Built on existing infrastructure — Covalent / GoldRush, Moralis, Alchemy, The Graph, Dune Analytics — to fetch address-level public data across chains (balances, transactions, protocol interactions, NFTs, staking and lending records). V1 prioritizes Ethereum, Base, and Solana, the highest-density chains. No custom indexer — keeping early engineering and ops cost minimal.
Transparent weighted rules: wallet age, transaction breadth and depth, protocol diversity, staking and liquidity history, asset distribution — users can see exactly why they got their score. No black box.Lightweight machine learning: small models (XGBoost / LightGBM) for anomaly detection — catching score farming, sybil attacks, and bot clusters that rule-based scoring alone can be gamed around. Training and inference need no GPU; an ordinary cloud server suffices. This is where “AI” earns its place rather than being a label.
Off-the-shelf large language models via API (not self-hosted, not self-trained) turn scores into plain-language readings — e.g., “You’re a long-term active DeFi participant who favors conservative lending and rarely touches speculative tokens.” Pay-per-call, low marginal cost, swappable for cheaper open-source models at any time.
3.4 On-Chain Credentials & Validators: An Honest Trust Assumption
Rings are issued as soulbound tokens / verifiable credentials, written to the user’s address — non-transferable, non-sellable.In V1, scores are co-signed and anchored by a small set of publicly identified, reputationally accountable signers (initially the team itself, later 1–2 trusted partners). We state the current trust assumption plainly — no pretending to be “fully decentralized” — and expand toward a stake-based validator network on the roadmap. We deliberately do not use an “anyone can run a node” model (see Appendix A).
