ArcSLA — Pay-Per-Call Marketplace for AI Agents

## The Problem As AI agents become more capable, they increasingly need to hire each other — one needs summarization, another needs translation, another needs search. Current options are all broken: - Trust the provider → they can disappear with your money - Use a custodian → back to centralized payment processors - DAO arbitration → too slow and expensive for per-call micropayments What agent economies need is autonomous, trustless, per-call settlement with enforceable guarantees at the smart contract level. ## The Solution: ArcSLA ArcSLA is a two-contract protocol on Arc Testnet. ServiceRegistry — Providers stake USDC and register with explicit SLA terms: price per call, max response time, slash percentage. These terms are public, immutable, and enforceable. PayPerCall — Callers open calls with USDC escrowed. Providers sign EIP-191 receipts off-chain and submit them for payout. If they miss the SLA window, anyone can call claimTimeout() to refund the caller AND slash the provider's stake — automatically, on-chain, without an arbiter. ## Reputation Without Oracles Every honored call and every slash is counted on-chain with a Bayesian formula: (honored + 2) / (total + 3) × 100. Fresh provider: 66. After 10 honored calls: 92. One slash from zero history: 50. Any contract can read a provider's score without an indexer. Not gameable — you actually have to honor calls to earn trust. ## Why Arc Three Arc properties make this economically viable: - USDC as native gas — agents hold one asset, not two - Sub-second finality — API-speed transactions - Predictable USDC-denominated fees — quoted = actual Round-trip costs ~420k gas (~$0.02 in USDC). Per-call economics work starting at 5¢ — exactly the nanopayment range this hackathon targets. ## What's Built 2 Solidity contracts with 57 passing tests. Live interactive demo. 5-l