Built a pre-trade risk check for Solana trading agents that traces insider wallet clusters via shared first funders — looking for feedback on the approach.

I'm a solo dev. For a while I've been building tooling for LLM agents that trade Solana tokens, and I kept running into the same problem: the agent would call a "rug checker" tool, get back "mint authority revoked, freeze authority revoked, not blacklisted, looks fine", buy in, and then watch 30% of supply dump from a dozen wallets that were obviously the same person. Most of the risk checkers I tried stop at authority flags and blacklists. Those checks are necessary but easy to pass. The dev revokes authorities and spreads supply across fresh wallets, and on paper the holders look distributed. So I built a JSON API meant to be called as a tool before an agent executes a buy. The part I'd most like feedback on is insider cluster detection: For the top holders, walk back to each wallet's first funder — the address that sent it the SOL it used to become active. Group holders that share a first funder, or a short funder chain, into a cluster. Report cluster count, the share of supply each cluster holds combined, and which holders belong to it. Ten wallets holding 3% each look fine one at a time. One funder behind all ten holding 30% does not. The response also includes the usual checks, so the agent gets it all in one call: 0–100 safety score, plus the individual signals behind it, so you can set your own thresholds instead of trusting my weighting Holder concentration (top-N share, excluding known pool and program accounts) Honeypot detection (can it actually be sold) LP lock/burn status Mint/freeze authority Live price and liquidity Known limitations and open questions: Wallets funded straight from CEX hot wallets look like they share a funder. I currently exclude known exchange addresses, but the list isn't complete, so there are false positives. Funder-tracing depth vs. latency: deeper tracing catches more laundering hops but slows the call. Curious what latency budget people give pre-trade tools. Solana only for now. Pricing is simple: a free tier (15 requests/day, no card), then pay-per-call or a subscription. It also supports x402, so an agent can pay per request in USDC without anyone managing an API key or account. I'm interested whether anyone here is actually running x402 in production agents yet. Link and docs in the first comment, per the sub's rules. I'd really value critique of the clustering logic, the score design, or how you'd want this exposed as a tool (function schema, MCP, etc.).