Best x402 services

Two stages. (1) RELEVANCE: when a free-text need is given, each candidate is scored on how well your query matches its AI-derived capability tags and summary plus its name/description/category (falls back to plain text match). (2) QUALITY: measured families combined with explicit weights - reliability (live status, uptime, response time), x402 compliance (the share of deterministic conformance checks the service passes, capped at 0.6 out of 1, the floor of the C band, when at least one of its EVM routes is missing the EIP-712 domain parameters a standard x402 client needs in order to sign a payment), economics (price and in-category price percentile), safety risk, and a SMALL ~10% weight on on-chain traction (per-service settlement volume, transaction count and unique buyers, measured over the service's known payTo via recognized settlers as a conservative undercount). Traction never dominates; a service with a shared payTo has only its settlement volume attributed pro-quota (divided across the services sharing that payout address), while transaction count and unique buyers stay whole operator-level figures, and a service on an unmeasured network or a shared member whose probe is failing carries no traction term (the other weights are renormalized). The term also gates on real recent settlement: it scores 0 when there was no settlement in the last 30 UTC days, or when 30d volume is below a $10 floor (sub-floor volume is indistinguishable from dust and never moves rank). Each recommendation also carries top_buyer_share_30d (0-1, the 30d volume share of the largest single buyer); a payout concentrated in one buyer is discounted, up to half the traction term when a single buyer accounts for all of it, since one buyer is one relationship rather than the broad demand the term is meant to reward. AI-derived fields are labeled {value,confidence,source:'ai'} and NEVER override a measured value.