CREDENCE

Every scam begins with trust placed in the wrong place. Now imagine an AI agent moving value for you without knowing who to TRUST. This is a problem I can't stop thinking about. @arc is built for an economy where AI agents move real money autonomously, settling in under a second, with no reversing it. But on-chain, trust is invisible, a scam wallet and an honest one look identical until it's too late. A human can pause and have a bad feeling about a counterparty. An autonomous agent can not. It needs a number. So I built @CREDENCE_intel a trust intelligence layer for the Arc ecosystem. You pass it any wallet, agent, or contract address, and it answers one question. Can this be trusted? • With a trust score from 0 to 100 • Risk level • A confidence score • Actual reasons behind the verdict • Recommended action. The hard part was not producing a score. It was being honest about one. A brand new wallet with no history isn't "trustworthy". it's unknown, and most systems quietly paper over that distinction. CREDENCE does not, It keeps trust and confidence separate on purpose, so it never hands you a confident verdict built on thin data. Absence of red flags is not the same as safety, and the score tells you which one you're looking at. Because Arc is built for agents, I integrated ERC-8004 built by arc. It is the on-chain agent identity and reputation standard so agents can register, find work, and get paid autonomously directly into CREDENCE. So it can read and reason about agent reputation, not just wallet activity. And that's where it got interesting. Digging into the ERC-8004 reputation data, I found out a lot of it was farmed. The same handful of accounts rating each other in loops to inflate scores. So instead of trusting raw reputation, CREDENCE weighs rater diversity: a score you can't simply buy. Under the hood it runs an unsupervised anomaly model alongside transparent, auditable rules, and deliberately no faked training labels, because no honest ground truth for "trustworthy" exists on-chain yet. Every verdict is explainable, and it scores any address live, not just a fixed dataset. CREDENCE is live now : credence-intelligence.netlify.app A web app, a public Trust API, and SDKs for Python and JavaScript so other builders can plug it into their own Arc apps. This is the kind of work I love, taking a vague, important problem and turning it into something real, careful, and usable. If you're building on Arc or you just want to see what the trust layer for an agentic economy looks like, I would genuinely value your eyes on it.