1. Name the decision
Write down the next decision the research should support. “Understand this company’s recent margin change” is specific. “Find the best stock” leaves the horizon, constraints and definition of success unresolved. Separate a research task from an instruction to trade.
2. Inspect one source chain
Pick one answer with a numerical claim. Open its source and check the reporting period, units and whether the figure is adjusted. Then reproduce any calculation. A tool that cannot show its source chain needs a different level of trust than one that can.
3. Check your actual coverage
Test the company, geography, filing type and history you use. A product’s broad “global markets” description does not establish that a particular exchange, quarterly filing or data field is available. Record missing data rather than filling it with an AI guess.
4. Evaluate the output you will keep
Ask for a compact memo with observations, sources, assumptions and open questions. If the output is a strategy, require explicit rules and a trade log. Save the date and model or product version when available so another researcher can understand what changed.
5. Take the next step in Premiss
Use Premiss when the research question should connect to a strategy you can test and inspect. Begin with the question and supported market, clarify the rules, then review the generated code and recorded trades. Source-linked research and a historical trading test answer different questions; keep both artifacts explicit.
Sources and next steps
- Premiss — official product ↗AI-assisted research, strategy creation and inspectable historical tests.
- SEC EDGAR ↗Original US public-company filings; a primary-source reference, not an AI research product.
- Premiss — current plans and boundaries ↗Historical research, backtests and paper trading; check current allowances.