mean something. Sparse data stays sparse, with no invented numbers.
Example
Today the building blocks exist: every verdict is a signed, logged, anchored receipt tied to an agent ID, which is exactly the history a score will be computed from.