relex-researchlisted

# Legal Research: You Discover, the Harness Grounds A Relex draft may only cite law whose **verbatim text sits in the global cache** — the server's verifier rejects anything else. Your web results and memory can never ground a citation; they only *find* what to ground. So: ``` discover (you: web + public legal MCPs) → directive (POST /research/scrape — the harness caches VERBATIM text) → ground (the case agent cites the cached text; verifier enforces) ``` Epistemic discipline while discovering is `relex-citations` — in short: no docket numbers, margin numbers or quotes from model memory, ever. ## Step 1 — know what the harness can fetch `execute GET /research/sources` — the live registry. `enabled: true` sources can be fetched-and-cached by directive. Everything else is **discovery-only**: you research it yourself and pass the best official URL as `sourceHint`, or the harness falls back to its generic fetch ladder. ## Step 2 — discover (your job) Use your own WebSearch/WebFetch and, when connected, public legal MCPs. Prefer official/free sources; identify the authority in **anchor form** — topic, court/ bench, date if known, official source to verify in — never a memorized citation. | Jurisdiction | Best discovery channels | |---|---| | US | CourtListener official MCP (mcp.courtlistener.com) — case law, dockets, citation check; GovInfo/eCFR for federal statutes/regs; LOCUS for local ordinances (below) | | DE | NeuRIS (rechtsinformationen.bund.de) — new official