Cases Workspace (Agent Law Cases)

The Cases Workspace is a paid-only, case-centric layer on top of FIFA jurisprudence and your own dossier. It pairs interactive AI chat over a per-case file store with longer deep-research jobs that run in Cursor Cloud and write a structured Agent Law Lab brief back to your account.

Indicative only — not legal advice

Every output is indicative. The Cases Workspace synthesizes public sources, your dossier, and a curated jurisprudence index. It is not a substitute for licensed FIFA agent representation, a club's own diligence, or professional legal/financial advice.

What's a "case"?

A case is a long-lived workspace owned by a paid user. It holds:

• Title, summary, status, tags — the core record (agent_cases).
• Dossier — uploaded PDF/DOCX/Markdown documents, external links, and free-text notes (case_documents). Files are stored in Cloudflare R2 and indexed in a per-case OpenAI vector store.
• Sources — source-first matrix of official FIFA, CAS/tribunal, court, press, and internal sources (case_sources).
• Findings — legal issues, facts, risks, arguments, next steps, and negative-search logs (case_findings).
• Chat messages — the back-and-forth with the case AI, with citations (case_messages).
• Research jobs — queued deep-research runs (case_research_jobs) that produce an Agent Law Lab brief.

Two AI shapes — chat and deep research

Shape	Latency	Where it runs	Best for
Case chat	~1–3 s	Cloudflare Worker → OpenAI Responses + file_search over your case store and the curated library	"Help me draft a position", "What does Article 15 say in this dossier?"
Deep research	1–3 min	Cursor Cloud agent (GitHub Action) → posts an Agent Law Lab brief back via the Worker	"Build me a full case brief with comparable jurisprudence and a recommended position"

Both use the same retrieval corpus: your case dossier (per-case vector store), structured sources/findings, and the shared curated jurisprudence index (links + short excerpts + metadata only).

Research depth

• Quick scan — lower cost, concise answer.
• Deep source search — official-source-first source collector, source matrix, negative searches, and Cursor briefing.
• Exhaustive legal dossier — longer run with legal issue map, facts vs assumptions, jurisdiction analysis, action plan, and manual verification checklist.

Completed research is saved in four places: law_lab_briefs, a case_documents excerpt indexed into chat, case_sources, and case_findings.

Curated jurisprudence — what we do not do

We do not re-host FIFA decision PDFs, articles, or full text. Instead, the curated library (juris_library_cases) contains:

• Source URL — link to the official FIFA portal or chamber decision page.
• Short excerpt — a few-sentence summary or key holding (used purely for retrieval).
• Metadata — tribunal, year, topic tags, parties (where public).

When the case AI cites a source, it must include the original URL. If the AI cannot find a grounded source, it says so rather than inventing an article number, case ID, or holding.

Compliance rules (always)

1. Never invent FIFA articles, case IDs, dates, or holdings. If retrieval is empty, say so.
2. Always cite the source URL for jurisprudence references.
3. Never re-host full FIFA decision PDFs in the dossier or the library.
4. Mark output as indicative. It is not legal advice.

Where it lives

• Web app: /cases (paid users only).
• MCP: list_cases, get_case, build_case_source_pack, list_case_sources, get_case_findings, attach_case_source, search_case, add_case_note, add_case_link, request_case_research, search_juris_library — see Cases tools.
• Internal: POST /api/internal/cases/:id and POST /api/internal/law-lab/brief are used by the Cursor runner; both require LAW_LAB_INGEST_SECRET.

Next — Cases MCP tools → · Agent Pro workflows →