I originally told the Librarian to find trading research. It found research all right. What I meant was: find strategies a tester can actually run. Those are not the same thing.
We changed the job. The Librarian now searches public books, academic papers, government and exchange documentation, trader websites, indicator vendors, YouTube transcripts, Reddit discussions, and social posts. It saves the exact claim, source, date, permission status, market, timeframe, and missing pieces.
Hype can be a lead
A trader with a loyal audience may have good instincts. A hidden Reddit group may notice something before a newspaper does. A vendor selling an indicator may explain a useful market structure. That can start an investigation. It cannot finish one.
The research also covers things that can confirm or cancel a technical setup: liquidity, flows, on-chain activity, volatility, Bitcoin dependence, market regime, sentiment, and scheduled macro events. “Everything looks good except Bitcoin is falling apart” is information the desk should not ignore.
The handoff has to be testable
A useful handoff says exactly what creates an entry, what exits it, which coin and timeframe it applies to, and what fees and slippage to assume. If the creator leaves out half the rules, the record stays incomplete. The tester does not fill the blanks with imagination.
Web pages are treated as untrusted input. A sentence hidden in a page cannot give the agent more tools or permission. Pirated books and bypassed subscriptions stay out.