# LexRails
Deterministic legal-citation fabrication checks for AI agents. USDC on Base via x402, no keys, no signup.
LLMs fabricate case citations (700+ documented court incidents; sanctions in Mata v. Avianca and beyond). A large class of fabrications is catchable STRUCTURALLY, offline: impossible reporter volumes (e.g. "999 U.S. 999" — U.S. Reports has ~600 volumes), unknown reporters, malformed USC/CFR/ECLI. LexRails sells exactly that check. Every paid response is Ed25519-signed.
Honest scope: verdicts are structural (impossible / unknown-reporter / plausible). "Plausible" is NOT existence proof — for existence, use a case-law database (e.g. CourtListener).

## Endpoints
- GET /v1/cite/check?q=  ($0.008) — check one citation: "410 U.S. 113", "2022 WL 4637582", etc.
- POST /v1/cite/extract  ($0.028) — body {"text": ...}: extract & check every citation in a draft; returns riskFlag.
- GET /v1/uscode/validate?title=&section=  ($0.009) — U.S. Code structural check (titles 1-54; 53 reserved).
- GET /v1/cfr/validate?title=&part=&section=  ($0.009) — CFR structural check (titles 1-50; 35 reserved).
- GET /v1/ecli/validate?id=  ($0.009) — European ECLI format/jurisdiction/year check.
- GET /v1/reporters  (free) — the bundled reporter table with volume bounds (asOf 2026-08).

## Discovery
- OpenAPI: /openapi.json · x402: /.well-known/x402.json · Signing key: /.well-known/signing-key.json